diff --git a/.gitignore b/.gitignore
index 8e0bb41..2a40dfe 100644
--- a/.gitignore
+++ b/.gitignore
@@ -11,3 +11,5 @@
 graphify-out/
 # doc-hygiene plugin state
 .dochygiene/
+# os-adr plugin state
+.os-adr/
diff --git a/CLAUDE.md b/CLAUDE.md
index 2dca195..8482fee 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -92,6 +92,14 @@ to those two and fix the stale doc.
 - Behavior: Monitors stale and bloated project documentation per-project under `.dochygiene/` state dir (gitignored). SessionStart reminder is deterministic (no AI tokens, once/day snooze). Skills (verb-first, no `commands/` dispatcher — invoked directly as `/os-doc-hygiene:<skill>`, per [[cc-os-plugin-skill-naming-convention]]): `check` (AI-assisted classification of staleness signals, emits machine+human reports), `clean` (AI-assisted or deterministic patch application with git-safe scoped cleanup), `status` (read-only lifecycle-timestamp read), `sweep` (check then clean in sequence). Reversion-protected via invariants.md + golden-example test fixtures.
 - Migration: migrated from standalone repo (`~/dev/cc-plugins/doc-hygiene/`, 2026-07-03) and integrated into cc-os. No content conflicts detected (doc-hygiene scope does not overlap cc-os memory system). Renamed from `doc-hygiene` to `os-doc-hygiene` per cc-os plugin naming convention. Skills renamed `hygiene-check`/`hygiene-clean` → `check`/`clean`, and the `commands/hygiene.md` dispatcher removed in favor of two new skills (`status`, `sweep`), aligning with the `os-vault`/`os-orchestration` pattern of no `commands/` directory (2026-07-03).
 
+**Global os-adr plugin** — `cc-os/plugins/os-adr/` (git-tracked, 2026-07-03); symlinked into `~/.claude/plugins/os-adr`
+- Purpose: Architecture Decision Records, consistently in every project — one customized-Nygard template (frontmatter `id`/`date`/`status`/`supersedes`/`superseded-by`/`affected-paths`/`affected-components`; Context/Decision/Consequences/Alternatives-rejected sections), `docs/adr/NNNN-kebab-title.md` one-file-per-decision + fully-regenerated `docs/adr/README.md` index. Built from `docs/adr-system/` (requirements `04`, PRD `05`, eval sketches `06`) via OpenSpec change `add-os-adr-plugin`.
+- Core: Ruby OO library `lib/adr/` (`Record`, `Repository`, `Index`, `Template`, `Detector`, `Migrator`, `MigrationReport`, `Finder`; Sandi Metz style, injected paths) behind thin CLIs in `bin/` (`adr-new`, `adr-init`, `adr-detect`, `adr-migrate`, `adr-find`). Tests: `ruby tests/all.rb` (47) + `python3 tests/hook_test.py` (15), all model-free. Behavioral invariants in `invariants.md`.
+- Hook: `hooks/session_start.py` wired via plugin-relative `hooks/hooks.json` (`${CLAUDE_PLUGIN_ROOT}`, no `settings.json` entries) — deterministic existence check: present → near-zero-token usage note naming `/os-adr:create` + `/os-adr:find`; absent → once-per-day init/migrate suggestion, silenced permanently by `.os-adr/suppress` (gitignored per-project state dir); silent outside git projects.
+- Skills (verb-first, no `commands/`): `create` (one-invocation templated+numbered+indexed ADR with mechanical supersession), `init`, `migrate` (non-destructive: detect surveyed shapes → mechanical heuristic fill → LLM fills only manifest-listed interpretive fields via `--apply-fills` → `migration_confidence` frontmatter + `docs/adr/migration-report.md`; old-system deletion only as an explicit separate user-approved step), `find` (deterministic-first retrieval: affected-paths/components match → Accepted-only filter → one-hop Graphify query-path expansion with graceful degradation → AI judgment over the candidate set only).
+- Migration pilot passed 2026-07-03 (sandboxed project copies; gate threshold 25% low-confidence flags): viking-warrior-training-log 0%, delta-refinery 0%, llf-schema 8.3%. Retrieval verified end-to-end against the real llf-schema project graph. Resolved: ADRs need no Graphify tag/edge convention — retrieval layer 3 expands the *query* paths via graph-node `source_file`s.
+- **Remaining (locked rollout order):** real-project migration/adoption one at a time via `/os-adr:migrate` — pilot projects first, then the cc-os retrofit (its 19-ADR monolithic file was deliberately excluded from the pilot), then wider; plus the deferred unprompted-behavior eval (held-out scenario shapes sketched in `docs/adr-system/06-eval-scenarios.md`; methodology locked in `04-plugin-requirements.md`).
+
 **Graphify** — v0.8.31 at `/home/jared/.local/bin/graphify`
 - Vault graph: `~/Documents/SecondBrain/graphify-out/` — disposable, structural index rebuilt by SessionStart hook; handles relational/graph queries. Distinct from vault-index.json (planned, Plan B Phase 1): a flat `{tag: [{path, title, summary}]}` lookup rebuilt at SessionEnd, queried by the `/memory-find` skill for fast tag-based SB note discovery.
 - Project graph: `<project-root>/graphify-out/` — same pattern; gitignore it in each project repo
diff --git a/docs/adr-system/00-README.md b/docs/adr-system/00-README.md
index 9ceb5fb..9072922 100644
--- a/docs/adr-system/00-README.md
+++ b/docs/adr-system/00-README.md
@@ -1,12 +1,13 @@
-# ADR System — Research Phase
+# ADR System — Research & Build
 
-_Last updated: 2026-07-03_
+_Last updated: 2026-07-03 — plugin BUILT (OpenSpec change `add-os-adr-plugin`, ADR-020); real-project migration/rollout remains, per the order below._
 
 This directory captures research for standardizing how Architecture Decision Records (ADRs)
-are organized and used across the user's projects, as a precursor to building a Claude Code
-plugin that automates ADR creation and relevance-scoped retrieval. **Nothing here is built
-yet** — this is the survey/research phase only, requested to answer "what do we have, and what
-could we have with a little engineering elbow grease."
+are organized and used across the user's projects, and the requirements/PRD the **now-built**
+`os-adr` plugin was implemented from (`cc-os/plugins/os-adr/`, installed as
+`os-adr@local-plugins`; see CLAUDE.md Implemented Components). The docs below remain the design
+record; the migration pilot passed 2026-07-03 (flag rates: viking 0%, delta-refinery 0%,
+llf-schema 8.3% vs. a 25% gate).
 
 - **`01-current-state-survey.md`** — audit of every project under `~/dev/`, `~/clients/`, and
   `~/projects/` for existing ADR practices (organization, templates, tooling, or absence
@@ -17,8 +18,35 @@ could we have with a little engineering elbow grease."
 - **`03-synthesis-and-opportunities.md`** — synthesis answering the three "how might we" goals
   (know when to create/query, systematize creation for consistency, and query with high
   relevance/low bloat) and naming concrete design options for a future `os-adr` plugin.
+- **`04-plugin-requirements.md`** — locked, pre-build requirements: single standardized template
+  (no client/tooling-mode split), `docs/adr/` location, the five plugin requirements (mechanical,
+  SessionStart existence-check hook, non-destructive migration, unprompted write/retrieval
+  effectiveness), eval-as-final-stage sequencing, and OOP/Sandi-Metz/Ruby-or-Python
+  implementation style.
+- **`05-plugin-prd.md`** — the PRD to hand to `openspec-propose` when ready to build: problem
+  statement, goals/non-goals, requirements broken into build phases, the deferred
+  unprompted-behavior eval stage, architecture/success criteria. Incorporates a Fable review pass
+  (2026-07-03) that sharpened the migration and sequencing sections.
+- **`06-eval-scenarios.md`** — held-out eval scenario *sketches* (write-trigger W1–W3, retrieval
+  R1–R4) written before the plugin skeleton froze; the evaluation itself remains a deferred,
+  separate stage per `04-plugin-requirements.md`.
 
 Provenance note: `02-external-research.md` cites web sources gathered by research agents in a
 single pass; several specific tool/repo names and one incident anecdote are flagged inline as
 **[unverified]** because they were not independently checked against the primary source in this
 pass. Verify before depending on any specific tool name in that file.
+
+## Build/rollout order (decided 2026-07-03)
+
+1. **Build the `os-adr` plugin first**, against a clean/small pilot — not against cc-os's own
+   19-ADR file. cc-os's `docs/memory-system/03-architecture-decisions.md` is known to need
+   splitting into one-file-per-decision (per the ~30-decision monolithic-file ceiling noted in
+   `02-external-research.md`), but that split is **deliberately deferred**, not forgotten — it
+   happens *using* the plugin once built, not by hand beforehand.
+2. **Then use the plugin to retrofit**: cc-os itself first, then other Claude Code projects on
+   this machine (per the `~/dev/`/`~/clients/`/`~/projects/` survey in `01-current-state-survey.md`),
+   then any new project encountered going forward — one at a time, matching the
+   build-first/migrate-incrementally precedent already set for the vault (ADR-013).
+
+Do not re-litigate this ordering in a future session — build the tool, then use the tool to fix
+the mess the tool was built to fix.
diff --git a/docs/adr-system/04-plugin-requirements.md b/docs/adr-system/04-plugin-requirements.md
new file mode 100644
index 0000000..f5d5224
--- /dev/null
+++ b/docs/adr-system/04-plugin-requirements.md
@@ -0,0 +1,87 @@
+# Plugin Requirements (locked, pre-build)
+
+_Last updated: 2026-07-03_
+
+Captured from the requirements-gathering conversation following the research phase
+(`00-README.md`–`03-synthesis-and-opportunities.md`). These are decided; do not re-litigate them
+in a future session without a new discussion that explicitly revisits one.
+
+## Template decision
+
+**One standardized template for every project, no client/tooling-mode split.** Superseding the
+"two modes" option floated in `03-synthesis-and-opportunities.md` — that split was an artifact of
+projects independently improvising with no shared plugin, not a real requirement. The user is not
+a solo operator but works with a team of AI agents across every project; those agents need to
+discover and write ADRs consistently, which a single template serves better than per-project
+variants ever could.
+
+- Customized Nygard: `Status` · `Context` · `Decision` · `Consequences` **+ `Alternatives
+  rejected`** as a first-class section (highest-value-per-word field per
+  `02-external-research.md`).
+- Frontmatter: `supersedes:` / `superseded-by:`, `affected-paths:`/`affected-components:` (for
+  deterministic retrieval), `date`, sequential ID.
+- MADR's RACI fields (Deciders/Consulted/Informed) explicitly rejected — no project run by this
+  user has the team-coordination problem those fields solve.
+- One-file-per-decision + a mechanically-maintained index, always — including for small/client
+  projects. A monolithic single-file log is rejected outright for every project, not just large
+  ones: even a small client project's decisions log will bloat over time, and per-file+index
+  costs nothing extra at low decision counts (an index with 2 rows is fine).
+
+## Location
+
+**`docs/adr/`**, not a root-level `adr/`. Matches the actual industry convention (Nygard,
+adr-tools, MADR, Log4brains, Backstage all default to `doc/adr/`/`docs/adr/`) and matches every
+one of this user's own surveyed projects that already nests decision docs under `docs/`. Riding
+the well-known convention matters here specifically because one goal is agent discoverability
+*without* the plugin's help.
+
+## Five requirements for the `os-adr` plugin
+
+1. **As mechanical as possible.** Deterministic scripts/parsing/index maintenance; reserve LLM
+   calls for genuine judgment (classification, ambiguous field-mapping during migration,
+   "is this ADR-worthy" nudges). Same shape as the `os-doc-hygiene` plugin's deterministic
+   scanner + generative-only-where-needed split.
+2. **SessionStart hook, existence-check only.** Deterministically checks whether the ADR system
+   already exists in the current project (path + index file check, no LLM). If present: inject
+   the smallest possible context note that the ADR system exists and is usable, plus how to use
+   it — no bloat. If absent: notify the user and suggest setting it up. Same shape as
+   `os-doc-hygiene`'s zero-token SessionStart reminder.
+3. **Non-destructive migration, dual-existence until explicit go-ahead.** Build the new system
+   fresh, then convert every existing ADR (whatever shape/project it's in) into the new template
+   — best-guess fill-in for missing expected attributes, flagged where the guess is uncertain.
+   The old system stays in place, untouched, until the user explicitly approves deleting it.
+   Matches the build-first/migrate-incrementally precedent already set for the vault (ADR-013).
+4. **Effective at *writing* — unprompted trigger recognition.** The AI should recognize on its
+   own, without being told directly, when a decision in progress warrants writing a new ADR.
+5. **Effective at *retrieval* — unprompted, correct surfacing.** The AI should recognize on its
+   own, without being told directly, when an existing ADR is relevant to a task or question in
+   progress (including cases where the right move is "stop, this conflicts with a locked
+   decision, adjust the approach") — and retrieve the *correct* one, not merely *a* plausible one.
+
+## Sequencing: eval is a distinct final stage, not folded into the build
+
+Requirements 4 and 5 are an **evaluation methodology**, not a build task, and happen **after**
+the plugin is built and the design is believed sound — not concurrently with initial design.
+Sequence:
+
+1. Build the plugin (template, hooks, migration tooling) against the requirements above.
+2. Only once there's a working, believed-correct design: design held-out test scenarios —
+   prompts/tasks that are relevant to a written ADR but never name it or the exact constraint
+   directly — and check whether the AI (a) consults the ADR system unprompted at all, and
+   (b) retrieves the specific correct ADR, not just a plausible-looking one.
+3. This eval stage uses the **`autoresearch` skill** (modify → verify → keep/discard against a
+   metric) to iterate on retrieval/trigger heuristics against that held-out test set.
+
+Do not attempt to validate retrieval/write-trigger effectiveness informally during initial build
+— it needs the dedicated held-out-scenario methodology above to mean anything.
+
+## Implementation style
+
+- **Object-oriented**, following **Sandi Metz's** approach to OOP (small classes/methods, single
+  responsibility, dependency injection over reaching for globals, tell-don't-ask, duck typing
+  over type-checking) — the user's standing preference, [[ruby-and-sandi-metz-oop-preference]].
+- Language: **Ruby preferred**; **Python acceptable** where it's the more native fit for Claude
+  Code tooling (e.g. hook scripts invoked directly by `settings.json`, matching the existing
+  `os-vault`/`os-doc-hygiene` hook pattern in this repo, which are Python). Decide per-component,
+  not globally — a hook entry point can be a thin Python script that still delegates to an
+  OO-designed core, in either language, if that core is more naturally Ruby.
diff --git a/docs/adr-system/05-plugin-prd.md b/docs/adr-system/05-plugin-prd.md
new file mode 100644
index 0000000..ac7bc6c
--- /dev/null
+++ b/docs/adr-system/05-plugin-prd.md
@@ -0,0 +1,146 @@
+# PRD: `os-adr` Plugin
+
+_Last updated: 2026-07-03_
+
+Product requirements for the `os-adr` Claude Code plugin — the "what and why," to direct a
+future `openspec-propose` when the user is ready to implement. Built on the research
+(`01`–`03`), the locked requirements (`04-plugin-requirements.md`), and a review pass from Fable
+(external reviewer consult, 2026-07-03) that shaped the gaps and sequencing called out below.
+Do not silently narrow or reverse anything in `04-plugin-requirements.md` — this doc specifies
+*how* those requirements get built, not a re-decision of *what* they are.
+
+## Problem statement
+
+~43% of this user's own projects have ADR-like content, in 5+ mutually-incompatible shapes; the
+rest have none. No project uses ADR tooling. The user works with AI agents across every one of
+these projects, not just themselves — so decision records need to be discoverable and writable
+*consistently* by an agent that has never seen a given project before, which no current project
+provides. `cc-os`'s own 19-ADR/601-line file is already near the point research says monolithic
+logs stop working. This plugin exists to fix that, once, and mechanically enough that it doesn't
+become another bespoke-per-project thing.
+
+## Goals
+
+- One template, one location convention, enforced the same way in every project.
+- New ADRs get written (by the AI, largely unprompted) using that template with no per-project
+  relearning.
+- Existing ADR content in every project — no matter its current shape — converts into the new
+  template non-destructively.
+- An AI agent working in any onboarded project surfaces the *correct* relevant ADR without being
+  told to look, and recognizes when its own in-progress decision warrants a new one.
+
+## Non-goals (explicit, out of scope for this PRD)
+
+- Per-project or per-client template variants (rejected in `04-plugin-requirements.md`; not
+  revisited here).
+- Multi-editor / handed-off-repo governance. This plugin assumes the user is the primary ADR
+  author (directly, or via their AI agents) in every project it's installed in. A repo where
+  other humans independently maintain ADR conventions is out of scope for Phase 1 — flag this
+  assumption in the plugin's own docs rather than silently breaking there.
+- Mirroring ADRs into an enforceable-constraints surface (AGENTS.md-style hard rules). The
+  research surfaced this as a real tension (ADRs record "why," not "must never happen"), but nothing
+  today gives ADRs a severity field or an enforcement mechanism. Note as a candidate Phase 2 if a
+  severity field gets added later and high-severity decisions accumulate — not part of this build.
+- Retrofitting `cc-os`'s own ADR file as part of building the plugin. Per the README's build
+  order note: build first against a small pilot, retrofit cc-os and everything else after.
+
+## Requirements → build phases
+
+### Phase 1 — Template + location (mechanical, no LLM)
+
+- Ship the customized-Nygard template and `docs/adr/NNNN-title.md` + index file convention from
+  `04-plugin-requirements.md`, as a copy-paste-ready generator (`os-adr:new` or similar skill).
+- Index file is maintained mechanically (append/update on every write) — never hand-edited.
+
+### Phase 2 — SessionStart existence-check hook
+
+- Deterministic only: check for `docs/adr/` + index file. No LLM call in the hook itself.
+- Present → inject the smallest possible context note (existence + one-line "how to use it").
+  Budget this like `os-doc-hygiene`'s reminder: near-zero tokens.
+- Absent → notify the user, suggest running the setup/migration skill. **Needs a suppression
+  mechanism** (flagged by Fable review) so a project that deliberately has no ADR system yet
+  doesn't re-nag every session — a per-project dismiss flag (e.g. a marker file or config line),
+  same shape as `os-doc-hygiene`'s snooze state.
+
+### Phase 3 — Non-destructive migration
+
+This is the highest-friction requirement (Fable review flags it as the one with the most edge
+cases) — resolve these before implementation starts, not during:
+
+- **Detection**: what counts as "an existing ADR" in an unknown project must be defined per
+  known shape from `01-current-state-survey.md` (numbered per-file dirs, dated single files,
+  monolithic multi-decision files, prose-embedded-in-other-docs) — plus a fallback path for
+  shapes not yet seen.
+- **Field-mapping threshold** (mechanical vs. LLM boundary — the vaguest part of requirement 4a
+  as originally stated): define a concrete rule before build, e.g. *heuristic-fill fields that
+  are structurally unambiguous (Status/Date via frontmatter or clear heading), LLM-fill only
+  fields requiring interpretation (Consequences, Alternatives-rejected) with the result flagged,
+  never silently invent a Decision or Context field from nothing.*
+- **Flagging format**: a single, consistent surface for uncertain guesses — a per-file frontmatter
+  field (e.g. `migration_confidence: low`) plus one migration report file, not scattered
+  inline comments. Decide the exact shape during build, but it must be one mechanism, not
+  several.
+- **Non-destructiveness**: old ADR content is never deleted or moved during migration — new files
+  are written alongside it under `docs/adr/`. Deletion of the old system happens only on an
+  explicit, separate user-approved step.
+- **Pilot before shipping broadly**: run the migration on 2-3 real projects (candidates: the ones
+  already surveyed with ADR content — viking-warrior-training-log, delta-refinery, llf-schema)
+  and check the uncertain-flag rate before treating the heuristic as good enough. If a large
+  share of fields end up flagged, tighten the heuristic before wider rollout — this is a build-phase
+  gate, not a shipped feature.
+
+### Phase 4 — Retrieval (deterministic-first, per the research's layering)
+
+1. Path/component match against each ADR's `affected-paths`/`affected-components` frontmatter.
+2. Status filter — only `Accepted` by default.
+3. Graph traversal via Graphify once ADRs are indexed as graph nodes (reuses the `os-vault`
+   infrastructure rather than a parallel retrieval system).
+4. AI judgment only over the already-narrowed candidate set from 1-3 — never the full corpus.
+
+## Deferred: unprompted-behavior evaluation (requirements 4d/4e)
+
+Per the locked sequencing in `04-plugin-requirements.md`: this is a distinct stage *after* the
+plugin is built and believed correct, not concurrent with the build. However, per Fable's review,
+sketch (not implement) the held-out test scenarios **before** finalizing the plugin skeleton, so
+Phase 4's retrieval surfaces aren't accidentally missing something the eval will need later.
+Two example scenario shapes to sketch now:
+
+- **Write-trigger scenario**: give the AI a task that touches a decision of real consequence
+  (irreversible, cross-cutting, contested) without saying "write an ADR" — check whether it
+  proposes one unprompted.
+- **Retrieval scenario**: give the AI a task or question that conflicts with an already-written,
+  Accepted ADR without naming the ADR or the constraint directly — check whether it (a) surfaces
+  the ADR system at all, and (b) retrieves the *specific correct* one, not merely a plausible one.
+
+If sketching these surfaces a plugin surface Phase 1-4 doesn't yet provide (e.g. a query tool
+callable mid-task, not just at SessionStart), fold that surface into the initial build. Do not
+fold the *evaluation itself* in — that stays a separate, later stage using the `autoresearch`
+skill's modify→verify→keep/discard loop against held-out scenarios, per the locked sequencing.
+
+## Architecture / implementation style
+
+- Object-oriented, Sandi Metz principles (small classes/methods, single responsibility,
+  dependency injection over globals, tell-don't-ask, duck typing over type-checks) —
+  see the [[ruby-and-sandi-metz-oop-preference]] memory.
+- Ruby preferred; Python acceptable for hook entry points where it's the more native fit
+  (matching the existing `os-vault`/`os-doc-hygiene` Python hook pattern) — decided
+  per-component, not as a global language choice. A Python hook script may thinly wrap an
+  OO-designed Ruby or Python core.
+
+## Success criteria
+
+- Every surveyed project with existing ADR content converts with zero data loss (old system
+  fully intact until user-approved deletion).
+- Migration uncertain-flag rate stays low enough to be trustworthy (define and check an explicit
+  threshold during the Phase 3 pilot — do not ship on an unmeasured guess).
+- SessionStart hook overhead is near-zero-token and adds no perceptible session-start latency.
+- A new ADR can be written, correctly templated and indexed, in one skill invocation.
+- (Deferred, Eval stage) unprompted write-trigger and retrieval accuracy measured against a
+  held-out scenario set — no numeric target set yet; the held-out methodology itself is the
+  Phase 5 deliverable.
+
+## Rollout order
+
+Matches the README's locked build order: build and pilot this plugin on a small/clean project
+first, then use it to retrofit `cc-os` itself, then other `~/dev/` projects, then
+`~/clients/`/`~/projects/`, then any project encountered going forward — one at a time.
diff --git a/docs/adr-system/06-eval-scenarios.md b/docs/adr-system/06-eval-scenarios.md
new file mode 100644
index 0000000..b6914ff
--- /dev/null
+++ b/docs/adr-system/06-eval-scenarios.md
@@ -0,0 +1,74 @@
+# Eval Scenario Sketches (held-out, deferred stage)
+
+_Last updated: 2026-07-03 — sketches only; the evaluation itself is a separate later stage
+(locked sequencing in `04-plugin-requirements.md`)._
+
+These sketch the *shape* of the held-out scenarios that will eventually measure requirements 4
+and 5 (unprompted write-trigger recognition, unprompted correct retrieval). They exist now, before
+the plugin skeleton freezes, for one purpose: to catch any plugin surface the eval will need that
+Phases 1–4 don't provide. They are **not** to be run, automated, or informally "tried out" during
+the build — that would contaminate the held-out methodology.
+
+## Ground rules for all scenarios
+
+- **Held-out**: scenario prompts never name the ADR system, the plugin, an ADR ID/title, or the
+  exact constraint text. Relevance must be discovered, not pointed at.
+- **Scored on two axes**: (a) did the agent consult/propose the ADR system *at all* unprompted;
+  (b) did it act on the *specific correct* ADR (retrieval) or propose a *well-scoped* new one
+  (write-trigger) — not merely a plausible-looking neighbor.
+- **Environment**: a real onboarded project (post-pilot) with a populated `docs/adr/`, the
+  SessionStart hook active, and — for graph-layer variants — a built `graphify-out/`.
+- The eval stage will iterate with the `autoresearch` skill (modify → verify → keep/discard)
+  against these scenarios once written out in full.
+
+## Write-trigger scenario shapes
+
+**W1 — Irreversible choice mid-task.** Task: "Add persistence to X" in a project where the
+storage engine choice is genuinely open and consequential (hard to reverse once data exists).
+Never mention ADRs. Pass: the agent proposes recording the engine decision via `/os-adr:create`
+(or asks whether to) before or immediately after implementing. Fail: silently picks one and moves on.
+
+**W2 — Cross-cutting convention change.** Task: a refactor request whose natural solution
+changes a convention used across many files (e.g. error-handling shape, module layout). Pass:
+the agent recognizes the decision outlives the task and proposes an ADR. Fail: does the refactor
+with no record.
+
+**W3 — Contested reversal.** Task: user asks for something that quietly reverses an earlier
+(non-ADR'd, but discoverable-in-code) approach. Pass: agent flags that this is a
+decision-of-record moment and proposes capturing it. This shape overlaps retrieval when the
+earlier approach *is* ADR'd — kept separate deliberately.
+
+## Retrieval scenario shapes
+
+**R1 — Direct conflict with an Accepted ADR.** Task: a change request whose obvious
+implementation violates an existing Accepted ADR whose `affected-paths` cover the files being
+edited. The prompt names neither the ADR nor the constraint. Pass: agent surfaces the specific
+ADR (via `/os-adr:find` or the SessionStart note) and adjusts or pushes back. Fail: implements
+the violation, or cites a different/wrong ADR.
+
+**R2 — Distractor discrimination.** Same as R1, but `docs/adr/` also contains 2–3 plausible
+near-miss ADRs (same component family, different decision; or Superseded status). Pass requires
+retrieving the *correct, Accepted* one — this is the "correct, not merely plausible" axis and
+the reason the status filter and path-match layers exist.
+
+**R3 — Mid-task question.** Partway through an ongoing task, the user asks a "how should we…"
+question already answered by an Accepted ADR, phrased in different vocabulary. Pass: agent
+queries the ADR system mid-task (not only at SessionStart) and answers from the ADR. This is the
+scenario that requires a **mid-task-callable query surface**.
+
+**R4 — Graph-layer reach.** Conflict exists but the edited files are *not* listed in any ADR's
+`affected-paths`; they are one graph hop away (import/reference) from files that are. Pass only
+reachable via the Graphify traversal layer. Also serves as the degradation check: without
+`graphify-out/`, R4 is expected to fail while R1–R3 still pass.
+
+## Plugin-surface requirements derived from these sketches
+
+All are already covered by the planned Phases 1–4 — **no plan change needed**:
+
+| Scenario need | Covering surface |
+| --- | --- |
+| Mid-task query, not SessionStart-only (R3) | `/os-adr:find` skill + `bin/adr-find` CLI (Phase 4) |
+| Discoverability without being told (all) | SessionStart note names both `create` and `find` (Phase 2) |
+| Correct-vs-plausible discrimination (R2) | path/component match + Accepted-by-default status filter (Phase 4, layers 1–2) |
+| One-hop relevance (R4) | Graphify traversal layer with graceful degradation (Phase 4, layer 3) |
+| Low-friction unprompted writing (W1–W3) | one-invocation `/os-adr:create` (Phase 1) |
diff --git a/docs/memory-system/03-architecture-decisions.md b/docs/memory-system/03-architecture-decisions.md
index 20892d8..46b171f 100644
--- a/docs/memory-system/03-architecture-decisions.md
+++ b/docs/memory-system/03-architecture-decisions.md
@@ -589,6 +589,20 @@ _Date: 2026-07-03_
 - **Cross-references**: ADR-016 (os-vault plugin sourced from cc-os git repo, establishing the git-tracked plugin pattern), ADR-018 (plugin marketplace mechanics for local plugins).
 - **Status**: Accepted 2026-07-03. cc-os's CLAUDE.md updated to remove the stricter orchestration section and note that behavior is now global; `os-orchestration` component added to Implemented Components.
 
+## ADR-020 — os-adr plugin built; implementation-time resolutions of the design's open questions
+
+_Date: 2026-07-03_
+
+- **Context**: The `os-adr` plugin (OpenSpec change `add-os-adr-plugin`, built from `docs/adr-system/04-plugin-requirements.md` + `05-plugin-prd.md`) was implemented. Its design deliberately left two questions open — the migration pilot's low-confidence flag-rate threshold, and whether ADR frontmatter needs a Graphify tag/edge convention for retrieval layer 3 — and the pilot forced several heuristic refinements within the design's declared mechanical/LLM boundary.
+- **Decision**: (1) **Pilot gate threshold: 25%** low-confidence flags, recorded in `Adr::MigrationReport::FLAG_RATE_THRESHOLD` and printed in every migration report. Final pilot rates (sandboxed copies of the three surveyed projects): viking-warrior-training-log 0%, delta-refinery 0%, llf-schema 8.3% — gate passed. (2) **No Graphify tag/edge convention for ADRs**: retrieval layer 3 expands the *query* file paths one hop through `graphify-out/graph.json` node `source_file` attributes and matches the expanded set against `affected-paths` frontmatter; ADRs never need to exist as graph nodes. Verified against the real llf-schema graph. (3) **Heuristic-boundary refinements from the pilot**: heading-synonym fallback mappings (Why/Rationale/Problem/Background→Context, Impact/Implications/Trade-offs→Consequences) are mechanical copies but cap the file at `migration_confidence: medium`; status synonyms SETTLED/Confirmed→Accepted, OPEN→Proposed map mechanically; bold-label paragraphs (`**Context.** …`) and bulleted labels (`- **Status:** …`) parse as sections; multi-decision files *without* mechanically splittable unit headings classify `unrecognized` (report-only) rather than converting as one blob.
+- **Rationale**: The threshold was set empirically at pilot time per the PRD ("do not ship on an unmeasured guess"); 25% sits well above the observed worst case (8.3%) while still failing a genuinely broken heuristic run (pre-tightening iterations hit 100% on delta-refinery and llf-schema). Query-path graph expansion keeps ADRs plain markdown with zero indexing obligations — retrieval reuses the existing project graph instead of imposing a new convention on it (and sidesteps the open ADR-014 facet-edge question entirely). Capping synonym-mapped files at medium keeps the mechanical-vs-judgment boundary honest: the copy is mechanical, the mapping is a judgment call, so it stays flagged.
+- **Alternatives rejected**:
+  - **Indexing ADRs as Graphify nodes with a tag/edge convention**: adds an extraction pass and a freshness obligation for no retrieval gain at current corpus sizes; revisit only if one-hop query expansion proves too shallow.
+  - **Treating heading synonyms as high-confidence mechanical fills**: hides an interpretive mapping from the flag mechanism.
+  - **Auto-splitting multi-Status files that lack unit headings**: the split points would themselves be guesses; `unrecognized` + manual handling is the non-destructive answer.
+- **Cross-references**: `docs/adr-system/04-plugin-requirements.md` (locked requirements), `05-plugin-prd.md` (phases), `06-eval-scenarios.md` (held-out eval sketches — the eval itself stays deferred), ADR-013 (build-first/migrate-incrementally precedent), ADR-018 (three-place plugin registration, followed during install), ADR-019 (plugin conventions).
+- **Status**: Accepted 2026-07-03. Plugin live at `cc-os/plugins/os-adr/`, installed as `os-adr@local-plugins`. Rollout order locked: pilot projects' real migration (interactive, per project) → cc-os retrofit → wider.
+
 ## Rejected tools (summary)
 
 | Tool | Why rejected for our use |
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/.openspec.yaml b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/.openspec.yaml
new file mode 100644
index 0000000..43e65ca
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/.openspec.yaml
@@ -0,0 +1,2 @@
+schema: spec-driven
+created: 2026-07-03
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/design.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/design.md
new file mode 100644
index 0000000..6ddcb44
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/design.md
@@ -0,0 +1,197 @@
+# Design: add-os-adr-plugin
+
+## Context
+
+The requirements are locked in `docs/adr-system/04-plugin-requirements.md` and phased in
+`docs/adr-system/05-plugin-prd.md` — this design decides *how* to build them, not *what* they
+are. cc-os already has three global plugins establishing the conventions this one must follow:
+
+- `os-vault` — Python hooks (thin entry points + shared `config.py`/`hook_io.py` modules), hooks
+  wired by absolute path in `settings.json`, verb-first skills, no `commands/` dir.
+- `os-doc-hygiene` — plugin-relative `hooks/hooks.json` using `${CLAUDE_PLUGIN_ROOT}`,
+  deterministic zero-token SessionStart reminder with per-project gitignored state dir
+  (`.dochygiene/`) and snooze, deterministic-scanner + generative-only-where-needed split,
+  invariants.md + golden-example fixtures.
+- `os-orchestration` — minimal SessionStart context injection.
+
+Constraints: single template everywhere (no variants), `docs/adr/` location, mechanical-first
+with LLM reserved for judgment, non-destructive migration, OO per Sandi Metz, eval deferred.
+
+## Goals / Non-Goals
+
+**Goals:**
+
+- A working `os-adr` plugin at `cc-os/plugins/os-adr/`, installed via the `local-plugins`
+  marketplace, covering PRD Phases 1–4 (authoring, session hook, migration, retrieval).
+- Every deterministic behavior testable without a model; every LLM-assisted behavior isolated
+  behind a skill so the mechanical core stays pure.
+- Eval scenario *sketches* written down so the plugin surface is eval-ready.
+
+**Non-Goals:**
+
+- The unprompted-behavior evaluation itself (separate later stage, `autoresearch` loop).
+- Retrofitting cc-os's own ADR file, bulk rollout beyond the pilot projects.
+- Per-project template variants, severity/enforcement fields, multi-editor governance.
+
+## Decisions
+
+### D1 — Plugin layout and packaging
+
+`cc-os/plugins/os-adr/` with:
+
+```
+os-adr/
+  .claude-plugin/plugin.json     # name: os-adr
+  hooks/hooks.json               # SessionStart wiring via ${CLAUDE_PLUGIN_ROOT}
+  hooks/session_start.py         # thin Python entry point (deterministic check)
+  lib/                           # Ruby OO core (adr/ classes)
+  bin/                           # Ruby CLI entry points the skills invoke
+  templates/adr.md               # the customized-Nygard template
+  skills/create/SKILL.md
+  skills/init/SKILL.md
+  skills/migrate/SKILL.md
+  skills/find/SKILL.md
+  tests/                         # unit tests + golden-example fixtures
+  invariants.md
+```
+
+Install follows the CLAUDE.md three-place procedure: symlink
+`~/.claude/plugins/os-adr → cc-os/plugins/os-adr/`, add `"os-adr"`/`"./os-adr"` to the
+`local-plugins` marketplace manifest, enable `os-adr@local-plugins` in `settings.json`, then
+`claude plugin marketplace update` + `claude plugin install`, and verify with
+`claude plugin details` (skills registered is the check that catches manifest mistakes).
+
+**Why:** matches all three existing plugins; the manifest step is the known failure mode
+(ADR-018) so it's explicit in tasks.
+
+### D2 — Hook wiring: plugin `hooks.json`, not `settings.json` absolute paths
+
+Use the `os-doc-hygiene` pattern (`hooks/hooks.json` with `${CLAUDE_PLUGIN_ROOT}`, matcher
+`startup|resume`, short timeout). **Why over the os-vault absolute-path pattern:** it keeps the
+plugin self-contained, survives repo moves, and hook registration then fails *with* plugin
+registration rather than silently diverging from it (the masking problem ADR-018 documented).
+
+### D3 — Language split: Ruby OO core, Python only for the hook entry point
+
+- `lib/` is Ruby, Sandi Metz style: small single-responsibility classes
+  (`Adr::Template`, `Adr::Record`, `Adr::Index`, `Adr::Repository`, `Adr::Detector` (migration
+  shape detection), `Adr::Migrator`, `Adr::Finder` (retrieval pipeline)), dependency injection
+  (paths and IO handed in, no globals), tell-don't-ask.
+- `bin/` exposes thin Ruby CLIs (`adr-new`, `adr-init`, `adr-detect`, `adr-migrate`,
+  `adr-find`) that skills call via Bash. (No standalone `adr-index` — index regeneration is
+  embedded in every writing CLI, per D4.)
+- The SessionStart hook is a thin Python script that does the existence check inline (stat two
+  paths, read one suppression file, emit JSON) — it does **not** shell out to Ruby.
+
+**Why:** the user's standing Ruby preference for the core; Python for the hook matches the
+existing hook pattern and keeps session-start latency free of Ruby interpreter spin-up. The hook
+logic is ~20 lines and purely mechanical, so duplicating "does `docs/adr/` exist" between Python
+and Ruby is acceptable coupling (alternative — Python wrapping the Ruby core — was rejected for
+latency and dependency-surface reasons at session start).
+
+### D4 — Template and ID/index mechanics
+
+- Template (`templates/adr.md`): frontmatter `id`, `date`, `status`
+  (Proposed/Accepted/Superseded/Deprecated), `supersedes`, `superseded-by`, `affected-paths`,
+  `affected-components`, `migration_confidence` (migration-written files only); body sections
+  `## Context`, `## Decision`, `## Consequences`, `## Alternatives rejected`.
+- Files: `docs/adr/NNNN-kebab-title.md`, 4-digit zero-padded sequential ID; next ID = max
+  existing + 1, derived by scanning the directory (never trusting the index).
+- Index: `docs/adr/README.md`, a generated table (ID, title, status, date) between markers;
+  regenerated in full from the directory on every write by `Adr::Index` — never hand-edited,
+  never appended-to incrementally. Setting `superseded-by` on an old ADR when a new one declares
+  `supersedes` is also mechanical (`adr-new` does both edits).
+
+**Why full regeneration over append:** idempotent, self-healing after manual file edits, trivial
+to test against golden fixtures; append-only indexes drift.
+
+### D5 — SessionStart behavior and suppression
+
+- `docs/adr/` + index present → inject one short additionalContext line: the system exists, how
+  to write (`/os-adr:create`) and query (`/os-adr:find`) it. Near-zero tokens, no LLM.
+- Absent → one-line suggestion to run `/os-adr:init` or `/os-adr:migrate`.
+- Suppression: per-project gitignored state dir `.os-adr/` containing `suppress` (permanent
+  opt-out for deliberately ADR-less projects) — same shape as `.dochygiene/`. `/os-adr:init`
+  clears it; hook exits silently when present. A once-per-day snooze stamp (`last_reminded`)
+  additionally rate-limits the absent-case nag, mirroring os-doc-hygiene.
+
+### D6 — Migration pipeline (mechanical/LLM boundary)
+
+`adr-detect` (mechanical) classifies existing content into the surveyed shapes — numbered
+per-file dirs, dated single files, monolithic multi-decision files, prose-embedded — plus
+`unrecognized` as the fallback (reported, never auto-converted). `adr-migrate` then, per source
+decision:
+
+- **Heuristic fill (mechanical):** fields structurally unambiguous in the source — Status/Date
+  from frontmatter or clear headings, title, ID assignment, source-file provenance link.
+- **LLM fill (skill-mediated, flagged):** interpretive fields only — Consequences,
+  Alternatives-rejected, splitting a monolithic file into per-decision units. The Ruby core
+  emits a work manifest of exactly which fields need LLM fill; the `migrate` skill has the model
+  fill them and hands results back to the core for writing. Never invent Decision or Context
+  from nothing — if absent in the source, the field carries an explicit `_not stated in
+  source_` marker and the file is flagged.
+- **Flagging (one mechanism):** `migration_confidence: low|medium|high` frontmatter per file +
+  one `docs/adr/migration-report.md` summarizing per-file confidence, source mapping, and the
+  overall flag rate. No scattered inline comments.
+- **Non-destructive:** new files only; source files untouched. Old-system deletion is a
+  separate, explicit user-approved step the skill offers only after the report is reviewed.
+- **Pilot gate:** run on 2–3 surveyed projects (viking-warrior-training-log, delta-refinery,
+  llf-schema); if the low-confidence flag rate exceeds the threshold set at pilot time
+  (recorded in the report), tighten heuristics before wider rollout. Build-phase gate, not a
+  shipped feature.
+
+### D7 — Retrieval stack (`adr-find` / `/os-adr:find`)
+
+Deterministic layers first, each narrowing the candidate set:
+
+1. Path/component match: query paths (e.g. files being edited) against `affected-paths` /
+   `affected-components` frontmatter across `docs/adr/`.
+2. Status filter: `Accepted` only by default (`--all-statuses` to override).
+3. Graphify traversal: when `graphify-out/` exists, expand the *query* paths one hop via
+   graph-node `source_file` attributes and re-match the expanded set against `affected-paths` —
+   ADRs themselves are never graph nodes (resolved during Phase 4; see ADR-020). Reuses
+   os-vault's project-graph infrastructure; degrades gracefully to layers 1–2 when absent.
+4. AI judgment (skill-mediated) only over the narrowed set — the `find` skill presents the
+   candidates and the model picks/ranks; it never sweeps the full corpus.
+
+This is a mid-task-callable surface (skill + CLI), satisfying the PRD's eval-readiness note —
+retrieval is not locked to SessionStart.
+
+### D8 — Eval-readiness sketches
+
+A `docs/adr-system/06-eval-scenarios.md` sketch (write-trigger + retrieval scenario shapes per
+the PRD) is written during this change, and the SessionStart context line explicitly names both
+surfaces (`create`, `find`) so an agent can discover them unprompted. The eval itself is out of
+scope.
+
+## Risks / Trade-offs
+
+- [Duplicated existence-check logic between Python hook and Ruby core] → it's two path checks;
+  covered by a shared fixture test asserting both agree on the same tree.
+- [Ruby runtime required at skill time] → acceptable: this is the user's own machine and
+  standing preference; hook path (the always-runs surface) has no Ruby dependency.
+- [Migration LLM fills may be wrong] → confidence flags + report + pilot gate before rollout;
+  non-destructiveness means a bad migration is discardable (delete `docs/adr/`, rerun).
+- [Graphify layer depends on os-vault onboarding state] → retrieval degrades to layers 1–2;
+  layer 3 is an enhancement, not a dependency.
+- [Monolithic-file splitting is the hardest detection case] → treated as LLM-boundary work with
+  per-unit flags; cc-os's own 19-ADR file is deliberately excluded from the pilot.
+- [Index regeneration overwrites manual index edits] → by design (index is declared
+  never-hand-edited in its own header comment).
+
+## Migration Plan
+
+Build order matches PRD phases: Phase 1 (template+create+index) → Phase 2 (hook) → Phase 3
+(migration) → Phase 4 (retrieval), each verifiable independently. Install/registration happens
+after Phase 1 so `create` is dogfoodable early. Rollback: uninstall plugin + remove symlink and
+manifest entry; projects keep their `docs/adr/` files (plain markdown, tool-independent).
+
+## Open Questions
+
+- Exact low-confidence flag-rate threshold for the pilot gate — set empirically at pilot time
+  and recorded in the migration report (deliberately not guessed now).
+- ~~Whether ADR frontmatter needs a dedicated Graphify tag/edge convention for layer-3
+  traversal~~ — **Resolved during Phase 4 (2026-07-03):** no convention needed. Layer 3 expands
+  the *query* paths one hop via graph-node `source_file`s; ADRs never need to be graph nodes and
+  match purely on `affected-paths` frontmatter (verified against the real llf-schema project
+  graph; recorded in ADR-020).
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/proposal.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/proposal.md
new file mode 100644
index 0000000..2de14d0
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/proposal.md
@@ -0,0 +1,76 @@
+# Proposal: add-os-adr-plugin
+
+## Why
+
+~43% of the user's projects carry ADR-like content in 5+ mutually-incompatible shapes; the rest
+have none, and no project uses ADR tooling. Because AI agents (not just the user) work across
+every project, decision records must be discoverable and writable *consistently* by an agent
+that has never seen a given repo — and cc-os's own 19-ADR/601-line monolithic file is already
+near the size where research says monolithic logs stop working. The requirements are locked
+(`docs/adr-system/04-plugin-requirements.md`) and the PRD is written
+(`docs/adr-system/05-plugin-prd.md`); this change builds the plugin.
+
+## What Changes
+
+- New global Claude Code plugin **`os-adr`** at `cc-os/plugins/os-adr/`, following the
+  established cc-os plugin conventions: `os-` prefix, verb-first skills, no `commands/`
+  dispatcher, symlink into `~/.claude/plugins/`, registered in the `local-plugins` marketplace
+  manifest (the three-place rename/registration procedure in CLAUDE.md).
+- **Phase 1 — authoring**: customized-Nygard template (Status/Context/Decision/Consequences +
+  Alternatives rejected; frontmatter with `supersedes`/`superseded-by`,
+  `affected-paths`/`affected-components`, `date`, sequential ID), `docs/adr/NNNN-title.md`
+  one-file-per-decision convention, mechanically-maintained index file, and a `new` skill that
+  writes a correctly templated + indexed ADR in one invocation. No LLM in the mechanical path.
+- **Phase 2 — SessionStart hook**: deterministic existence check for `docs/adr/` + index. Present
+  → near-zero-token context note; absent → suggest setup, with a per-project suppression flag so
+  a deliberately ADR-less project doesn't re-nag (same shape as `os-doc-hygiene` snooze state).
+- **Phase 3 — non-destructive migration**: a `migrate` skill converting existing ADR content of
+  any surveyed shape into the new template alongside (never replacing) the old system; heuristic
+  fill for structurally unambiguous fields, LLM fill only for interpretive fields, one consistent
+  uncertainty-flagging mechanism (`migration_confidence` frontmatter + a migration report file);
+  deletion of the old system only as an explicit, separate user-approved step. Piloted on 2–3
+  real projects with an uncertain-flag-rate gate before wider rollout.
+- **Phase 4 — retrieval**: deterministic-first layering — path/component frontmatter match →
+  status filter (Accepted by default) → Graphify graph traversal (reusing `os-vault`
+  infrastructure) → AI judgment only over the already-narrowed candidate set. Exposed as a query
+  surface callable mid-task, not just at SessionStart.
+- Implementation style: OO per Sandi Metz principles; Ruby-preferred core with Python acceptable
+  for hook entry points, decided per-component (thin hook scripts wrapping an OO core).
+- Held-out eval scenarios (write-trigger, retrieval) are **sketched** as part of this change so
+  the plugin surface isn't missing what the eval needs — but the evaluation itself stays a
+  separate later stage (locked sequencing in `04-plugin-requirements.md`).
+
+## Capabilities
+
+### New Capabilities
+
+- `adr-authoring`: the ADR template, `docs/adr/NNNN-title.md` + index location convention, and
+  the one-invocation new-ADR generator skill with mechanical index maintenance.
+- `adr-session-awareness`: the deterministic SessionStart existence-check hook — minimal context
+  injection when the ADR system exists, setup suggestion with per-project suppression when it
+  doesn't.
+- `adr-migration`: non-destructive conversion of existing ADR content (all surveyed shapes plus
+  a fallback) into the new template, with the mechanical/LLM field-mapping boundary, single
+  uncertainty-flagging mechanism, and pilot gate.
+- `adr-retrieval`: the deterministic-first retrieval stack (frontmatter path match → status
+  filter → Graphify traversal → narrowed AI judgment) exposed as a mid-task query surface.
+- `adr-plugin-packaging`: plugin structure, naming, marketplace registration, and install
+  procedure per cc-os conventions (`os-` prefix, verb-first skills, symlink + `local-plugins`
+  manifest + `settings.json` enablement).
+
+### Modified Capabilities
+
+<!-- none — os-adr is a new plugin; no existing spec's requirements change -->
+
+## Impact
+
+- New code: `cc-os/plugins/os-adr/` (hooks, skills, OO core, template assets, tests/fixtures).
+- Host config: `~/.claude/plugins/os-adr` symlink, `~/.claude/plugins/.claude-plugin/marketplace.json`
+  entry, `settings.json` `enabledPlugins` (and hook wiring if hooks are registered by absolute
+  path like os-vault, or via plugin `hooks.json` like os-doc-hygiene — decided in design).
+- Reuses: Graphify / `os-vault` graph infrastructure for the retrieval layer; `os-doc-hygiene`
+  patterns for deterministic SessionStart reminders and snooze state.
+- Target projects (pilot, later phases): viking-warrior-training-log, delta-refinery, llf-schema;
+  cc-os itself is retrofitted only after the pilot, per the locked rollout order.
+- Out of scope: per-project template variants, multi-editor governance, enforceable-constraints
+  mirroring, the unprompted-behavior evaluation itself, and retrofitting cc-os's ADR file.
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-authoring/spec.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-authoring/spec.md
new file mode 100644
index 0000000..bc4b278
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-authoring/spec.md
@@ -0,0 +1,70 @@
+# adr-authoring
+
+## ADDED Requirements
+
+### Requirement: Standard ADR template
+The plugin SHALL provide a single customized-Nygard ADR template used identically in every
+project, with frontmatter fields `id`, `date`, `status`, `supersedes`, `superseded-by`,
+`affected-paths`, `affected-components`, and body sections `Context`, `Decision`,
+`Consequences`, and `Alternatives rejected`. The plugin SHALL NOT provide per-project or
+per-client template variants.
+
+#### Scenario: Template contains all required fields
+- **WHEN** a new ADR is generated in any project
+- **THEN** the resulting file contains every frontmatter field and body section listed above,
+  in template order
+
+#### Scenario: No project-specific variation
+- **WHEN** ADRs are generated in two different projects
+- **THEN** both files follow the identical template structure
+
+### Requirement: One file per decision at the standard location
+ADRs SHALL be stored one-file-per-decision at `docs/adr/NNNN-kebab-title.md`, where `NNNN` is a
+4-digit zero-padded sequential ID derived by scanning the existing directory contents (max
+existing ID + 1), never by trusting the index file.
+
+#### Scenario: Sequential ID assignment
+- **WHEN** `docs/adr/` contains ADRs `0001` and `0002` and a new ADR titled "Use Postgres" is
+  created
+- **THEN** the new file is `docs/adr/0003-use-postgres.md`
+
+#### Scenario: ID derived from directory, not index
+- **WHEN** the index file is stale or missing but `docs/adr/0007-*.md` exists
+- **THEN** the next generated ADR receives ID `0008`
+
+### Requirement: Mechanically maintained index
+The plugin SHALL maintain an index file at `docs/adr/README.md` listing every ADR's ID, title,
+status, and date. The index SHALL be regenerated in full from the directory contents on every
+ADR write, with no LLM involvement, and SHALL declare itself generated/never-hand-edited.
+
+#### Scenario: Index updated on ADR creation
+- **WHEN** a new ADR is created via the generator
+- **THEN** the index file is regenerated and contains a row for the new ADR
+
+#### Scenario: Index self-heals
+- **WHEN** an ADR file was added or edited manually and the generator next runs
+- **THEN** the regenerated index reflects the actual directory contents
+
+### Requirement: One-invocation ADR creation skill
+The plugin SHALL provide a `create` skill (`/os-adr:create`) that produces a correctly templated,
+correctly numbered, and indexed ADR in a single invocation, and an `init` skill (`/os-adr:init`)
+that sets up `docs/adr/` plus the index in a project that has none.
+
+#### Scenario: Create in one invocation
+- **WHEN** the user (or agent) invokes `/os-adr:create` with a decision to record
+- **THEN** a new templated ADR file exists under `docs/adr/` and the index includes it, with no
+  further steps required
+
+#### Scenario: Init on a fresh project
+- **WHEN** `/os-adr:init` runs in a project without `docs/adr/`
+- **THEN** `docs/adr/` and an empty index exist and any session-hook suppression flag for the
+  project is cleared
+
+### Requirement: Supersession is mechanical
+WHEN a new ADR declares `supersedes: NNNN`, the generator SHALL set `superseded-by` on the
+referenced ADR and update its status accordingly, without manual edits.
+
+#### Scenario: Superseding an existing ADR
+- **WHEN** a new ADR `0009` is created with `supersedes: 0004`
+- **THEN** ADR `0004`'s frontmatter gains `superseded-by: 0009` and status `Superseded`, and the
+  regenerated index shows both statuses
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-migration/spec.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-migration/spec.md
new file mode 100644
index 0000000..0bddcb2
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-migration/spec.md
@@ -0,0 +1,71 @@
+# adr-migration
+
+## ADDED Requirements
+
+### Requirement: Shape detection with a fallback
+The migration tooling SHALL mechanically detect existing ADR-like content and classify it into
+the surveyed shapes — numbered per-file directories, dated single files, monolithic
+multi-decision files, prose embedded in other docs — plus an `unrecognized` fallback
+classification. Unrecognized content SHALL be reported for manual handling, never auto-converted.
+
+#### Scenario: Known shape detected
+- **WHEN** detection runs on a project with a monolithic multi-decision file
+- **THEN** the file is classified as monolithic and its individual decisions are enumerated as
+  migration units
+
+#### Scenario: Unknown shape falls back safely
+- **WHEN** detection encounters ADR-like content matching no known shape
+- **THEN** it is classified `unrecognized` and listed in the report without any conversion
+  attempt
+
+### Requirement: Mechanical/LLM field-mapping boundary
+During conversion, fields that are structurally unambiguous in the source (Status and Date from
+frontmatter or clear headings; title; ID assignment; source provenance) SHALL be filled
+mechanically. Only interpretive fields (Consequences, Alternatives rejected, monolithic-file
+decision splitting) MAY be LLM-filled, and every LLM-filled field SHALL be flagged. The
+migration SHALL never invent a Decision or Context field absent from the source — such fields
+SHALL carry an explicit not-stated-in-source marker and lower the file's confidence.
+
+#### Scenario: Unambiguous field filled mechanically
+- **WHEN** a source ADR has a `Status: Accepted` heading and a date
+- **THEN** the converted file's `status` and `date` are set without any LLM call
+
+#### Scenario: Missing core field is not invented
+- **WHEN** a source decision has no discernible Context
+- **THEN** the converted Context section contains a not-stated-in-source marker and the file is
+  flagged low-confidence
+
+### Requirement: Single uncertainty-flagging mechanism
+Migration uncertainty SHALL be surfaced through exactly one mechanism: a
+`migration_confidence: low|medium|high` frontmatter field per converted file, plus one
+migration report file (`docs/adr/migration-report.md`) summarizing per-file confidence, the
+source-to-new mapping, and the overall low-confidence flag rate. Scattered inline uncertainty
+comments SHALL NOT be used.
+
+#### Scenario: Report aggregates confidence
+- **WHEN** a migration completes
+- **THEN** the report lists every converted file with its confidence, its source location, and
+  the overall flag rate
+
+### Requirement: Non-destructive, dual-existence conversion
+Migration SHALL only write new files under `docs/adr/`; existing ADR content SHALL never be
+deleted, moved, or edited by the migration. Removal of the old system SHALL happen only as an
+explicit, separate, user-approved step offered after the migration report exists.
+
+#### Scenario: Old system untouched
+- **WHEN** a migration run completes
+- **THEN** every pre-existing ADR source file is byte-identical to its pre-migration state
+
+#### Scenario: Deletion requires explicit approval
+- **WHEN** the user has not explicitly approved old-system removal
+- **THEN** no tooling path deletes the old content, and the removal offer appears only after the
+  report is available for review
+
+### Requirement: Pilot gate before broad rollout
+The migration SHALL be piloted on 2–3 real surveyed projects, and the low-confidence flag rate
+SHALL be measured against an explicit threshold recorded at pilot time before the heuristics are
+treated as good enough for wider rollout.
+
+#### Scenario: High flag rate blocks rollout
+- **WHEN** a pilot run's low-confidence rate exceeds the recorded threshold
+- **THEN** the heuristics are tightened and the pilot re-run before migrating further projects
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-plugin-packaging/spec.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-plugin-packaging/spec.md
new file mode 100644
index 0000000..7a03dab
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-plugin-packaging/spec.md
@@ -0,0 +1,61 @@
+# adr-plugin-packaging
+
+## ADDED Requirements
+
+### Requirement: Plugin naming and skill naming conventions
+The plugin SHALL be named `os-adr` (kebab-case, `os-` prefix) and its skills SHALL use
+verb-first kebab-case names invoked as `/os-adr:<verb>` (`create`, `init`, `migrate`, `find`),
+with no `commands/` dispatcher directory — matching the cc-os plugin/skill naming convention.
+
+#### Scenario: Skills register under verb-first names
+- **WHEN** the plugin is installed and `claude plugin details os-adr@local-plugins` is run
+- **THEN** the listed skills are `create`, `init`, `migrate`, and `find`, and no `commands/`
+  directory exists in the plugin
+
+### Requirement: Source of truth in cc-os with symlinked install
+The plugin source SHALL live git-tracked at `cc-os/plugins/os-adr/` and be installed via a
+symlink `~/.claude/plugins/os-adr` pointing at it, matching the `os-vault`/`os-doc-hygiene`/
+`os-orchestration` pattern.
+
+#### Scenario: Symlinked install
+- **WHEN** installation is complete
+- **THEN** `~/.claude/plugins/os-adr` is a symlink resolving to the git-tracked plugin directory
+
+### Requirement: Full three-place marketplace registration
+Installation SHALL update all three registration surfaces — the plugin directory/symlink, the
+`local-plugins` marketplace manifest (`~/.claude/plugins/.claude-plugin/marketplace.json`), and
+`settings.json` `enabledPlugins` — then refresh plugin-manager state via
+`claude plugin marketplace update` and `claude plugin install`, and verify skill registration
+with `claude plugin list` and `claude plugin details`.
+
+#### Scenario: Skills resolve after install
+- **WHEN** the registration procedure completes
+- **THEN** `claude plugin list` shows `os-adr@local-plugins` enabled and its skills are
+  invocable as slash commands in a fresh session
+
+### Requirement: Hooks wired through plugin hooks.json
+Plugin hooks SHALL be declared in `hooks/hooks.json` using `${CLAUDE_PLUGIN_ROOT}`-relative
+commands (the `os-doc-hygiene` pattern), not by absolute paths in `settings.json`, so hook
+registration cannot silently outlive or diverge from plugin registration.
+
+#### Scenario: Hook path is plugin-relative
+- **WHEN** the plugin's hook configuration is inspected
+- **THEN** the SessionStart command references `${CLAUDE_PLUGIN_ROOT}` and no os-adr hook entry
+  exists in `settings.json`
+
+### Requirement: OO core with tested deterministic behavior
+The plugin's core logic SHALL be implemented as an object-oriented Ruby library (small
+single-responsibility classes, dependency-injected paths/IO, no globals) exposed through thin
+CLI entry points, with a thin Python script only for the SessionStart hook. All deterministic
+behaviors (ID assignment, index regeneration, shape detection, existence check, retrieval layers
+1–2) SHALL have automated tests runnable without any LLM, including golden-example fixtures.
+
+#### Scenario: Deterministic tests run model-free
+- **WHEN** the plugin's test suite runs on a machine with no model access
+- **THEN** all tests for ID assignment, index regeneration, detection, existence check, and
+  frontmatter retrieval pass
+
+#### Scenario: Skills delegate to CLI entry points
+- **WHEN** a skill performs a mechanical operation (numbering, indexing, detection, filtering)
+- **THEN** it does so by invoking a `bin/` CLI rather than re-implementing the logic in prompt
+  instructions
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-retrieval/spec.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-retrieval/spec.md
new file mode 100644
index 0000000..9b0a8d3
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-retrieval/spec.md
@@ -0,0 +1,49 @@
+# adr-retrieval
+
+## ADDED Requirements
+
+### Requirement: Deterministic-first retrieval layering
+ADR retrieval SHALL narrow candidates through ordered layers: (1) path/component match of query
+inputs against each ADR's `affected-paths`/`affected-components` frontmatter; (2) status filter
+admitting only `Accepted` ADRs by default, with an explicit override to include other statuses;
+(3) Graphify graph traversal when a project graph exists; (4) AI judgment applied only to the
+already-narrowed candidate set. AI judgment SHALL never be applied to the full ADR corpus.
+
+#### Scenario: Path match narrows first
+- **WHEN** retrieval is queried with file paths being edited
+- **THEN** ADRs whose `affected-paths` match are selected before any model involvement
+
+#### Scenario: Default status filter
+- **WHEN** retrieval runs without an override
+- **THEN** only `Accepted` ADRs appear in the candidate set
+
+#### Scenario: AI sees only the narrowed set
+- **WHEN** the AI-judgment layer runs
+- **THEN** it receives only the candidates surviving layers 1–3, not the full corpus
+
+### Requirement: Graph layer degrades gracefully
+WHEN no `graphify-out/` project graph exists (or ADRs are not yet indexed as graph nodes),
+retrieval SHALL still function using layers 1, 2, and 4, without error.
+
+#### Scenario: No project graph present
+- **WHEN** retrieval runs in a project that has never been onboarded to Graphify
+- **THEN** results are produced from frontmatter matching and status filtering alone
+
+### Requirement: Mid-task-callable query surface
+Retrieval SHALL be exposed as a `find` skill (`/os-adr:find`) backed by a deterministic CLI,
+callable at any point during a task — not only at SessionStart — so an agent can check
+in-progress work against existing decisions.
+
+#### Scenario: Query during a task
+- **WHEN** an agent mid-task invokes `/os-adr:find` with the files or topic it is working on
+- **THEN** it receives the relevant Accepted ADR candidates for those inputs
+
+### Requirement: Eval-readiness sketches accompany the build
+Held-out evaluation scenario shapes (write-trigger and retrieval) SHALL be sketched in a design
+doc before the plugin skeleton is finalized, and any plugin surface those sketches require SHALL
+be part of the initial build. The evaluation itself SHALL NOT be part of this change.
+
+#### Scenario: Sketches exist before skeleton freeze
+- **WHEN** the plugin skeleton is finalized
+- **THEN** a scenario-sketch document exists covering at least one write-trigger and one
+  retrieval scenario shape
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-session-awareness/spec.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-session-awareness/spec.md
new file mode 100644
index 0000000..6a6fffb
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/specs/adr-session-awareness/spec.md
@@ -0,0 +1,45 @@
+# adr-session-awareness
+
+## ADDED Requirements
+
+### Requirement: Deterministic SessionStart existence check
+The plugin SHALL register a SessionStart hook (matcher `startup|resume`) that deterministically
+checks for `docs/adr/` and its index file in the current project. The hook SHALL make no LLM
+call, SHALL complete within its configured timeout (5s), and SHALL add no perceptible
+session-start latency.
+
+#### Scenario: Hook is deterministic
+- **WHEN** the SessionStart hook runs
+- **THEN** its output is a pure function of the project's filesystem state (ADR dir, index,
+  suppression state) with no model invocation
+
+### Requirement: Minimal context injection when the ADR system exists
+WHEN `docs/adr/` and the index file both exist, the hook SHALL inject a near-zero-token
+additionalContext note stating that the ADR system exists and naming both usage surfaces:
+writing (`/os-adr:create`) and retrieval (`/os-adr:find`).
+
+#### Scenario: Present project gets the usage note
+- **WHEN** a session starts in a project with `docs/adr/` and an index
+- **THEN** the injected context is a short note (a few lines at most) naming `/os-adr:create`
+  and `/os-adr:find`
+
+### Requirement: Setup suggestion with suppression when absent
+WHEN the ADR system is absent, the hook SHALL notify the user and suggest `/os-adr:init` or
+`/os-adr:migrate` — unless a per-project suppression flag is set, in which case it SHALL stay
+silent. The suggestion SHALL also be rate-limited to at most once per day per project via a
+snooze timestamp. Suppression state SHALL live in a gitignored per-project state directory
+(`.os-adr/`), following the `os-doc-hygiene` state-dir shape.
+
+#### Scenario: Absent and unsuppressed
+- **WHEN** a session starts in a project with no `docs/adr/` and no suppression flag, and no
+  reminder was shown today
+- **THEN** the hook emits a one-line suggestion to run `/os-adr:init` or `/os-adr:migrate` and
+  stamps the snooze timestamp
+
+#### Scenario: Suppressed project stays silent
+- **WHEN** a session starts in a project whose `.os-adr/suppress` flag is set
+- **THEN** the hook emits nothing
+
+#### Scenario: Same-day repeat session
+- **WHEN** a second session starts the same day in an absent, unsuppressed project
+- **THEN** no repeat suggestion is emitted
diff --git a/openspec/changes/archive/2026-07-03-add-os-adr-plugin/tasks.md b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/tasks.md
new file mode 100644
index 0000000..a8547a2
--- /dev/null
+++ b/openspec/changes/archive/2026-07-03-add-os-adr-plugin/tasks.md
@@ -0,0 +1,51 @@
+# Tasks: add-os-adr-plugin
+
+## 1. Plugin skeleton and eval-readiness sketches
+
+- [x] 1.1 Sketch held-out eval scenario shapes (write-trigger + retrieval) in `docs/adr-system/06-eval-scenarios.md`; note any plugin surface they require beyond Phases 1–4 and fold it into this plan before proceeding
+- [x] 1.2 Create `cc-os/plugins/os-adr/` skeleton: `.claude-plugin/plugin.json` (name `os-adr`), `hooks/`, `lib/`, `bin/`, `templates/`, `skills/`, `tests/`, `invariants.md` stub
+- [x] 1.3 Write `templates/adr.md` — customized-Nygard template with full frontmatter (`id`, `date`, `status`, `supersedes`, `superseded-by`, `affected-paths`, `affected-components`) and body sections (Context, Decision, Consequences, Alternatives rejected)
+
+## 2. Phase 1 — authoring core (Ruby, mechanical)
+
+- [x] 2.1 Implement `Adr::Record` (parse/serialize one ADR file: frontmatter + sections) with unit tests
+- [x] 2.2 Implement `Adr::Repository` (scan `docs/adr/`, derive next 4-digit ID from directory contents, injected root path) with unit tests including stale/missing-index cases
+- [x] 2.3 Implement `Adr::Index` (full regeneration of `docs/adr/README.md` table between markers, never-hand-edited header) with golden-fixture tests including the self-heal-after-manual-edit case
+- [x] 2.4 Implement `Adr::Template` + `bin/adr-new` (create templated ADR, regenerate index, mechanical supersession: set `superseded-by` + status on the superseded ADR) with tests
+- [x] 2.5 Implement `bin/adr-init` (create `docs/adr/` + empty index, clear `.os-adr/suppress`) with tests
+- [x] 2.6 Write `skills/create/SKILL.md` and `skills/init/SKILL.md` — skills gather decision content, then delegate all mechanical work to `bin/adr-new`/`bin/adr-init`
+
+## 3. Phase 1 — install and register the plugin
+
+- [x] 3.1 Symlink `~/.claude/plugins/os-adr → cc-os/plugins/os-adr/`
+- [x] 3.2 Add `os-adr`/`./os-adr` to `~/.claude/plugins/.claude-plugin/marketplace.json` and enable `os-adr@local-plugins` in `settings.json`
+- [x] 3.3 Run `claude plugin marketplace update local-plugins` + `claude plugin install os-adr@local-plugins`; verify with `claude plugin list` and `claude plugin details` that all skills resolve (ADR-018 failure mode check)
+- [x] 3.4 Dogfood: in a scratch project, run `/os-adr:init` then `/os-adr:create` and confirm one-invocation creation with correct template, ID, and index (CLI path the skills delegate to, verified in scratchpad; skills confirmed registered via `claude plugin details` — interactive slash invocation re-checked in 4.4's fresh-session pass)
+
+## 4. Phase 2 — SessionStart hook
+
+- [x] 4.1 Implement `hooks/session_start.py` — deterministic existence check (`docs/adr/` + index), present→minimal usage note naming `/os-adr:create` and `/os-adr:find`, absent→init/migrate suggestion gated by `.os-adr/suppress` flag and once-per-day snooze stamp
+- [x] 4.2 Write `hooks/hooks.json` (SessionStart, matcher `startup|resume`, `${CLAUDE_PLUGIN_ROOT}` command, 5s timeout); confirm no os-adr entries needed in `settings.json`
+- [x] 4.3 Add hook tests: shared fixture tree asserting the Python check and Ruby `Adr::Repository` agree on existence; suppression, snooze, and present/absent output cases
+- [x] 4.4 Verify in fresh sessions: present project injects the note, absent project nags once then snoozes, suppressed project stays silent (verified by invoking the hook exactly as the harness does, in fixture git projects; hook registration confirmed via `claude plugin details` — Hooks (1) SessionStart)
+
+## 5. Phase 3 — migration
+
+- [x] 5.1 Implement `Adr::Detector` + `bin/adr-detect` — classify existing content into the four surveyed shapes + `unrecognized` fallback (report-only), enumerate migration units, with fixture tests per shape
+- [x] 5.2 Implement `Adr::Migrator` + `bin/adr-migrate` mechanical pass — heuristic-fill unambiguous fields (status/date/title/ID/provenance), emit a work manifest of fields needing LLM fill, write nothing outside `docs/adr/`, with byte-identical-source tests
+- [x] 5.3 Implement flagging + report: `migration_confidence` frontmatter, not-stated-in-source markers for absent Decision/Context, and `docs/adr/migration-report.md` generation (per-file confidence, source mapping, flag rate) with tests
+- [x] 5.4 Write `skills/migrate/SKILL.md` — runs detect, has the model fill only manifest-listed interpretive fields, hands results back to the core for writing, presents the report, and offers old-system deletion only as an explicit separate user-approved step
+- [x] 5.5 Pilot the migration on 2–3 surveyed projects (viking-warrior-training-log, delta-refinery, llf-schema); record the low-confidence flag rate and threshold in each report; tighten heuristics and re-run if the rate exceeds the threshold (gate — do not proceed to broad use until passed). **Pilot 2026-07-03 (sandboxed copies, real repos untouched): threshold 25%; two tightening rounds (bulleted labels, bold-label paragraphs `**Context.**`, Why/Rationale→Context fallback at medium confidence, SETTLED/OPEN status mapping, multi-Status-no-headings→unrecognized guard). Final rates: viking 0% (6/6 high), delta-refinery 0% (12 medium), llf-schema 8.3% (1 low: the superseded 2026-06-04 index doc, legitimately lacking Decision/Context). Gate PASSED. Real-repo migration happens interactively via `/os-adr:migrate` per project.**
+
+## 6. Phase 4 — retrieval
+
+- [x] 6.1 Implement `Adr::Finder` + `bin/adr-find` layers 1–2: path/component frontmatter matching and Accepted-by-default status filter (with override flag), with tests
+- [x] 6.2 Add layer 3: Graphify traversal over `graphify-out/` when present, expanding the candidate set via graph edges; graceful degradation to layers 1–2 when absent, with tests for both paths; resolve whether ADR frontmatter needs a tag/edge convention against the llf-schema graph. **Resolved: no tag/edge convention needed — layer 3 expands the *query* paths one hop via graph-node `source_file` attributes; ADRs never need to be graph nodes and match purely on `affected-paths` frontmatter.**
+- [x] 6.3 Write `skills/find/SKILL.md` — mid-task-callable: run the deterministic CLI, apply AI judgment only over the returned candidate set, never the full corpus
+- [x] 6.4 Verify retrieval end-to-end in a piloted project: query with real edited-file paths and confirm the correct Accepted ADR surfaces (llf-schema pilot copy + its real project graph: `tests/IdMinterTest.php` query surfaced the correct ADR via `graph` match; `src/IdMinter.php` via `path`; unrelated path → 0 candidates)
+
+## 7. Wrap-up
+
+- [x] 7.1 Fill in `invariants.md` (non-destructiveness, index-regeneration, no-LLM-in-hook, single-flag-mechanism invariants) and ensure the test suite covers each (9 invariants, each mapped to its covering tests)
+- [x] 7.2 Update cc-os `CLAUDE.md` Implemented Components with the os-adr pointer and note the locked rollout order (pilot → cc-os retrofit → wider) as remaining work
+- [x] 7.3 Record the build in `docs/memory-system`-style ADR log (`03-architecture-decisions.md`) if any design decision changed during implementation; keep `docs/adr-system/` docs' status lines current (ADR-020 added: 25% pilot threshold, no-graph-convention resolution, heuristic-boundary refinements; `00-README.md` status updated, `06-eval-scenarios.md` added to its doc list)
diff --git a/openspec/specs/adr-authoring/spec.md b/openspec/specs/adr-authoring/spec.md
new file mode 100644
index 0000000..9e73886
--- /dev/null
+++ b/openspec/specs/adr-authoring/spec.md
@@ -0,0 +1,74 @@
+# Spec: ADR Authoring
+
+## Purpose
+
+Defines how Architecture Decision Records are created and maintained by the `os-adr` plugin: one standard template, one file per decision at a standard location, a mechanically maintained index, one-invocation creation skills, and mechanical supersession.
+
+## Requirements
+
+### Requirement: Standard ADR template
+The plugin SHALL provide a single customized-Nygard ADR template used identically in every
+project, with frontmatter fields `id`, `date`, `status`, `supersedes`, `superseded-by`,
+`affected-paths`, `affected-components`, and body sections `Context`, `Decision`,
+`Consequences`, and `Alternatives rejected`. The plugin SHALL NOT provide per-project or
+per-client template variants.
+
+#### Scenario: Template contains all required fields
+- **WHEN** a new ADR is generated in any project
+- **THEN** the resulting file contains every frontmatter field and body section listed above,
+  in template order
+
+#### Scenario: No project-specific variation
+- **WHEN** ADRs are generated in two different projects
+- **THEN** both files follow the identical template structure
+
+### Requirement: One file per decision at the standard location
+ADRs SHALL be stored one-file-per-decision at `docs/adr/NNNN-kebab-title.md`, where `NNNN` is a
+4-digit zero-padded sequential ID derived by scanning the existing directory contents (max
+existing ID + 1), never by trusting the index file.
+
+#### Scenario: Sequential ID assignment
+- **WHEN** `docs/adr/` contains ADRs `0001` and `0002` and a new ADR titled "Use Postgres" is
+  created
+- **THEN** the new file is `docs/adr/0003-use-postgres.md`
+
+#### Scenario: ID derived from directory, not index
+- **WHEN** the index file is stale or missing but `docs/adr/0007-*.md` exists
+- **THEN** the next generated ADR receives ID `0008`
+
+### Requirement: Mechanically maintained index
+The plugin SHALL maintain an index file at `docs/adr/README.md` listing every ADR's ID, title,
+status, and date. The index SHALL be regenerated in full from the directory contents on every
+ADR write, with no LLM involvement, and SHALL declare itself generated/never-hand-edited.
+
+#### Scenario: Index updated on ADR creation
+- **WHEN** a new ADR is created via the generator
+- **THEN** the index file is regenerated and contains a row for the new ADR
+
+#### Scenario: Index self-heals
+- **WHEN** an ADR file was added or edited manually and the generator next runs
+- **THEN** the regenerated index reflects the actual directory contents
+
+### Requirement: One-invocation ADR creation skill
+The plugin SHALL provide a `create` skill (`/os-adr:create`) that produces a correctly templated,
+correctly numbered, and indexed ADR in a single invocation, and an `init` skill (`/os-adr:init`)
+that sets up `docs/adr/` plus the index in a project that has none.
+
+#### Scenario: Create in one invocation
+- **WHEN** the user (or agent) invokes `/os-adr:create` with a decision to record
+- **THEN** a new templated ADR file exists under `docs/adr/` and the index includes it, with no
+  further steps required
+
+#### Scenario: Init on a fresh project
+- **WHEN** `/os-adr:init` runs in a project without `docs/adr/`
+- **THEN** `docs/adr/` and an empty index exist and any session-hook suppression flag for the
+  project is cleared
+
+### Requirement: Supersession is mechanical
+WHEN a new ADR declares `supersedes: NNNN`, the generator SHALL set `superseded-by` on the
+referenced ADR and update its status accordingly, without manual edits.
+
+#### Scenario: Superseding an existing ADR
+- **WHEN** a new ADR `0009` is created with `supersedes: 0004`
+- **THEN** ADR `0004`'s frontmatter gains `superseded-by: 0009` and status `Superseded`, and the
+  regenerated index shows both statuses
diff --git a/openspec/specs/adr-migration/spec.md b/openspec/specs/adr-migration/spec.md
new file mode 100644
index 0000000..82b91ae
--- /dev/null
+++ b/openspec/specs/adr-migration/spec.md
@@ -0,0 +1,75 @@
+# Spec: ADR Migration
+
+## Purpose
+
+Defines how existing ADR-like content in a project is migrated into the standard `os-adr` format: mechanical shape detection with a safe fallback, a strict mechanical/LLM field-mapping boundary, a single uncertainty-flagging mechanism, non-destructive dual-existence conversion, and a pilot gate before broad rollout.
+
+## Requirements
+
+### Requirement: Shape detection with a fallback
+The migration tooling SHALL mechanically detect existing ADR-like content and classify it into
+the surveyed shapes — numbered per-file directories, dated single files, monolithic
+multi-decision files, prose embedded in other docs — plus an `unrecognized` fallback
+classification. Unrecognized content SHALL be reported for manual handling, never auto-converted.
+
+#### Scenario: Known shape detected
+- **WHEN** detection runs on a project with a monolithic multi-decision file
+- **THEN** the file is classified as monolithic and its individual decisions are enumerated as
+  migration units
+
+#### Scenario: Unknown shape falls back safely
+- **WHEN** detection encounters ADR-like content matching no known shape
+- **THEN** it is classified `unrecognized` and listed in the report without any conversion
+  attempt
+
+### Requirement: Mechanical/LLM field-mapping boundary
+During conversion, fields that are structurally unambiguous in the source (Status and Date from
+frontmatter or clear headings; title; ID assignment; source provenance) SHALL be filled
+mechanically. Only interpretive fields (Consequences, Alternatives rejected, monolithic-file
+decision splitting) MAY be LLM-filled, and every LLM-filled field SHALL be flagged. The
+migration SHALL never invent a Decision or Context field absent from the source — such fields
+SHALL carry an explicit not-stated-in-source marker and lower the file's confidence.
+
+#### Scenario: Unambiguous field filled mechanically
+- **WHEN** a source ADR has a `Status: Accepted` heading and a date
+- **THEN** the converted file's `status` and `date` are set without any LLM call
+
+#### Scenario: Missing core field is not invented
+- **WHEN** a source decision has no discernible Context
+- **THEN** the converted Context section contains a not-stated-in-source marker and the file is
+  flagged low-confidence
+
+### Requirement: Single uncertainty-flagging mechanism
+Migration uncertainty SHALL be surfaced through exactly one mechanism: a
+`migration_confidence: low|medium|high` frontmatter field per converted file, plus one
+migration report file (`docs/adr/migration-report.md`) summarizing per-file confidence, the
+source-to-new mapping, and the overall low-confidence flag rate. Scattered inline uncertainty
+comments SHALL NOT be used.
+
+#### Scenario: Report aggregates confidence
+- **WHEN** a migration completes
+- **THEN** the report lists every converted file with its confidence, its source location, and
+  the overall flag rate
+
+### Requirement: Non-destructive, dual-existence conversion
+Migration SHALL only write new files under `docs/adr/`; existing ADR content SHALL never be
+deleted, moved, or edited by the migration. Removal of the old system SHALL happen only as an
+explicit, separate, user-approved step offered after the migration report exists.
+
+#### Scenario: Old system untouched
+- **WHEN** a migration run completes
+- **THEN** every pre-existing ADR source file is byte-identical to its pre-migration state
+
+#### Scenario: Deletion requires explicit approval
+- **WHEN** the user has not explicitly approved old-system removal
+- **THEN** no tooling path deletes the old content, and the removal offer appears only after the
+  report is available for review
+
+### Requirement: Pilot gate before broad rollout
+The migration SHALL be piloted on 2–3 real surveyed projects, and the low-confidence flag rate
+SHALL be measured against an explicit threshold recorded at pilot time before the heuristics are
+treated as good enough for wider rollout.
+
+#### Scenario: High flag rate blocks rollout
+- **WHEN** a pilot run's low-confidence rate exceeds the recorded threshold
+- **THEN** the heuristics are tightened and the pilot re-run before migrating further projects
diff --git a/openspec/specs/adr-plugin-packaging/spec.md b/openspec/specs/adr-plugin-packaging/spec.md
new file mode 100644
index 0000000..8c64a6b
--- /dev/null
+++ b/openspec/specs/adr-plugin-packaging/spec.md
@@ -0,0 +1,65 @@
+# Spec: ADR Plugin Packaging
+
+## Purpose
+
+Defines how the `os-adr` plugin is packaged, named, installed, and structured: cc-os naming conventions, git-tracked source with a symlinked install, full three-place marketplace registration, plugin-relative hook wiring, and an object-oriented Ruby core with model-free deterministic tests.
+
+## Requirements
+
+### Requirement: Plugin naming and skill naming conventions
+The plugin SHALL be named `os-adr` (kebab-case, `os-` prefix) and its skills SHALL use
+verb-first kebab-case names invoked as `/os-adr:<verb>` (`create`, `init`, `migrate`, `find`),
+with no `commands/` dispatcher directory — matching the cc-os plugin/skill naming convention.
+
+#### Scenario: Skills register under verb-first names
+- **WHEN** the plugin is installed and `claude plugin details os-adr@local-plugins` is run
+- **THEN** the listed skills are `create`, `init`, `migrate`, and `find`, and no `commands/`
+  directory exists in the plugin
+
+### Requirement: Source of truth in cc-os with symlinked install
+The plugin source SHALL live git-tracked at `cc-os/plugins/os-adr/` and be installed via a
+symlink `~/.claude/plugins/os-adr` pointing at it, matching the `os-vault`/`os-doc-hygiene`/
+`os-orchestration` pattern.
+
+#### Scenario: Symlinked install
+- **WHEN** installation is complete
+- **THEN** `~/.claude/plugins/os-adr` is a symlink resolving to the git-tracked plugin directory
+
+### Requirement: Full three-place marketplace registration
+Installation SHALL update all three registration surfaces — the plugin directory/symlink, the
+`local-plugins` marketplace manifest (`~/.claude/plugins/.claude-plugin/marketplace.json`), and
+`settings.json` `enabledPlugins` — then refresh plugin-manager state via
+`claude plugin marketplace update` and `claude plugin install`, and verify skill registration
+with `claude plugin list` and `claude plugin details`.
+
+#### Scenario: Skills resolve after install
+- **WHEN** the registration procedure completes
+- **THEN** `claude plugin list` shows `os-adr@local-plugins` enabled and its skills are
+  invocable as slash commands in a fresh session
+
+### Requirement: Hooks wired through plugin hooks.json
+Plugin hooks SHALL be declared in `hooks/hooks.json` using `${CLAUDE_PLUGIN_ROOT}`-relative
+commands (the `os-doc-hygiene` pattern), not by absolute paths in `settings.json`, so hook
+registration cannot silently outlive or diverge from plugin registration.
+
+#### Scenario: Hook path is plugin-relative
+- **WHEN** the plugin's hook configuration is inspected
+- **THEN** the SessionStart command references `${CLAUDE_PLUGIN_ROOT}` and no os-adr hook entry
+  exists in `settings.json`
+
+### Requirement: OO core with tested deterministic behavior
+The plugin's core logic SHALL be implemented as an object-oriented Ruby library (small
+single-responsibility classes, dependency-injected paths/IO, no globals) exposed through thin
+CLI entry points, with a thin Python script only for the SessionStart hook. All deterministic
+behaviors (ID assignment, index regeneration, shape detection, existence check, retrieval layers
+1–2) SHALL have automated tests runnable without any LLM, including golden-example fixtures.
+
+#### Scenario: Deterministic tests run model-free
+- **WHEN** the plugin's test suite runs on a machine with no model access
+- **THEN** all tests for ID assignment, index regeneration, detection, existence check, and
+  frontmatter retrieval pass
+
+#### Scenario: Skills delegate to CLI entry points
+- **WHEN** a skill performs a mechanical operation (numbering, indexing, detection, filtering)
+- **THEN** it does so by invoking a `bin/` CLI rather than re-implementing the logic in prompt
+  instructions
diff --git a/openspec/specs/adr-retrieval/spec.md b/openspec/specs/adr-retrieval/spec.md
new file mode 100644
index 0000000..18aac20
--- /dev/null
+++ b/openspec/specs/adr-retrieval/spec.md
@@ -0,0 +1,53 @@
+# Spec: ADR Retrieval
+
+## Purpose
+
+Defines how relevant ADRs are found by the `os-adr` plugin: deterministic-first layered narrowing (frontmatter match, status filter, Graphify expansion) with AI judgment applied only to the narrowed candidate set, graceful degradation without a project graph, a mid-task-callable query surface, and eval-readiness sketches accompanying the build.
+
+## Requirements
+
+### Requirement: Deterministic-first retrieval layering
+ADR retrieval SHALL narrow candidates through ordered layers: (1) path/component match of query
+inputs against each ADR's `affected-paths`/`affected-components` frontmatter; (2) status filter
+admitting only `Accepted` ADRs by default, with an explicit override to include other statuses;
+(3) Graphify graph traversal when a project graph exists; (4) AI judgment applied only to the
+already-narrowed candidate set. AI judgment SHALL never be applied to the full ADR corpus.
+
+#### Scenario: Path match narrows first
+- **WHEN** retrieval is queried with file paths being edited
+- **THEN** ADRs whose `affected-paths` match are selected before any model involvement
+
+#### Scenario: Default status filter
+- **WHEN** retrieval runs without an override
+- **THEN** only `Accepted` ADRs appear in the candidate set
+
+#### Scenario: AI sees only the narrowed set
+- **WHEN** the AI-judgment layer runs
+- **THEN** it receives only the candidates surviving layers 1–3, not the full corpus
+
+### Requirement: Graph layer degrades gracefully
+WHEN no `graphify-out/` project graph exists (or ADRs are not yet indexed as graph nodes),
+retrieval SHALL still function using layers 1, 2, and 4, without error.
+
+#### Scenario: No project graph present
+- **WHEN** retrieval runs in a project that has never been onboarded to Graphify
+- **THEN** results are produced from frontmatter matching and status filtering alone
+
+### Requirement: Mid-task-callable query surface
+Retrieval SHALL be exposed as a `find` skill (`/os-adr:find`) backed by a deterministic CLI,
+callable at any point during a task — not only at SessionStart — so an agent can check
+in-progress work against existing decisions.
+
+#### Scenario: Query during a task
+- **WHEN** an agent mid-task invokes `/os-adr:find` with the files or topic it is working on
+- **THEN** it receives the relevant Accepted ADR candidates for those inputs
+
+### Requirement: Eval-readiness sketches accompany the build
+Held-out evaluation scenario shapes (write-trigger and retrieval) SHALL be sketched in a design
+doc before the plugin skeleton is finalized, and any plugin surface those sketches require SHALL
+be part of the initial build. The evaluation itself SHALL NOT be part of this change.
+
+#### Scenario: Sketches exist before skeleton freeze
+- **WHEN** the plugin skeleton is finalized
+- **THEN** a scenario-sketch document exists covering at least one write-trigger and one
+  retrieval scenario shape
diff --git a/openspec/specs/adr-session-awareness/spec.md b/openspec/specs/adr-session-awareness/spec.md
new file mode 100644
index 0000000..10d6253
--- /dev/null
+++ b/openspec/specs/adr-session-awareness/spec.md
@@ -0,0 +1,49 @@
+# Spec: ADR Session Awareness
+
+## Purpose
+
+Defines the `os-adr` SessionStart behavior: a deterministic, model-free existence check that injects a near-zero-token usage note when a project's ADR system exists, and a suppressible, once-per-day setup suggestion when it doesn't.
+
+## Requirements
+
+### Requirement: Deterministic SessionStart existence check
+The plugin SHALL register a SessionStart hook (matcher `startup|resume`) that deterministically
+checks for `docs/adr/` and its index file in the current project. The hook SHALL make no LLM
+call, SHALL complete within its configured timeout (5s), and SHALL add no perceptible
+session-start latency.
+
+#### Scenario: Hook is deterministic
+- **WHEN** the SessionStart hook runs
+- **THEN** its output is a pure function of the project's filesystem state (ADR dir, index,
+  suppression state) with no model invocation
+
+### Requirement: Minimal context injection when the ADR system exists
+WHEN `docs/adr/` and the index file both exist, the hook SHALL inject a near-zero-token
+additionalContext note stating that the ADR system exists and naming both usage surfaces:
+writing (`/os-adr:create`) and retrieval (`/os-adr:find`).
+
+#### Scenario: Present project gets the usage note
+- **WHEN** a session starts in a project with `docs/adr/` and an index
+- **THEN** the injected context is a short note (a few lines at most) naming `/os-adr:create`
+  and `/os-adr:find`
+
+### Requirement: Setup suggestion with suppression when absent
+WHEN the ADR system is absent, the hook SHALL notify the user and suggest `/os-adr:init` or
+`/os-adr:migrate` — unless a per-project suppression flag is set, in which case it SHALL stay
+silent. The suggestion SHALL also be rate-limited to at most once per day per project via a
+snooze timestamp. Suppression state SHALL live in a gitignored per-project state directory
+(`.os-adr/`), following the `os-doc-hygiene` state-dir shape.
+
+#### Scenario: Absent and unsuppressed
+- **WHEN** a session starts in a project with no `docs/adr/` and no suppression flag, and no
+  reminder was shown today
+- **THEN** the hook emits a one-line suggestion to run `/os-adr:init` or `/os-adr:migrate` and
+  stamps the snooze timestamp
+
+#### Scenario: Suppressed project stays silent
+- **WHEN** a session starts in a project whose `.os-adr/suppress` flag is set
+- **THEN** the hook emits nothing
+
+#### Scenario: Same-day repeat session
+- **WHEN** a second session starts the same day in an absent, unsuppressed project
+- **THEN** no repeat suggestion is emitted
diff --git a/plugins/os-adr/.claude-plugin/plugin.json b/plugins/os-adr/.claude-plugin/plugin.json
new file mode 100644
index 0000000..c684e00
--- /dev/null
+++ b/plugins/os-adr/.claude-plugin/plugin.json
@@ -0,0 +1,5 @@
+{
+  "name": "os-adr",
+  "version": "0.1.0",
+  "description": "Architecture Decision Records, consistently in every project: one template, docs/adr/NNNN-title.md + mechanical index, deterministic SessionStart existence check, non-destructive migration, and deterministic-first retrieval."
+}
diff --git a/plugins/os-adr/bin/adr-detect b/plugins/os-adr/bin/adr-detect
new file mode 100755
index 0000000..ada4274
--- /dev/null
+++ b/plugins/os-adr/bin/adr-detect
@@ -0,0 +1,16 @@
+#!/usr/bin/env ruby
+# Classify existing ADR-like content into surveyed shapes and enumerate
+# migration units. Report-only: writes nothing.
+#
+# Usage: adr-detect [--root DIR]   ->  JSON on stdout
+
+require "json"
+require_relative "../lib/adr"
+
+root = Dir.pwd
+ARGV.each_with_index { |arg, i| root = ARGV[i + 1] if arg == "--root" }
+
+units = Adr::Detector.new(root: root).units.map do |unit|
+  { "shape" => unit.shape.to_s, "source" => unit.source_ref, "title" => unit.title }
+end
+puts JSON.pretty_generate("units" => units)
diff --git a/plugins/os-adr/bin/adr-find b/plugins/os-adr/bin/adr-find
new file mode 100755
index 0000000..20aa8b0
--- /dev/null
+++ b/plugins/os-adr/bin/adr-find
@@ -0,0 +1,30 @@
+#!/usr/bin/env ruby
+# Deterministic ADR retrieval (layers 1-3). AI judgment belongs in the `find`
+# skill, applied only to this command's output.
+#
+# Usage: adr-find [--root DIR] [--paths a,b] [--components x,y] [--all-statuses]
+#   ->  JSON {"candidates": [{id, title, status, date, file, matched_on}]}
+
+require "json"
+require_relative "../lib/adr"
+
+root = Dir.pwd
+paths = []
+components = []
+all_statuses = false
+ARGV.each_with_index do |arg, i|
+  case arg
+  when "--root"          then root = ARGV[i + 1]
+  when "--paths"         then paths = ARGV[i + 1].to_s.split(",")
+  when "--components"    then components = ARGV[i + 1].to_s.split(",")
+  when "--all-statuses"  then all_statuses = true
+  end
+end
+
+repo = Adr::Repository.new(root: root)
+finder = Adr::Finder.new(
+  repository: repo,
+  graph_path: File.join(root, "graphify-out", "graph.json")
+)
+candidates = finder.find(paths: paths, components: components, all_statuses: all_statuses)
+puts JSON.pretty_generate("candidates" => candidates)
diff --git a/plugins/os-adr/bin/adr-init b/plugins/os-adr/bin/adr-init
new file mode 100755
index 0000000..305407a
--- /dev/null
+++ b/plugins/os-adr/bin/adr-init
@@ -0,0 +1,20 @@
+#!/usr/bin/env ruby
+# Initialize docs/adr/ + empty generated index in a project, and clear any
+# session-hook suppression flag (.os-adr/suppress).
+#
+# Usage: adr-init [--root DIR]
+
+require "fileutils"
+require_relative "../lib/adr"
+
+root = Dir.pwd
+ARGV.each_with_index { |arg, i| root = ARGV[i + 1] if arg == "--root" }
+
+repo = Adr::Repository.new(root: root)
+FileUtils.mkdir_p(repo.dir)
+Adr::Index.new(repository: repo).regenerate!
+
+suppress = File.join(root, ".os-adr", "suppress")
+File.delete(suppress) if File.exist?(suppress)
+
+puts repo.dir
diff --git a/plugins/os-adr/bin/adr-migrate b/plugins/os-adr/bin/adr-migrate
new file mode 100755
index 0000000..8d870b8
--- /dev/null
+++ b/plugins/os-adr/bin/adr-migrate
@@ -0,0 +1,35 @@
+#!/usr/bin/env ruby
+# Non-destructive mechanical migration pass.
+#
+#   adr-migrate [--root DIR]                  convert detected units into new
+#                                             ADRs under docs/adr/; print the
+#                                             work manifest (JSON) of fields
+#                                             needing LLM fill
+#   adr-migrate [--root DIR] --apply-fills F  apply LLM fills from JSON file F
+#                                             ({relpath: {field: text}}); only
+#                                             pending-marker fields are fillable
+#
+# Never deletes, moves, or edits source files; writes only under docs/adr/.
+
+require "json"
+require "fileutils"
+require_relative "../lib/adr"
+
+root = Dir.pwd
+fills_path = nil
+ARGV.each_with_index do |arg, i|
+  root = ARGV[i + 1] if arg == "--root"
+  fills_path = ARGV[i + 1] if arg == "--apply-fills"
+end
+
+repo = Adr::Repository.new(root: root)
+FileUtils.mkdir_p(repo.dir)
+migrator = Adr::Migrator.new(repository: repo, detector: Adr::Detector.new(root: root))
+
+if fills_path
+  migrator.apply_fills!(JSON.parse(File.read(fills_path)))
+  puts JSON.pretty_generate("applied" => true,
+                            "report" => File.join("docs", "adr", Adr::MigrationReport::FILENAME))
+else
+  puts JSON.pretty_generate(migrator.migrate!)
+end
diff --git a/plugins/os-adr/bin/adr-new b/plugins/os-adr/bin/adr-new
new file mode 100755
index 0000000..cd7b831
--- /dev/null
+++ b/plugins/os-adr/bin/adr-new
@@ -0,0 +1,58 @@
+#!/usr/bin/env ruby
+# Create a templated, numbered, indexed ADR in one invocation.
+#
+# Usage: echo '{"title": "...", ...}' | adr-new [--root DIR]
+#
+# JSON input fields:
+#   title       (required)  decision title
+#   status      (optional)  default "Accepted"
+#   date        (optional)  ISO date, default today
+#   supersedes  (optional)  4-digit ID of the ADR this one supersedes
+#   affected_paths, affected_components  (optional)  arrays
+#   context, decision, consequences, alternatives  (optional)  section bodies
+#
+# Prints the created file path. Also mechanically sets superseded-by + status
+# on the superseded ADR and regenerates the index.
+
+require "json"
+require "date"
+require_relative "../lib/adr"
+
+root = Dir.pwd
+ARGV.each_with_index { |arg, i| root = ARGV[i + 1] if arg == "--root" }
+
+input = JSON.parse($stdin.read)
+title = input["title"].to_s.strip
+abort "adr-new: 'title' is required" if title.empty?
+
+repo = Adr::Repository.new(root: root)
+abort "adr-new: #{repo.dir} is not initialized (run adr-init first)" unless Dir.exist?(repo.dir)
+
+id = repo.next_id
+template = Adr::Template.new(path: File.expand_path("../templates/adr.md", __dir__))
+content = template.render(
+  id: id,
+  title: title,
+  date: input["date"] || Date.today.iso8601,
+  status: input["status"] || "Accepted",
+  supersedes: input["supersedes"] ? %("#{input['supersedes']}") : nil,
+  affected_paths: input["affected_paths"] || [],
+  affected_components: input["affected_components"] || [],
+  context: input["context"],
+  decision: input["decision"],
+  consequences: input["consequences"],
+  alternatives: input["alternatives"]
+)
+
+path = File.join(repo.dir, "#{id}-#{Adr.slugify(title)}.md")
+File.write(path, content)
+
+if (old_id = input["supersedes"])
+  entry = repo.find(old_id.to_s)
+  abort "adr-new: superseded ADR #{old_id} not found" unless entry
+  entry.record.supersede_with(id)
+  repo.write_entry(entry)
+end
+
+Adr::Index.new(repository: repo).regenerate!
+puts path
diff --git a/plugins/os-adr/hooks/__pycache__/session_start.cpython-314.pyc b/plugins/os-adr/hooks/__pycache__/session_start.cpython-314.pyc
new file mode 100644
index 0000000..677cb7c
Binary files /dev/null and b/plugins/os-adr/hooks/__pycache__/session_start.cpython-314.pyc differ
diff --git a/plugins/os-adr/hooks/hooks.json b/plugins/os-adr/hooks/hooks.json
new file mode 100644
index 0000000..f142c39
--- /dev/null
+++ b/plugins/os-adr/hooks/hooks.json
@@ -0,0 +1,16 @@
+{
+  "hooks": {
+    "SessionStart": [
+      {
+        "matcher": "startup|resume",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "python3 ${CLAUDE_PLUGIN_ROOT}/hooks/session_start.py",
+            "timeout": 5
+          }
+        ]
+      }
+    ]
+  }
+}
diff --git a/plugins/os-adr/hooks/session_start.py b/plugins/os-adr/hooks/session_start.py
new file mode 100644
index 0000000..7fc81fa
--- /dev/null
+++ b/plugins/os-adr/hooks/session_start.py
@@ -0,0 +1,121 @@
+#!/usr/bin/env python3
+"""session_start.py — deterministic SessionStart existence check for os-adr.
+
+Pure function of the project's filesystem state; no model, no network, no Ruby.
+
+  present (docs/adr/ + index)  -> additionalContext note naming /os-adr:create
+                                  and /os-adr:find (near-zero tokens)
+  absent, suppressed           -> silent (.os-adr/suppress flag)
+  absent, reminded today       -> silent (.os-adr/last_reminded snooze stamp)
+  absent otherwise             -> one-line systemMessage suggesting
+                                  /os-adr:init or /os-adr:migrate; stamps snooze
+  not a git project            -> silent (never creates state dirs outside
+                                  real projects)
+
+Always exits 0 — the hook must never block a session.
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+from datetime import date
+from pathlib import Path
+from typing import Optional, Tuple
+
+PRESENT_NOTE = (
+    "[os-adr] This project records architecture decisions in docs/adr/ "
+    "(index: docs/adr/README.md). Record a new decision of consequence with "
+    "/os-adr:create; before changing decided-on behavior, check relevant "
+    "decisions with /os-adr:find."
+)
+
+ABSENT_NOTE = (
+    "[os-adr] No ADR system in this project. /os-adr:init sets one up; "
+    "/os-adr:migrate converts existing decision docs. "
+    "(touch .os-adr/suppress to silence this permanently)"
+)
+
+
+def find_project_root(cwd: Path) -> Optional[Path]:
+    """Nearest ancestor (including cwd) containing .git, else None."""
+    for candidate in [cwd, *cwd.parents]:
+        if (candidate / ".git").exists():
+            return candidate
+    return None
+
+
+def adr_system_present(root: Path) -> bool:
+    """The same existence rule as Ruby Adr::Repository#exists?."""
+    adr_dir = root / "docs" / "adr"
+    return adr_dir.is_dir() and (adr_dir / "README.md").is_file()
+
+
+def decide(
+    present: bool,
+    suppressed: bool,
+    last_reminded: Optional[date],
+    today: date,
+) -> Tuple[str, Optional[str]]:
+    """Pure decision: ('context', note) | ('banner', note) | ('silent', None)."""
+    if present:
+        return ("context", PRESENT_NOTE)
+    if suppressed:
+        return ("silent", None)
+    if last_reminded == today:
+        return ("silent", None)
+    return ("banner", ABSENT_NOTE)
+
+
+class StateDir:
+    """The gitignored per-project .os-adr/ state directory."""
+
+    def __init__(self, root: Path) -> None:
+        self._dir = root / ".os-adr"
+
+    def suppressed(self) -> bool:
+        return (self._dir / "suppress").exists()
+
+    def last_reminded(self) -> Optional[date]:
+        try:
+            return date.fromisoformat(
+                (self._dir / "last_reminded").read_text().strip()
+            )
+        except Exception:
+            return None
+
+    def stamp_reminded(self, today: date) -> None:
+        self._dir.mkdir(exist_ok=True)
+        (self._dir / "last_reminded").write_text(today.isoformat())
+
+
+def run(root: Path, today: date, out) -> str:
+    state = StateDir(root)
+    outcome, note = decide(
+        adr_system_present(root), state.suppressed(), state.last_reminded(), today
+    )
+    if outcome == "context":
+        out.write(json.dumps({
+            "hookSpecificOutput": {
+                "hookEventName": "SessionStart",
+                "additionalContext": note,
+            }
+        }))
+    elif outcome == "banner":
+        out.write(json.dumps({"systemMessage": note}))
+        state.stamp_reminded(today)
+    return outcome
+
+
+def main() -> int:
+    try:
+        root = find_project_root(Path.cwd())
+        if root is not None:
+            run(root, date.today(), sys.stdout)
+    except Exception:
+        pass  # never block a session
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/plugins/os-adr/invariants.md b/plugins/os-adr/invariants.md
new file mode 100644
index 0000000..9fc0001
--- /dev/null
+++ b/plugins/os-adr/invariants.md
@@ -0,0 +1,53 @@
+# os-adr invariants
+
+Behavioral invariants of the plugin. Changing any of these requires explicit human approval and
+a matching update to the tests named beside it. Suite entry points: `ruby tests/all.rb` and
+`python3 tests/hook_test.py` (both model-free).
+
+1. **Non-destructiveness (migration).** Migration writes only new files under `docs/adr/`;
+   existing source content is never deleted, moved, or edited by any tooling path. Old-system
+   removal is a separate, explicit, user-approved step (skill-mediated, never automatic).
+   — `migrator_test.rb#test_sources_stay_byte_identical`, pilot sha256 checks.
+
+2. **Index regeneration, never accretion.** `docs/adr/README.md`'s table is regenerated in full
+   from directory contents on every ADR write; it is never appended to or hand-edited, and
+   self-heals after manual edits. Content outside the markers is preserved.
+   — `index_test.rb` (golden fixture, self-heal, prose preservation, generated-header cases).
+
+3. **IDs come from the directory, never the index.** Next ID = max existing filename ID + 1,
+   derived by scanning `docs/adr/`; a stale or missing index never influences numbering.
+   — `repository_test.rb#test_next_id_ignores_{missing,stale}_index`.
+
+4. **No LLM in the hook.** The SessionStart hook is a pure function of filesystem state (ADR
+   dir, index, suppression state, snooze stamp) — no model call, no network, no Ruby spin-up;
+   always exits 0; emits nothing (and creates no state) outside a git project. Its existence
+   rule is identical to `Adr::Repository#exists?`.
+   — `hook_test.py` (all cases; `PythonRubyAgreementTest` pins the shared rule).
+
+5. **Single uncertainty-flagging mechanism.** Migration uncertainty surfaces only via
+   `migration_confidence: low|medium|high` frontmatter plus `docs/adr/migration-report.md`.
+   No scattered inline uncertainty comments.
+   — `migrator_test.rb#test_report_lists_files_confidence_sources_and_flag_rate`.
+
+6. **Core fields are never invented.** A Decision or Context absent from a migration source is
+   written as `_not stated in source_` and flags the file low-confidence; LLM fills apply only
+   to pending-marker fields (enforced structurally by `apply_fills!` — a stray fill key is a
+   no-op), and an LLM-filled file stays flagged (≤ medium).
+   — `migrator_test.rb#test_missing_core_fields_*`, `#test_apply_fills_only_touches_pending_fields`.
+
+7. **Unrecognized content is reported, never auto-converted.** Content matching no known shape
+   — including multi-decision files without mechanically splittable unit headings — is listed
+   in the report for manual handling only.
+   — `detector_test.rb#test_multi_decision_file_without_unit_headings_is_unrecognized`,
+   `migrator_test.rb#test_unrecognized_content_is_reported_never_converted`.
+
+8. **Deterministic-first retrieval.** `adr-find` narrows by frontmatter path/component match,
+   then Accepted-only status filter (explicit override), then optional one-hop graph expansion
+   that degrades gracefully (absent or corrupt graph → layers 1–2). AI judgment happens only in
+   the `find` skill, only over the CLI's returned candidates.
+   — `finder_test.rb` (all cases, including both degradation paths).
+
+9. **Skills delegate mechanics to `bin/` CLIs.** Numbering, templating, indexing, supersession
+   bookkeeping, detection, conversion, and retrieval filtering are never re-implemented in
+   prompt instructions.
+   — by construction of `skills/*/SKILL.md`; `cli_test.rb` pins the CLI contracts.
diff --git a/plugins/os-adr/lib/adr.rb b/plugins/os-adr/lib/adr.rb
new file mode 100644
index 0000000..683f212
--- /dev/null
+++ b/plugins/os-adr/lib/adr.rb
@@ -0,0 +1,14 @@
+require_relative "adr/record"
+require_relative "adr/repository"
+require_relative "adr/index"
+require_relative "adr/template"
+require_relative "adr/detector"
+require_relative "adr/migrator"
+require_relative "adr/migration_report"
+require_relative "adr/finder"
+
+module Adr
+  def self.slugify(title)
+    title.downcase.gsub(/[^a-z0-9]+/, "-").gsub(/\A-|-\z/, "")
+  end
+end
diff --git a/plugins/os-adr/lib/adr/detector.rb b/plugins/os-adr/lib/adr/detector.rb
new file mode 100644
index 0000000..5207fbf
--- /dev/null
+++ b/plugins/os-adr/lib/adr/detector.rb
@@ -0,0 +1,108 @@
+module Adr
+  # Mechanical classification of existing ADR-like content into the surveyed
+  # shapes, enumerating migration units. Report-only for unrecognized content —
+  # nothing here writes anything.
+  class Detector
+    Unit = Struct.new(:shape, :source_path, :anchor, :title, :body, keyword_init: true) do
+      def source_ref
+        anchor ? "#{source_path}##{anchor}" : source_path
+      end
+    end
+
+    CANDIDATE_DIRS = [%w[docs adr], %w[docs decisions], %w[docs decision]].freeze
+    GENERATED_BASENAMES = %w[README.md migration-report.md].freeze
+    UNIT_HEADING = /^\#{1,4}\s*(?:ADR[-\s]?\d+|D-\d+)\b/i
+    PROSE_DECISION = /^(?:\#{1,4}\s*|\*\*)?Decision\s+\d+\s*[:—–-]/i
+    DECISION_MARKERS = /^(?:\#{1,4}\s*|\*\*)?(?:Status|Decision|Decided|Context)\b/i
+
+    def initialize(root:)
+      @root = root
+    end
+
+    def units
+      candidate_files.flat_map { |path| classify_file(path) }
+    end
+
+    def candidate_files
+      files = CANDIDATE_DIRS.flat_map do |parts|
+        Dir.glob(File.join(@root, *parts, "*.md"))
+      end
+      files += Dir.glob(File.join(@root, "DECISIONS.md"))
+      files.uniq.sort.reject { |p| generated?(p) || already_migrated?(p) }
+    end
+
+    private
+
+    def generated?(path)
+      GENERATED_BASENAMES.include?(File.basename(path))
+    end
+
+    # A file already in the os-adr format (frontmatter with an id) is not a
+    # migration source.
+    def already_migrated?(path)
+      text = File.read(path)
+      return false unless text.start_with?("---\n")
+      !Record.parse(text).id.nil?
+    rescue Record::ParseError, StandardError
+      false
+    end
+
+    def classify_file(path)
+      text = File.read(path)
+      base = File.basename(path)
+      relative = relative(path)
+
+      if text.scan(UNIT_HEADING).size >= 2
+        split_units(relative, text, :monolithic, UNIT_HEADING)
+      elsif text.scan(PROSE_DECISION).size >= 2
+        split_units(relative, text, :prose_embedded, PROSE_DECISION)
+      elsif text.scan(/^\**Status\**\s*[:：]/i).size >= 2
+        # Multiple decisions but no mechanically splittable unit headings:
+        # report-only, never auto-converted.
+        [Unit.new(shape: :unrecognized, source_path: relative, anchor: nil,
+                  title: base, body: text)]
+      elsif text.match?(DECISION_MARKERS)
+        shape =
+          if base.match?(/\A\d{4}-\d{2}(-\d{2})?-/) then :dated_single_file
+          elsif base.match?(/\A\d+-/)               then :numbered_per_file
+          else                                           :topic_single_file
+          end
+        [Unit.new(shape: shape, source_path: relative, anchor: nil,
+                  title: single_title(text, base), body: text)]
+      else
+        [Unit.new(shape: :unrecognized, source_path: relative, anchor: nil,
+                  title: base, body: text)]
+      end
+    end
+
+    def split_units(relative, text, shape, pattern)
+      lines = text.lines
+      starts = lines.each_index.select { |i| lines[i].match?(pattern) }
+      starts.each_with_index.map do |start, n|
+        stop = starts[n + 1] || lines.size
+        heading = lines[start].strip
+        Unit.new(shape: shape, source_path: relative, anchor: heading,
+                 title: unit_title(heading), body: lines[start...stop].join)
+      end
+    end
+
+    def unit_title(heading)
+      heading
+        .sub(/\A\#{1,4}\s*/, "")
+        .gsub("**", "")
+        .sub(/\A(?:ADR[-\s]?\d+|D-\d+|Decision\s+\d+)\s*[:—–-]?\s*/i, "")
+        .strip
+    end
+
+    def single_title(text, base)
+      h1 = text[/^# +(.+)$/, 1]
+      return h1.strip.sub(/\A(?:ADR[-\s]?\d+|D-\d+|\d{1,4})\s*[:—–-]\s*/i, "") if h1
+
+      base.sub(/\.md\z/, "").sub(/\A[\d-]+/, "").tr("-", " ").strip.capitalize
+    end
+
+    def relative(path)
+      path.sub(%r{\A#{Regexp.escape(@root)}/?}, "")
+    end
+  end
+end
diff --git a/plugins/os-adr/lib/adr/finder.rb b/plugins/os-adr/lib/adr/finder.rb
new file mode 100644
index 0000000..92b2f1f
--- /dev/null
+++ b/plugins/os-adr/lib/adr/finder.rb
@@ -0,0 +1,83 @@
+require "json"
+require "set"
+
+module Adr
+  # Deterministic-first ADR retrieval. Layers, in order:
+  #   1. path/component match against affected-paths / affected-components
+  #   2. status filter (Accepted only unless all_statuses)
+  #   3. Graphify one-hop expansion of the query paths (when graphify-out/
+  #      exists; degrades gracefully to layers 1-2 when absent or unreadable)
+  # AI judgment (layer 4) happens in the `find` skill, only over what this
+  # returns — never the full corpus.
+  class Finder
+    def initialize(repository:, graph_path: nil)
+      @repository = repository
+      @graph_path = graph_path
+    end
+
+    def find(paths: [], components: [], all_statuses: false)
+      paths = normalize(paths)
+      graph_paths = normalize(graph_neighbors(paths)) - paths
+      @repository.entries.filter_map do |entry|
+        record = entry.record
+        next unless all_statuses || record.status == "Accepted"
+
+        matched = match_reasons(record, paths, graph_paths, components)
+        next if matched.empty?
+
+        { "id" => record.id, "title" => record.title, "status" => record.status,
+          "date" => record.date.to_s, "file" => entry.path, "matched_on" => matched }
+      end
+    end
+
+    private
+
+    def match_reasons(record, paths, graph_paths, components)
+      affected_paths = Array(record.frontmatter["affected-paths"]).map { |p| p.to_s.chomp("/") }
+      affected_components = Array(record.frontmatter["affected-components"]).map(&:to_s)
+      reasons = []
+      reasons << "path" if overlap?(affected_paths, paths)
+      reasons << "graph" if reasons.empty? && overlap?(affected_paths, graph_paths)
+      reasons << "component" if (affected_components & components).any?
+      reasons
+    end
+
+    def overlap?(affected, queried)
+      affected.any? { |a| queried.any? { |q| path_match?(a, q) } }
+    end
+
+    # Prefix containment either way: an ADR affecting `lib/storage` matches a
+    # query for `lib/storage/adapter.rb`, and one affecting a specific file
+    # matches a query for its directory.
+    def path_match?(a, q)
+      a == q || q.start_with?("#{a}/") || a.start_with?("#{q}/")
+    end
+
+    def normalize(paths)
+      paths.map { |p| p.to_s.chomp("/") }.reject(&:empty?).uniq
+    end
+
+    # Files one graph hop from the queried files, via graphify-out/graph.json
+    # node source_file attributes. Any failure degrades to no expansion.
+    def graph_neighbors(paths)
+      return [] unless @graph_path && File.exist?(@graph_path)
+
+      graph = JSON.parse(File.read(@graph_path))
+      nodes = graph.fetch("nodes", [])
+      id_to_file = nodes.to_h { |n| [n["id"], n["source_file"]] }
+      seeds = nodes.select do |n|
+        n["source_file"] && paths.any? { |q| path_match?(n["source_file"].chomp("/"), q) }
+      end.map { |n| n["id"] }.to_set
+      return [] if seeds.empty?
+
+      graph.fetch("links", []).flat_map do |link|
+        if seeds.include?(link["source"]) then [link["target"]]
+        elsif seeds.include?(link["target"]) then [link["source"]]
+        else []
+        end
+      end.filter_map { |id| id_to_file[id] }.uniq
+    rescue StandardError
+      []
+    end
+  end
+end
diff --git a/plugins/os-adr/lib/adr/index.rb b/plugins/os-adr/lib/adr/index.rb
new file mode 100644
index 0000000..f1fa316
--- /dev/null
+++ b/plugins/os-adr/lib/adr/index.rb
@@ -0,0 +1,61 @@
+module Adr
+  # The generated docs/adr/README.md. Always regenerated in full from the
+  # directory contents — never appended to, never hand-edited. Content outside
+  # the markers is preserved so projects can add prose around the table.
+  class Index
+    BEGIN_MARKER = "<!-- adr-index:begin -->"
+    END_MARKER = "<!-- adr-index:end -->"
+
+    DEFAULT_SKELETON = <<~MD
+      <!-- Generated by os-adr. Do not hand-edit the table: it is regenerated in full on every ADR write. -->
+
+      # Architecture Decision Records
+
+      One file per decision, `NNNN-kebab-title.md`, created via `/os-adr:create`.
+
+      #{BEGIN_MARKER}
+      #{END_MARKER}
+    MD
+
+    def initialize(repository:)
+      @repository = repository
+    end
+
+    def regenerate!
+      File.write(@repository.index_path, content)
+    end
+
+    def content
+      base = if File.exist?(@repository.index_path) && File.read(@repository.index_path).include?(BEGIN_MARKER)
+               File.read(@repository.index_path)
+             else
+               DEFAULT_SKELETON
+             end
+      before, _stale, after = split(base)
+      "#{before}#{BEGIN_MARKER}\n#{table}#{END_MARKER}#{after}"
+    end
+
+    private
+
+    def split(text)
+      before, rest = text.split(BEGIN_MARKER, 2)
+      _stale, after = rest.split(END_MARKER, 2)
+      [before, _stale, after]
+    end
+
+    def table
+      rows = @repository.entries.map do |entry|
+        record = entry.record
+        "| #{record.id} | [#{record.title}](#{File.basename(entry.path)}) " \
+          "| #{record.status} | #{record.date} |"
+      end
+      return "" if rows.empty?
+
+      <<~MD
+        | ID | Title | Status | Date |
+        | --- | --- | --- | --- |
+        #{rows.join("\n")}
+      MD
+    end
+  end
+end
diff --git a/plugins/os-adr/lib/adr/migration_report.rb b/plugins/os-adr/lib/adr/migration_report.rb
new file mode 100644
index 0000000..5e24abc
--- /dev/null
+++ b/plugins/os-adr/lib/adr/migration_report.rb
@@ -0,0 +1,67 @@
+module Adr
+  # Generates docs/adr/migration-report.md — the single aggregate surface for
+  # migration uncertainty (alongside per-file migration_confidence frontmatter).
+  class MigrationReport
+    FILENAME = "migration-report.md"
+
+    # Pilot-gate threshold (task 5.5): a run whose low-confidence flag rate
+    # exceeds this is not good enough for wider rollout — tighten heuristics
+    # and re-run. Set empirically during the 2026-07-03 pilot.
+    FLAG_RATE_THRESHOLD = 0.25
+
+    def initialize(repository:, detector:)
+      @repository = repository
+      @detector = detector
+    end
+
+    def path = File.join(@repository.dir, FILENAME)
+
+    def write!
+      File.write(path, content)
+    end
+
+    def migrated_entries
+      @repository.entries.select { |e| e.record.frontmatter["migration_source"] }
+    end
+
+    def flag_rate
+      entries = migrated_entries
+      return 0.0 if entries.empty?
+
+      low = entries.count { |e| e.record.frontmatter["migration_confidence"] == "low" }
+      (low.to_f / entries.size).round(3)
+    end
+
+    def content
+      entries = migrated_entries
+      unrecognized = @detector.units.select { |u| u.shape == :unrecognized }
+      rows = entries.map do |entry|
+        record = entry.record
+        pending = record.sections.select { |_, v| v == Migrator::PENDING_MARKER }.keys
+        not_stated = record.sections.select { |_, v| v == Migrator::NOT_STATED_MARKER }.keys
+        "| #{record.id} | [#{record.title}](#{File.basename(entry.path)}) " \
+          "| #{record.frontmatter['migration_source']} " \
+          "| #{record.frontmatter['migration_confidence']} " \
+          "| #{(pending.map { |s| "#{s} (pending)" } + not_stated.map { |s| "#{s} (not stated)" }).join(', ')} |"
+      end
+
+      <<~MD
+        <!-- Generated by os-adr migration. Regenerated on every migration pass. -->
+
+        # ADR Migration Report
+
+        Converted: #{entries.size} · Low-confidence flag rate: #{(flag_rate * 100).round(1)}% (gate threshold: #{(FLAG_RATE_THRESHOLD * 100).round}% — #{flag_rate > FLAG_RATE_THRESHOLD ? 'EXCEEDED, tighten heuristics before wider rollout' : 'within gate'})
+
+        | ID | Title | Source | Confidence | Open items |
+        | --- | --- | --- | --- | --- |
+        #{rows.join("\n")}
+
+        ## Unrecognized content (manual handling required)
+
+        #{unrecognized.empty? ? '_none_' : unrecognized.map { |u| "- #{u.source_ref}" }.join("\n")}
+
+        _Sources are untouched. Old-system removal is a separate, explicitly user-approved step._
+      MD
+    end
+  end
+end
diff --git a/plugins/os-adr/lib/adr/migrator.rb b/plugins/os-adr/lib/adr/migrator.rb
new file mode 100644
index 0000000..0c51399
--- /dev/null
+++ b/plugins/os-adr/lib/adr/migrator.rb
@@ -0,0 +1,226 @@
+require "fileutils"
+
+module Adr
+  # Mechanical migration pass. Converts detected units into new-format ADRs
+  # under docs/adr/ — heuristic-filling only structurally unambiguous fields —
+  # and emits a manifest of the interpretive fields an LLM must fill. Writes
+  # nothing outside docs/adr/; never touches source files.
+  class Migrator
+    PENDING_MARKER = "_pending LLM fill (see migration manifest)_"
+    NOT_STATED_MARKER = "_not stated in source_"
+
+    # Mechanical status normalization: canonical lifecycle values plus the one
+    # unambiguous client-facing synonym. Anything else is copied raw.
+    STATUS_MAP = {
+      "accepted" => "Accepted", "proposed" => "Proposed",
+      "superseded" => "Superseded", "deprecated" => "Deprecated",
+      "confirmed" => "Accepted", "settled" => "Accepted", "open" => "Proposed"
+    }.freeze
+
+    # Source heading -> target section. Interpretive targets (consequences,
+    # alternatives) get PENDING when absent; core targets (context, decision)
+    # get NOT_STATED — they are never invented, by an LLM or otherwise.
+    SECTION_SOURCES = {
+      "Context" => [/\Acontext\z/i],
+      "Decision" => [/\Adecision\z/i],
+      "Consequences" => [/\Aconsequences\z/i],
+      "Alternatives rejected" => [/\Aalternatives( (rejected|considered))?\z/i]
+    }.freeze
+
+    # Heading synonyms mapped as a heuristic: the content is copied verbatim
+    # from the source, but because the mapping is a judgment call the file's
+    # confidence is capped at medium.
+    SECTION_FALLBACKS = {
+      "Context" => [/\A(why|rationale|problem|background)\z/i],
+      "Consequences" => [/\A(impact|implications|trade-?offs)\z/i]
+    }.freeze
+    CORE_SECTIONS = ["Context", "Decision"].freeze
+    FILL_FIELDS = { "consequences" => "Consequences",
+                    "alternatives" => "Alternatives rejected" }.freeze
+
+    def initialize(repository:, detector:)
+      @repository = repository
+      @detector = detector
+    end
+
+    def migrate!
+      FileUtils.mkdir_p(@repository.dir)
+      recognized, unrecognized = @detector.units.partition { |u| u.shape != :unrecognized }
+      fresh = recognized.reject { |u| migrated_sources.include?(u.source_ref) }
+      results = fresh.map { |unit| convert(unit) }
+      finalize!
+      manifest(results, unrecognized)
+    end
+
+    # fills: {"docs/adr/NNNN-x.md" => {"consequences" => "...", "status" => "..."}}
+    # Only pending-marker sections and an empty status are fillable — the
+    # manifest is enforced structurally, not trusted from the caller.
+    def apply_fills!(fills)
+      fills.each do |relpath, fields|
+        path = File.join(@repository.root, relpath)
+        record = Record.parse(File.read(path))
+        fields.each { |field, value| apply_fill(record, field, value) }
+        File.write(path, record.to_markdown)
+      end
+      finalize!
+    end
+
+    private
+
+    def migrated_sources
+      @repository.entries.filter_map { |e| e.record.frontmatter["migration_source"] }
+    end
+
+    def convert(unit)
+      extraction = Extraction.new(unit)
+      sections, llm_fields, fallback_used = build_sections(extraction)
+      confidence = confidence_for(sections, llm_fields, extraction, fallback_used)
+      llm_fields << "status" if extraction.status.to_s.empty?
+
+      record = Record.new(
+        frontmatter: {
+          "id" => @repository.next_id,
+          "date" => extraction.date,
+          "status" => extraction.status,
+          "supersedes" => nil, "superseded-by" => nil,
+          "affected-paths" => [], "affected-components" => [],
+          "migration_confidence" => confidence,
+          "migration_source" => unit.source_ref
+        },
+        title: unit.title,
+        sections: sections
+      )
+      path = @repository.write(record)
+      { "file" => relative(path), "source" => unit.source_ref, "shape" => unit.shape.to_s,
+        "confidence" => confidence, "llm_fields" => llm_fields,
+        "source_material" => llm_fields.empty? ? nil : unit.body }.compact
+    end
+
+    def build_sections(extraction)
+      llm_fields = []
+      sections = {}
+      fallback_used = false
+      Record::SECTION_ORDER.each do |name|
+        found = extraction.section(name)
+        if found.nil? && (fallback = extraction.fallback_section(name))
+          found = fallback
+          fallback_used = true
+        end
+        sections[name] =
+          if found
+            found
+          elsif CORE_SECTIONS.include?(name)
+            NOT_STATED_MARKER
+          else
+            llm_fields << FILL_FIELDS.key(name)
+            PENDING_MARKER
+          end
+      end
+      [sections, llm_fields, fallback_used]
+    end
+
+    def confidence_for(sections, llm_fields, extraction, fallback_used)
+      return "low" if CORE_SECTIONS.any? { |name| sections[name] == NOT_STATED_MARKER }
+      return "medium" if llm_fields.any? || fallback_used || extraction.status.to_s.empty?
+
+      "high"
+    end
+
+    def apply_fill(record, field, value)
+      if field == "status"
+        record.frontmatter["status"] = value if record.status.to_s.empty?
+      elsif (section = FILL_FIELDS[field])
+        record.sections[section] = value if record.sections[section] == PENDING_MARKER
+      end
+    end
+
+    def finalize!
+      Index.new(repository: @repository).regenerate!
+      MigrationReport.new(repository: @repository, detector: @detector).write!
+    end
+
+    def manifest(results, unrecognized)
+      {
+        "migrated" => results,
+        "unrecognized" => unrecognized.map { |u| { "source" => u.source_ref, "title" => u.title } },
+        "flag_rate" => MigrationReport.new(repository: @repository, detector: @detector).flag_rate
+      }
+    end
+
+    def relative(path)
+      path.sub(%r{\A#{Regexp.escape(@repository.root)}/?}, "")
+    end
+
+    # Mechanical field extraction from one unit's text: only structurally
+    # unambiguous values (labelled lines, exact headings, ISO dates).
+    class Extraction
+      def initialize(unit)
+        @unit = unit
+        @body = unit.body
+      end
+
+      def status
+        raw = labelled_value("Status") || heading_section("Status")&.lines&.first&.strip
+        return nil if raw.nil? || raw.empty?
+
+        STATUS_MAP.fetch(raw.split(/[\s(]/).first.downcase, raw)
+      end
+
+      def date
+        raw = labelled_value("Date") || labelled_value("Decided")
+        iso = raw&.[](/\d{4}-\d{2}-\d{2}/) ||
+              @unit.source_path[/\d{4}-\d{2}-\d{2}/]
+        iso && Date.parse(iso)
+      end
+
+      def section(target)
+        match_section(SECTION_SOURCES.fetch(target))
+      end
+
+      def fallback_section(target)
+        match_section(SECTION_FALLBACKS.fetch(target, []))
+      end
+
+      private
+
+      def match_section(patterns)
+        patterns.each do |pattern|
+          headings.each do |name, content|
+            return content if name.match?(pattern) && !content.empty?
+          end
+        end
+        nil
+      end
+
+      def labelled_value(label)
+        @body[/^[-*]?\s*\**#{label}\**\s*[:：]\s*\**([^\n*]+)/i, 1]&.strip
+      end
+
+      def heading_section(name)
+        headings.find { |heading, _| heading.match?(/\A#{name}\z/i) }&.last
+      end
+
+      def headings
+        @headings ||= heading_sections + bold_label_sections
+      end
+
+      def heading_sections
+        @body.scan(/^\#{2,4} +([^\n]+)\n(.*?)(?=^\#{2,4} |\z)/m)
+             .map { |name, content| [name.strip, strip_trailing_rule(content)] }
+      end
+
+      # "**Context.** text..." paragraphs (bold-label style, e.g. llf-schema):
+      # the label acts as a section heading; content runs to the next bold
+      # label, heading, or end of unit.
+      def bold_label_sections
+        @body.scan(/^\*\*([A-Za-z][^*\n]{0,40}?)[.:]?\*\*[.:]?\s*(.*?)(?=^\*\*[A-Za-z]|^\#{1,4} |\z)/m)
+             .map { |name, content| [name.strip, strip_trailing_rule(content)] }
+             .reject { |_, content| content.empty? }
+      end
+
+      def strip_trailing_rule(content)
+        content.strip.sub(/\n+---+\s*\z/, "").strip
+      end
+    end
+  end
+end
diff --git a/plugins/os-adr/lib/adr/record.rb b/plugins/os-adr/lib/adr/record.rb
new file mode 100644
index 0000000..2ae1f64
--- /dev/null
+++ b/plugins/os-adr/lib/adr/record.rb
@@ -0,0 +1,82 @@
+require "yaml"
+require "date"
+
+module Adr
+  # One ADR file: frontmatter + titled body sections. Parses and serializes;
+  # knows nothing about the filesystem.
+  class Record
+    ParseError = Class.new(StandardError)
+
+    FRONTMATTER_ORDER = %w[
+      id date status supersedes superseded-by
+      affected-paths affected-components migration_confidence
+    ].freeze
+    QUOTED_FIELDS = %w[id supersedes superseded-by].freeze
+    SECTION_ORDER = ["Context", "Decision", "Consequences", "Alternatives rejected"].freeze
+
+    attr_reader :frontmatter, :title, :sections
+
+    def self.parse(text)
+      match = text.match(/\A---\n(.*?)\n---\n(.*)\z/m)
+      raise ParseError, "no frontmatter block found" unless match
+
+      frontmatter = YAML.safe_load(match[1], permitted_classes: [Date]) || {}
+      body = match[2]
+      heading = body[/^# +(.+)$/, 1].to_s.strip
+      title = heading.sub(/\A\S+\s+—\s+/, "")
+      sections = {}
+      body.scan(/^## ([^\n]+)\n(.*?)(?=^## |\z)/m) do |name, content|
+        sections[name.strip] = content.strip
+      end
+      new(frontmatter: frontmatter, title: title, sections: sections)
+    end
+
+    def initialize(frontmatter:, title:, sections:)
+      @frontmatter = frontmatter
+      @title = title
+      @sections = sections
+    end
+
+    def id      = frontmatter["id"]&.to_s
+    def status  = frontmatter["status"]
+    def date    = frontmatter["date"]
+
+    def supersede_with(new_id)
+      frontmatter["superseded-by"] = new_id
+      frontmatter["status"] = "Superseded"
+    end
+
+    def to_markdown
+      lines = ["---"]
+      FRONTMATTER_ORDER.each do |field|
+        next unless frontmatter.key?(field)
+        lines << "#{field}: #{format_value(field, frontmatter[field])}".rstrip
+      end
+      (frontmatter.keys - FRONTMATTER_ORDER).each do |field|
+        lines << "#{field}: #{format_value(field, frontmatter[field])}".rstrip
+      end
+      lines << "---" << "" << "# #{id} — #{title}"
+      sections.each do |name, content|
+        lines << "" << "## #{name}" << "" << content
+      end
+      lines.join("\n") + "\n"
+    end
+
+    private
+
+    def format_value(field, value)
+      case value
+      when nil   then ""
+      when Array then "[#{value.join(', ')}]"
+      when Date  then value.iso8601
+      else
+        text = value.to_s
+        if QUOTED_FIELDS.include?(field) || text.match?(/[:#]/)
+          text.inspect
+        else
+          text
+        end
+      end
+    end
+  end
+end
diff --git a/plugins/os-adr/lib/adr/repository.rb b/plugins/os-adr/lib/adr/repository.rb
new file mode 100644
index 0000000..0f94bb4
--- /dev/null
+++ b/plugins/os-adr/lib/adr/repository.rb
@@ -0,0 +1,55 @@
+module Adr
+  # The docs/adr/ directory of one project. Scans files, derives the next ID
+  # from directory contents (never the index), reads and writes records.
+  class Repository
+    INDEX_FILENAME = "README.md"
+
+    Entry = Struct.new(:path, :record)
+
+    def initialize(root:)
+      @root = root
+    end
+
+    attr_reader :root
+
+    def dir        = File.join(root, "docs", "adr")
+    def index_path = File.join(dir, INDEX_FILENAME)
+
+    def exists?
+      Dir.exist?(dir) && File.exist?(index_path)
+    end
+
+    def adr_paths
+      Dir.glob(File.join(dir, "[0-9][0-9][0-9][0-9]-*.md")).sort
+    end
+
+    # Only files in the os-adr format count as entries; legacy frontmatter-less
+    # files sharing docs/adr/ (pre-migration sources) are skipped, not errors.
+    def entries
+      adr_paths.filter_map do |path|
+        Entry.new(path, Record.parse(File.read(path)))
+      rescue Record::ParseError, Psych::SyntaxError
+        nil
+      end
+    end
+
+    def next_id
+      max = adr_paths.map { |path| File.basename(path)[0, 4].to_i }.max || 0
+      format("%04d", max + 1)
+    end
+
+    def find(id)
+      entries.find { |entry| entry.record.id == id }
+    end
+
+    def write(record)
+      path = File.join(dir, "#{record.id}-#{Adr.slugify(record.title)}.md")
+      File.write(path, record.to_markdown)
+      path
+    end
+
+    def write_entry(entry)
+      File.write(entry.path, entry.record.to_markdown)
+    end
+  end
+end
diff --git a/plugins/os-adr/lib/adr/template.rb b/plugins/os-adr/lib/adr/template.rb
new file mode 100644
index 0000000..b912c8d
--- /dev/null
+++ b/plugins/os-adr/lib/adr/template.rb
@@ -0,0 +1,25 @@
+module Adr
+  # Renders templates/adr.md by substituting {{key}} placeholders.
+  # Arrays render as YAML flow sequences; missing keys render empty.
+  class Template
+    def initialize(path:)
+      @path = path
+    end
+
+    def render(values)
+      File.read(@path).gsub(/\{\{(\w+)\}\}/) do
+        format_value(values[Regexp.last_match(1).to_sym])
+      end.gsub(/ +$/, "")
+    end
+
+    private
+
+    def format_value(value)
+      case value
+      when nil   then ""
+      when Array then "[#{value.join(', ')}]"
+      else value.to_s
+      end
+    end
+  end
+end
diff --git a/plugins/os-adr/skills/create/SKILL.md b/plugins/os-adr/skills/create/SKILL.md
new file mode 100644
index 0000000..bf17859
--- /dev/null
+++ b/plugins/os-adr/skills/create/SKILL.md
@@ -0,0 +1,43 @@
+---
+name: create
+description: Record an architecture decision as a correctly templated, numbered, and indexed ADR in one invocation. Use when a decision of real consequence (irreversible, cross-cutting, contested, or convention-setting) has just been made or is being made — including unprompted, when you recognize such a decision mid-task. Invoked by `/os-adr:create`.
+---
+
+Create a new ADR under `docs/adr/`. All mechanical work (ID assignment, templating, index
+regeneration, supersession bookkeeping) is done by the plugin's CLI — never hand-author the
+file, number, or index.
+
+## 1. Gather the decision content
+
+From the conversation (ask only for what you cannot infer):
+
+- **title** — short, imperative or noun-phrase ("Use Postgres for persistence")
+- **context** — the forces/problem that made a decision necessary
+- **decision** — what was decided, stated actively
+- **consequences** — what becomes easier/harder as a result
+- **alternatives** — options considered and *why each was rejected* (highest-value field; push
+  for real reasons, not placeholders)
+- **affected_paths** / **affected_components** — repo paths and component names this decision
+  governs (used for deterministic retrieval later; be concrete, e.g. `["src/db/", "config/database.yml"]`)
+- **status** — default `Accepted`; use `Proposed` only if the user says it isn't settled
+- **supersedes** — 4-digit ID, only if this replaces an existing ADR (check `docs/adr/README.md`)
+
+If the project has no `docs/adr/` yet, run `/os-adr:init` first (or its CLI, step 2's
+`adr-init`).
+
+## 2. Invoke the CLI
+
+```bash
+echo '{"title": "...", "context": "...", "decision": "...", "consequences": "...",
+  "alternatives": "...", "affected_paths": ["..."], "affected_components": ["..."]}' \
+  | ruby ${CLAUDE_PLUGIN_ROOT}/bin/adr-new --root <project-root>
+```
+
+Optional JSON fields: `status`, `date` (ISO), `supersedes` ("NNNN"). The CLI prints the created
+file path, sets `superseded-by` + `Superseded` status on any superseded ADR, and regenerates
+`docs/adr/README.md`. Prefer writing the JSON to a temp file and piping it (`cat file | ...`)
+when section bodies contain quotes.
+
+## 3. Confirm
+
+Show the user the created path and a one-line summary. Do not edit the generated index by hand.
diff --git a/plugins/os-adr/skills/find/SKILL.md b/plugins/os-adr/skills/find/SKILL.md
new file mode 100644
index 0000000..d512d5a
--- /dev/null
+++ b/plugins/os-adr/skills/find/SKILL.md
@@ -0,0 +1,37 @@
+---
+name: find
+description: Retrieve the architecture decision records relevant to files, components, or a topic being worked on — callable at any point mid-task, not just at session start. Use unprompted before changing behavior that may already be decided (a locked choice, a convention, a rejected alternative). Invoked by `/os-adr:find`.
+---
+
+Find the ADRs relevant to what is being worked on. Retrieval is deterministic-first: run the
+CLI, then apply judgment ONLY to the candidates it returns — never read/rank the full ADR
+corpus yourself.
+
+## 1. Run the deterministic CLI
+
+```bash
+ruby ${CLAUDE_PLUGIN_ROOT}/bin/adr-find --root <project-root> \
+  --paths <comma-separated file/dir paths being edited> \
+  --components <comma-separated component names>   # optional
+```
+
+The CLI layers: frontmatter path/component matching → Accepted-only status filter →
+one-hop Graphify graph expansion of the query paths (automatic when `graphify-out/graph.json`
+exists; silently skipped otherwise). Add `--all-statuses` only when history matters (e.g.
+"why was X superseded?").
+
+## 2. Judge the candidate set
+
+Read the returned candidate files (they are few) and decide which actually bear on the task.
+`matched_on` tells you why each surfaced (`path`, `graph`, `component`). Prefer direct path
+matches; treat `graph` matches as "nearby decisions worth a look".
+
+## 3. Act on what you find
+
+- If the in-progress change **conflicts** with an Accepted ADR: stop, surface the specific ADR
+  (ID + title + the constraint), and either adjust the approach or propose superseding it via
+  `/os-adr:create` with `supersedes`.
+- If a candidate merely informs the work, cite it and continue.
+- If the CLI returns nothing, say so — do not fall back to sweeping `docs/adr/` unless the user
+  asks; an empty result usually means no decision governs those paths (or `affected-paths`
+  frontmatter is missing on older ADRs — worth mentioning once).
diff --git a/plugins/os-adr/skills/init/SKILL.md b/plugins/os-adr/skills/init/SKILL.md
new file mode 100644
index 0000000..5731441
--- /dev/null
+++ b/plugins/os-adr/skills/init/SKILL.md
@@ -0,0 +1,26 @@
+---
+name: init
+description: Set up the ADR system (docs/adr/ + generated index) in a project that has none, and clear any session-reminder suppression. Invoked by `/os-adr:init`.
+---
+
+Initialize the standard ADR layout in the current project. Everything is mechanical — run the
+CLI, don't create the files yourself:
+
+```bash
+ruby ${CLAUDE_PLUGIN_ROOT}/bin/adr-init --root <project-root>
+```
+
+This creates `docs/adr/` with a generated `README.md` index (empty table between markers,
+regenerated in full on every ADR write — never hand-edited), and deletes `.os-adr/suppress` if
+present so the SessionStart reminder resumes normal behavior.
+
+It is idempotent: safe to run in a project that already has ADRs (the index is regenerated from
+the directory contents).
+
+Afterwards:
+
+1. Ensure `.os-adr/` is gitignored (append to `.gitignore` if the project has one and the entry
+   is missing).
+2. If the project has existing ADR-like content in another shape, suggest `/os-adr:migrate`
+   instead of re-authoring.
+3. Offer `/os-adr:create` to record the first decision.
diff --git a/plugins/os-adr/skills/migrate/SKILL.md b/plugins/os-adr/skills/migrate/SKILL.md
new file mode 100644
index 0000000..440aeb4
--- /dev/null
+++ b/plugins/os-adr/skills/migrate/SKILL.md
@@ -0,0 +1,61 @@
+---
+name: migrate
+description: Non-destructively convert a project's existing ADR-like content (numbered files, dated files, monolithic decision logs, prose-embedded decisions) into the standard docs/adr/ format. Sources are never touched; uncertainty is flagged via migration_confidence + a migration report. Invoked by `/os-adr:migrate`.
+---
+
+Migrate existing decision records into the standard ADR format. The mechanical work (detection,
+splitting on unit headings, heuristic field fill, numbering, indexing, report generation) is all
+done by the CLI; your job is ONLY to fill the interpretive fields the CLI's manifest explicitly
+lists, and to present the results.
+
+## 1. Detect (report-only)
+
+```bash
+ruby ${CLAUDE_PLUGIN_ROOT}/bin/adr-detect --root <project-root>
+```
+
+Show the user the detected units and shapes (`numbered_per_file`, `dated_single_file`,
+`topic_single_file`, `monolithic`, `prose_embedded`, `unrecognized`). Confirm they want to
+proceed before converting anything.
+
+## 2. Mechanical pass
+
+```bash
+ruby ${CLAUDE_PLUGIN_ROOT}/bin/adr-migrate --root <project-root>
+```
+
+This writes new ADRs under `docs/adr/` only (sources stay byte-identical), heuristic-fills
+structurally unambiguous fields (status, date, title, ID, provenance, exactly-matching
+sections), regenerates the index, writes `docs/adr/migration-report.md`, and prints a JSON
+**work manifest**.
+
+## 3. LLM fill — manifest-listed fields ONLY
+
+For each manifest entry with non-empty `llm_fields`, write the listed fields (`consequences`,
+`alternatives`, sometimes `status`) from that entry's `source_material` — summarize what the
+source actually says, don't invent. Rules:
+
+- Fill ONLY fields the manifest lists. Fields marked `_not stated in source_` (Decision,
+  Context) must NEVER be invented — leave them; the file stays flagged low-confidence.
+- `status` fills must be one of Proposed/Accepted/Superseded/Deprecated.
+
+Hand the results back to the core for writing — do not edit the ADR files directly:
+
+```bash
+# fills.json: {"docs/adr/0001-x.md": {"consequences": "...", "alternatives": "..."}}
+ruby ${CLAUDE_PLUGIN_ROOT}/bin/adr-migrate --root <project-root> --apply-fills fills.json
+```
+
+(The core only accepts fills into pending-marker fields, so a stray key is a no-op.)
+
+## 4. Present the report
+
+Show `docs/adr/migration-report.md`: per-file confidence, source mapping, the low-confidence
+flag rate, and any `unrecognized` sources needing manual handling. Suggest the user review
+low-confidence files.
+
+## 5. Old-system deletion — separate, explicit approval only
+
+Do NOT delete, move, or edit the old decision files as part of migration. Only after the user
+has reviewed the report, offer removal of the old system as a distinct step, and perform it only
+on their explicit approval of the specific files to remove.
diff --git a/plugins/os-adr/templates/adr.md b/plugins/os-adr/templates/adr.md
new file mode 100644
index 0000000..4ff6321
--- /dev/null
+++ b/plugins/os-adr/templates/adr.md
@@ -0,0 +1,27 @@
+---
+id: "{{id}}"
+date: {{date}}
+status: {{status}}
+supersedes: {{supersedes}}
+superseded-by:
+affected-paths: {{affected_paths}}
+affected-components: {{affected_components}}
+---
+
+# {{id}} — {{title}}
+
+## Context
+
+{{context}}
+
+## Decision
+
+{{decision}}
+
+## Consequences
+
+{{consequences}}
+
+## Alternatives rejected
+
+{{alternatives}}
diff --git a/plugins/os-adr/tests/all.rb b/plugins/os-adr/tests/all.rb
new file mode 100644
index 0000000..8336904
--- /dev/null
+++ b/plugins/os-adr/tests/all.rb
@@ -0,0 +1,2 @@
+# Run the whole suite: ruby tests/all.rb
+Dir.glob(File.join(__dir__, "*_test.rb")).sort.each { |f| require f }
diff --git a/plugins/os-adr/tests/cli_test.rb b/plugins/os-adr/tests/cli_test.rb
new file mode 100644
index 0000000..6c698a4
--- /dev/null
+++ b/plugins/os-adr/tests/cli_test.rb
@@ -0,0 +1,89 @@
+require_relative "test_helper"
+require "open3"
+require "json"
+
+class CliTest < Minitest::Test
+  include AdrTestHelpers
+
+  ADR_NEW = File.join(PLUGIN_ROOT, "bin", "adr-new")
+  ADR_INIT = File.join(PLUGIN_ROOT, "bin", "adr-init")
+
+  def run_new(root, input)
+    out, err, status = Open3.capture3("ruby", ADR_NEW, "--root", root, stdin_data: JSON.generate(input))
+    [out.strip, err, status]
+  end
+
+  def run_init(root)
+    Open3.capture3("ruby", ADR_INIT, "--root", root)
+  end
+
+  def test_init_creates_dir_and_index_and_clears_suppress
+    Dir.mktmpdir do |root|
+      FileUtils.mkdir_p(File.join(root, ".os-adr"))
+      File.write(File.join(root, ".os-adr", "suppress"), "")
+      _out, _err, status = run_init(root)
+      assert status.success?
+      repo = Adr::Repository.new(root: root)
+      assert repo.exists?
+      refute File.exist?(File.join(root, ".os-adr", "suppress"))
+    end
+  end
+
+  def test_init_is_idempotent_and_preserves_existing_adrs
+    with_project("0001-a.md" => sample_adr) do |root|
+      _out, _err, status = run_init(root)
+      assert status.success?
+      assert_includes File.read(Adr::Repository.new(root: root).index_path), "0001"
+    end
+  end
+
+  def test_new_creates_templated_numbered_indexed_adr
+    with_project("0001-a.md" => sample_adr(id: "0001")) do |root|
+      out, err, status = run_new(root, title: "Use Postgres", context: "We need a DB.",
+                                       decision: "Postgres.", consequences: "Ops.",
+                                       alternatives: "SQLite.", affected_paths: ["db/"],
+                                       affected_components: ["storage"])
+      assert status.success?, err
+      assert out.end_with?("0002-use-postgres.md")
+      record = Adr::Record.parse(File.read(out))
+      assert_equal "0002", record.id
+      assert_equal "Accepted", record.status
+      assert_equal "Use Postgres", record.title
+      assert_equal ["db/"], record.frontmatter["affected-paths"]
+      Adr::Record::SECTION_ORDER.each { |s| assert record.sections.key?(s), "missing section #{s}" }
+      assert_includes File.read(Adr::Repository.new(root: root).index_path),
+                      "[Use Postgres](0002-use-postgres.md)"
+    end
+  end
+
+  def test_new_supersession_is_mechanical
+    with_project("0004-old-way.md" => sample_adr(id: "0004", title: "Old way")) do |root|
+      out, err, status = run_new(root, title: "New way", supersedes: "0004")
+      assert status.success?, err
+      new_record = Adr::Record.parse(File.read(out))
+      assert_equal "0004", new_record.frontmatter["supersedes"]
+      old = Adr::Repository.new(root: root).find("0004")
+      assert_equal new_record.id, old.record.frontmatter["superseded-by"]
+      assert_equal "Superseded", old.record.status
+      index = File.read(Adr::Repository.new(root: root).index_path)
+      assert_includes index, "| Superseded |"
+      assert_includes index, "| Accepted |"
+    end
+  end
+
+  def test_new_requires_title
+    with_project do |root|
+      _out, err, status = run_new(root, {})
+      refute status.success?
+      assert_includes err, "title"
+    end
+  end
+
+  def test_new_fails_cleanly_without_init
+    Dir.mktmpdir do |root|
+      _out, err, status = run_new(root, title: "X")
+      refute status.success?
+      assert_includes err, "adr-init"
+    end
+  end
+end
diff --git a/plugins/os-adr/tests/detector_test.rb b/plugins/os-adr/tests/detector_test.rb
new file mode 100644
index 0000000..d9495eb
--- /dev/null
+++ b/plugins/os-adr/tests/detector_test.rb
@@ -0,0 +1,175 @@
+require_relative "test_helper"
+
+class DetectorTest < Minitest::Test
+  include AdrTestHelpers
+
+  NUMBERED = <<~MD
+    # 0001 — PocketBase backend
+
+    ## Status
+
+    Accepted
+
+    ## Context
+
+    Need a backend.
+
+    ## Decision
+
+    PocketBase.
+
+    ## Consequences
+
+    Single binary.
+
+    ## Alternatives considered
+
+    Rails — too heavy.
+  MD
+
+  DATED = <<~MD
+    # JSX conversion pivot
+
+    **Date:** 2024-12-15
+    **Status:** Accepted
+
+    ## Context
+
+    Conversion stalled.
+
+    ## Decision
+
+    Pivot to JSX.
+  MD
+
+  TOPIC = <<~MD
+    # bin/lint-fix
+
+    Status: Confirmed
+    Decided: 2026-02-01
+
+    ## Decision
+
+    One lint entry point.
+
+    ## Why
+
+    Consistency across repos.
+  MD
+
+  MONOLITHIC = <<~MD
+    # Decision log
+
+    ## ADR-001: Use SQLite
+
+    Status: Accepted
+
+    ### Context
+
+    Need storage.
+
+    ### Decision
+
+    SQLite.
+
+    ## ADR-002: Drop SQLite
+
+    Status: Accepted
+
+    ### Decision
+
+    Postgres instead.
+  MD
+
+  PROSE = <<~MD
+    # Gaps and decisions (LOCKED)
+
+    Long narrative preamble.
+
+    Decision 26: Single entry point
+
+    We route everything through one command.
+
+    Decision 27: No config file
+
+    Defaults only.
+  MD
+
+  UNRECOGNIZED = "# Meeting notes\n\nWe talked about stuff.\n"
+
+  MULTI_STATUS_NO_HEADINGS = <<~MD
+    # DECISIONS
+
+    2026-01-05 — Hosting
+    Status: Confirmed
+    We host on X.
+
+    2026-02-10 — Pricing
+    Status: Deferred
+    Revisit in Q3.
+  MD
+
+  def fixture_project
+    Dir.mktmpdir do |root|
+      decisions = File.join(root, "docs", "decisions")
+      adr = File.join(root, "docs", "adr")
+      FileUtils.mkdir_p(decisions)
+      FileUtils.mkdir_p(adr)
+      File.write(File.join(adr, "0001-pocketbase-backend.md"), NUMBERED)
+      File.write(File.join(adr, "README.md"), "index")
+      File.write(File.join(decisions, "2024-12-15-jsx-pivot.md"), DATED)
+      File.write(File.join(decisions, "bin-lint-fix.md"), TOPIC)
+      File.write(File.join(decisions, "decision-log.md"), MONOLITHIC)
+      File.write(File.join(decisions, "gaps-and-decisions.md"), PROSE)
+      File.write(File.join(decisions, "notes.md"), UNRECOGNIZED)
+      File.write(File.join(root, "DECISIONS.md"), MULTI_STATUS_NO_HEADINGS)
+      yield root, Adr::Detector.new(root: root).units
+    end
+  end
+
+  def shapes_by_source(units)
+    units.group_by(&:source_path).transform_values { |us| us.map(&:shape).uniq }
+  end
+
+  def test_classifies_each_surveyed_shape
+    fixture_project do |_root, units|
+      shapes = shapes_by_source(units)
+      assert_equal [:numbered_per_file], shapes["docs/adr/0001-pocketbase-backend.md"]
+      assert_equal [:dated_single_file], shapes["docs/decisions/2024-12-15-jsx-pivot.md"]
+      assert_equal [:topic_single_file], shapes["docs/decisions/bin-lint-fix.md"]
+      assert_equal [:monolithic], shapes["docs/decisions/decision-log.md"]
+      assert_equal [:prose_embedded], shapes["docs/decisions/gaps-and-decisions.md"]
+      assert_equal [:unrecognized], shapes["docs/decisions/notes.md"]
+    end
+  end
+
+  def test_monolithic_units_are_enumerated_per_heading
+    fixture_project do |_root, units|
+      mono = units.select { |u| u.source_path == "docs/decisions/decision-log.md" }
+      assert_equal 2, mono.size
+      assert_equal ["Use SQLite", "Drop SQLite"], mono.map(&:title)
+      assert_includes mono.first.body, "SQLite."
+      refute_includes mono.first.body, "Postgres"
+    end
+  end
+
+  def test_prose_embedded_units_are_enumerated
+    fixture_project do |_root, units|
+      prose = units.select { |u| u.source_path == "docs/decisions/gaps-and-decisions.md" }
+      assert_equal ["Single entry point", "No config file"], prose.map(&:title)
+    end
+  end
+
+  def test_multi_decision_file_without_unit_headings_is_unrecognized
+    fixture_project do |_root, units|
+      assert_equal [:unrecognized], shapes_by_source(units)["DECISIONS.md"]
+    end
+  end
+
+  def test_index_and_already_migrated_files_are_excluded
+    with_project("0001-a.md" => sample_adr) do |root|
+      File.write(File.join(root, "docs", "adr", "README.md"), "index")
+      assert_empty Adr::Detector.new(root: root).units
+    end
+  end
+end
diff --git a/plugins/os-adr/tests/finder_test.rb b/plugins/os-adr/tests/finder_test.rb
new file mode 100644
index 0000000..9e2be6a
--- /dev/null
+++ b/plugins/os-adr/tests/finder_test.rb
@@ -0,0 +1,133 @@
+require_relative "test_helper"
+require "json"
+
+class FinderTest < Minitest::Test
+  include AdrTestHelpers
+
+  def adr(id:, title:, status: "Accepted", paths: [], components: [])
+    <<~MD
+      ---
+      id: "#{id}"
+      date: 2026-07-03
+      status: #{status}
+      supersedes:
+      superseded-by:
+      affected-paths: [#{paths.join(', ')}]
+      affected-components: [#{components.join(', ')}]
+      ---
+
+      # #{id} — #{title}
+
+      ## Context
+
+      c
+
+      ## Decision
+
+      d
+
+      ## Consequences
+
+      q
+
+      ## Alternatives rejected
+
+      a
+    MD
+  end
+
+  def storage_project(&block)
+    with_project(
+      "0001-storage.md" => adr(id: "0001", title: "Storage", paths: %w[lib/storage db/],
+                               components: %w[storage]),
+      "0002-old-storage.md" => adr(id: "0002", title: "Old storage", status: "Superseded",
+                                   paths: %w[lib/storage]),
+      "0003-auth.md" => adr(id: "0003", title: "Auth", paths: %w[lib/auth],
+                            components: %w[auth]),
+      &block
+    )
+  end
+
+  def finder(root, graph: nil)
+    Adr::Finder.new(repository: Adr::Repository.new(root: root),
+                    graph_path: graph || File.join(root, "graphify-out", "graph.json"))
+  end
+
+  def test_path_prefix_match_narrows
+    storage_project do |root|
+      results = finder(root).find(paths: ["lib/storage/adapter.rb"])
+      assert_equal ["0001"], results.map { |r| r["id"] }
+      assert_equal ["path"], results.first["matched_on"]
+    end
+  end
+
+  def test_directory_query_matches_file_scoped_adr
+    storage_project do |root|
+      results = finder(root).find(paths: ["db"])
+      assert_equal ["0001"], results.map { |r| r["id"] }
+    end
+  end
+
+  def test_component_match
+    storage_project do |root|
+      results = finder(root).find(components: ["auth"])
+      assert_equal ["0003"], results.map { |r| r["id"] }
+      assert_equal ["component"], results.first["matched_on"]
+    end
+  end
+
+  def test_accepted_only_by_default_with_override
+    storage_project do |root|
+      default = finder(root).find(paths: ["lib/storage"])
+      assert_equal ["0001"], default.map { |r| r["id"] }
+      all = finder(root).find(paths: ["lib/storage"], all_statuses: true)
+      assert_equal %w[0001 0002], all.map { |r| r["id"] }.sort
+    end
+  end
+
+  def test_no_match_returns_empty
+    storage_project do |root|
+      assert_empty finder(root).find(paths: ["src/ui"])
+    end
+  end
+
+  GRAPH = {
+    "nodes" => [
+      { "id" => "n1", "source_file" => "src/api/handler.rb" },
+      { "id" => "n2", "source_file" => "lib/storage/adapter.rb" },
+      { "id" => "n3", "source_file" => "lib/auth/session.rb" }
+    ],
+    "links" => [
+      { "source" => "n1", "target" => "n2" }
+    ]
+  }.freeze
+
+  def test_graph_layer_expands_one_hop
+    storage_project do |root|
+      graph_dir = File.join(root, "graphify-out")
+      FileUtils.mkdir_p(graph_dir)
+      File.write(File.join(graph_dir, "graph.json"), JSON.generate(GRAPH))
+      # src/api/handler.rb is in no ADR's affected-paths, but is one hop from
+      # lib/storage/adapter.rb which 0001 covers.
+      results = finder(root).find(paths: ["src/api/handler.rb"])
+      assert_equal ["0001"], results.map { |r| r["id"] }
+      assert_equal ["graph"], results.first["matched_on"]
+    end
+  end
+
+  def test_degrades_gracefully_without_graph
+    storage_project do |root|
+      assert_empty finder(root).find(paths: ["src/api/handler.rb"])
+      assert_equal ["0001"], finder(root).find(paths: ["lib/storage"]).map { |r| r["id"] }
+    end
+  end
+
+  def test_degrades_gracefully_on_corrupt_graph
+    storage_project do |root|
+      graph_dir = File.join(root, "graphify-out")
+      FileUtils.mkdir_p(graph_dir)
+      File.write(File.join(graph_dir, "graph.json"), "{not json")
+      assert_equal ["0001"], finder(root).find(paths: ["lib/storage"]).map { |r| r["id"] }
+    end
+  end
+end
diff --git a/plugins/os-adr/tests/hook_test.py b/plugins/os-adr/tests/hook_test.py
new file mode 100644
index 0000000..bc5a487
--- /dev/null
+++ b/plugins/os-adr/tests/hook_test.py
@@ -0,0 +1,160 @@
+"""Tests for hooks/session_start.py. Run: python3 tests/hook_test.py"""
+
+import io
+import json
+import subprocess
+import sys
+import unittest
+from datetime import date
+from pathlib import Path
+from tempfile import TemporaryDirectory
+
+PLUGIN_ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(PLUGIN_ROOT / "hooks"))
+
+from session_start import (  # noqa: E402
+    ABSENT_NOTE,
+    PRESENT_NOTE,
+    StateDir,
+    adr_system_present,
+    decide,
+    find_project_root,
+    run,
+)
+
+TODAY = date(2026, 7, 3)
+YESTERDAY = date(2026, 7, 2)
+
+
+class DecideTest(unittest.TestCase):
+    def test_present_injects_context_naming_both_surfaces(self):
+        outcome, note = decide(True, False, None, TODAY)
+        self.assertEqual("context", outcome)
+        self.assertIn("/os-adr:create", note)
+        self.assertIn("/os-adr:find", note)
+
+    def test_present_wins_even_if_suppressed(self):
+        self.assertEqual("context", decide(True, True, None, TODAY)[0])
+
+    def test_absent_unsuppressed_banners_with_init_and_migrate(self):
+        outcome, note = decide(False, False, None, TODAY)
+        self.assertEqual("banner", outcome)
+        self.assertIn("/os-adr:init", note)
+        self.assertIn("/os-adr:migrate", note)
+
+    def test_absent_suppressed_is_silent(self):
+        self.assertEqual(("silent", None), decide(False, True, None, TODAY))
+
+    def test_absent_reminded_today_is_silent(self):
+        self.assertEqual(("silent", None), decide(False, False, TODAY, TODAY))
+
+    def test_absent_reminded_yesterday_banners_again(self):
+        self.assertEqual("banner", decide(False, False, YESTERDAY, TODAY)[0])
+
+
+class RunTest(unittest.TestCase):
+    def project(self, tmp, with_adr=False):
+        root = Path(tmp)
+        (root / ".git").mkdir()
+        if with_adr:
+            (root / "docs" / "adr").mkdir(parents=True)
+            (root / "docs" / "adr" / "README.md").write_text("index")
+        return root
+
+    def test_present_emits_additional_context(self):
+        with TemporaryDirectory() as tmp:
+            root = self.project(tmp, with_adr=True)
+            out = io.StringIO()
+            self.assertEqual("context", run(root, TODAY, out))
+            payload = json.loads(out.getvalue())
+            self.assertEqual(
+                PRESENT_NOTE,
+                payload["hookSpecificOutput"]["additionalContext"],
+            )
+            self.assertFalse((root / ".os-adr").exists(), "present case must not create state")
+
+    def test_absent_emits_banner_and_stamps_snooze(self):
+        with TemporaryDirectory() as tmp:
+            root = self.project(tmp)
+            out = io.StringIO()
+            self.assertEqual("banner", run(root, TODAY, out))
+            self.assertEqual(ABSENT_NOTE, json.loads(out.getvalue())["systemMessage"])
+            self.assertEqual(
+                TODAY.isoformat(), (root / ".os-adr" / "last_reminded").read_text()
+            )
+
+    def test_second_run_same_day_is_silent(self):
+        with TemporaryDirectory() as tmp:
+            root = self.project(tmp)
+            run(root, TODAY, io.StringIO())
+            out = io.StringIO()
+            self.assertEqual("silent", run(root, TODAY, out))
+            self.assertEqual("", out.getvalue())
+
+    def test_next_day_banners_again(self):
+        with TemporaryDirectory() as tmp:
+            root = self.project(tmp)
+            run(root, YESTERDAY, io.StringIO())
+            self.assertEqual("banner", run(root, TODAY, io.StringIO()))
+
+    def test_suppressed_project_stays_silent(self):
+        with TemporaryDirectory() as tmp:
+            root = self.project(tmp)
+            (root / ".os-adr").mkdir()
+            (root / ".os-adr" / "suppress").write_text("")
+            out = io.StringIO()
+            self.assertEqual("silent", run(root, TODAY, out))
+            self.assertEqual("", out.getvalue())
+
+    def test_corrupt_snooze_stamp_is_treated_as_unset(self):
+        with TemporaryDirectory() as tmp:
+            root = self.project(tmp)
+            (root / ".os-adr").mkdir()
+            (root / ".os-adr" / "last_reminded").write_text("not-a-date")
+            self.assertEqual("banner", run(root, TODAY, io.StringIO()))
+
+
+class ProjectRootTest(unittest.TestCase):
+    def test_walks_up_to_git_root(self):
+        with TemporaryDirectory() as tmp:
+            root = Path(tmp)
+            (root / ".git").mkdir()
+            nested = root / "a" / "b"
+            nested.mkdir(parents=True)
+            self.assertEqual(root, find_project_root(nested))
+
+    def test_none_outside_any_git_project(self):
+        with TemporaryDirectory() as tmp:
+            self.assertIsNone(find_project_root(Path(tmp)))
+
+
+class PythonRubyAgreementTest(unittest.TestCase):
+    """Shared fixture tree: the Python existence check and Ruby
+    Adr::Repository#exists? must give the same answer (design D3 risk)."""
+
+    def ruby_exists(self, root):
+        script = (
+            f'require "{PLUGIN_ROOT}/lib/adr"; '
+            f'puts Adr::Repository.new(root: {str(root)!r}).exists?'
+        )
+        out = subprocess.run(
+            ["ruby", "-e", script], capture_output=True, text=True, check=True
+        )
+        return out.stdout.strip() == "true"
+
+    def assert_agreement(self, root, expected):
+        self.assertEqual(expected, adr_system_present(root), "python check")
+        self.assertEqual(expected, self.ruby_exists(root), "ruby check")
+
+    def test_agreement_across_fixture_states(self):
+        with TemporaryDirectory() as tmp:
+            root = Path(tmp)
+            self.assert_agreement(root, False)          # nothing
+            (root / "docs" / "adr").mkdir(parents=True)
+            self.assert_agreement(root, False)          # dir, no index
+            (root / "docs" / "adr" / "README.md").write_text("index")
+            self.assert_agreement(root, True)           # dir + index
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/plugins/os-adr/tests/index_test.rb b/plugins/os-adr/tests/index_test.rb
new file mode 100644
index 0000000..c6addcd
--- /dev/null
+++ b/plugins/os-adr/tests/index_test.rb
@@ -0,0 +1,79 @@
+require_relative "test_helper"
+
+class IndexTest < Minitest::Test
+  include AdrTestHelpers
+
+  GOLDEN = <<~MD
+    <!-- Generated by os-adr. Do not hand-edit the table: it is regenerated in full on every ADR write. -->
+
+    # Architecture Decision Records
+
+    One file per decision, `NNNN-kebab-title.md`, created via `/os-adr:create`.
+
+    <!-- adr-index:begin -->
+    | ID | Title | Status | Date |
+    | --- | --- | --- | --- |
+    | 0001 | [Use Postgres](0001-use-postgres.md) | Accepted | 2026-07-03 |
+    | 0002 | [Use Redis](0002-use-redis.md) | Accepted | 2026-07-03 |
+    <!-- adr-index:end -->
+  MD
+
+  def two_adr_project(&block)
+    with_project("0001-use-postgres.md" => sample_adr(id: "0001", title: "Use Postgres"),
+                 "0002-use-redis.md" => sample_adr(id: "0002", title: "Use Redis"), &block)
+  end
+
+  def test_regenerates_full_table_matching_golden
+    two_adr_project do |root|
+      repo = Adr::Repository.new(root: root)
+      Adr::Index.new(repository: repo).regenerate!
+      assert_equal GOLDEN, File.read(repo.index_path)
+    end
+  end
+
+  def test_self_heals_after_manual_table_edit
+    two_adr_project do |root|
+      repo = Adr::Repository.new(root: root)
+      index = Adr::Index.new(repository: repo)
+      index.regenerate!
+      vandalized = File.read(repo.index_path)
+        .sub("| 0001 | [Use Postgres](0001-use-postgres.md) | Accepted | 2026-07-03 |\n", "")
+        .sub("Use Redis", "Hand-Edited Title")
+      File.write(repo.index_path, vandalized)
+      index.regenerate!
+      assert_equal GOLDEN, File.read(repo.index_path)
+    end
+  end
+
+  def test_preserves_prose_outside_markers
+    two_adr_project do |root|
+      repo = Adr::Repository.new(root: root)
+      index = Adr::Index.new(repository: repo)
+      index.regenerate!
+      File.write(repo.index_path, "Custom preamble.\n\n" + File.read(repo.index_path) + "\nCustom footer.\n")
+      index.regenerate!
+      content = File.read(repo.index_path)
+      assert_includes content, "Custom preamble."
+      assert_includes content, "Custom footer."
+      assert_includes content, "| 0001 | [Use Postgres](0001-use-postgres.md) | Accepted | 2026-07-03 |"
+    end
+  end
+
+  def test_empty_directory_yields_empty_table
+    with_project do |root|
+      repo = Adr::Repository.new(root: root)
+      Adr::Index.new(repository: repo).regenerate!
+      content = File.read(repo.index_path)
+      assert_includes content, Adr::Index::BEGIN_MARKER
+      refute_includes content, "| ID |"
+    end
+  end
+
+  def test_index_declares_itself_generated
+    with_project do |root|
+      repo = Adr::Repository.new(root: root)
+      Adr::Index.new(repository: repo).regenerate!
+      assert_match(/do not hand-edit/i, File.read(repo.index_path))
+    end
+  end
+end
diff --git a/plugins/os-adr/tests/migrator_test.rb b/plugins/os-adr/tests/migrator_test.rb
new file mode 100644
index 0000000..5255b81
--- /dev/null
+++ b/plugins/os-adr/tests/migrator_test.rb
@@ -0,0 +1,208 @@
+require_relative "test_helper"
+require "digest"
+
+class MigratorTest < Minitest::Test
+  include AdrTestHelpers
+
+  FULL_SOURCE = <<~MD
+    # 0001 — PocketBase backend
+
+    ## Status
+
+    Accepted
+
+    ## Context
+
+    Need a backend.
+
+    ## Decision
+
+    PocketBase.
+
+    ## Consequences
+
+    Single binary.
+
+    ## Alternatives considered
+
+    Rails — too heavy.
+  MD
+
+  PARTIAL_SOURCE = <<~MD
+    # bin/lint-fix
+
+    Status: Confirmed
+    Decided: 2026-02-01
+
+    ## Context
+
+    Lint config drifted per repo.
+
+    ## Decision
+
+    One lint entry point.
+
+    ## Why
+
+    Consistency across repos.
+  MD
+
+  BARE_SOURCE = <<~MD
+    # Something happened
+
+    Status: Accepted
+
+    We did a thing, narratively.
+  MD
+
+  def project(sources)
+    Dir.mktmpdir do |root|
+      decisions = File.join(root, "docs", "decisions")
+      FileUtils.mkdir_p(decisions)
+      sources.each { |name, body| File.write(File.join(decisions, name), body) }
+      repo = Adr::Repository.new(root: root)
+      migrator = Adr::Migrator.new(repository: repo, detector: Adr::Detector.new(root: root))
+      yield root, repo, migrator
+    end
+  end
+
+  def test_sources_stay_byte_identical
+    project("0001-full.md" => FULL_SOURCE, "partial.md" => PARTIAL_SOURCE) do |root, _repo, migrator|
+      before = Dir.glob(File.join(root, "docs", "decisions", "*"))
+                  .to_h { |p| [p, Digest::SHA256.file(p).hexdigest] }
+      migrator.migrate!
+      before.each { |path, sha| assert_equal sha, Digest::SHA256.file(path).hexdigest }
+    end
+  end
+
+  def test_unambiguous_fields_fill_mechanically_high_confidence
+    project("0001-full.md" => FULL_SOURCE) do |_root, repo, migrator|
+      manifest = migrator.migrate!
+      entry = repo.entries.first
+      record = entry.record
+      assert_equal "Accepted", record.status
+      assert_equal "PocketBase backend", record.title
+      assert_equal "Need a backend.", record.sections["Context"]
+      assert_equal "Rails — too heavy.", record.sections["Alternatives rejected"]
+      assert_equal "high", record.frontmatter["migration_confidence"]
+      assert_equal "docs/decisions/0001-full.md", record.frontmatter["migration_source"]
+      assert_empty manifest["migrated"].first["llm_fields"]
+    end
+  end
+
+  def test_interpretive_fields_get_pending_markers_and_manifest
+    project("partial.md" => PARTIAL_SOURCE) do |_root, repo, migrator|
+      manifest = migrator.migrate!
+      record = repo.entries.first.record
+      assert_equal "Accepted", record.status, "Confirmed maps mechanically to Accepted"
+      assert_equal Date.new(2026, 2, 1), record.date
+      assert_equal Adr::Migrator::PENDING_MARKER, record.sections["Consequences"]
+      assert_equal Adr::Migrator::PENDING_MARKER, record.sections["Alternatives rejected"]
+      assert_equal "medium", record.frontmatter["migration_confidence"]
+      entry = manifest["migrated"].first
+      assert_equal %w[consequences alternatives], entry["llm_fields"]
+      assert_includes entry["source_material"], "Consistency across repos."
+    end
+  end
+
+  def test_fallback_heading_mapping_caps_confidence_at_medium
+    source = <<~MD
+      # Topic
+
+      Status: Accepted
+
+      ## Decision
+
+      Do the thing.
+
+      ## Why
+
+      Because reasons.
+
+      ## Consequences
+
+      Some.
+
+      ## Alternatives rejected
+
+      None seriously.
+    MD
+    project("topic.md" => source) do |_root, repo, migrator|
+      migrator.migrate!
+      record = repo.entries.first.record
+      assert_equal "Because reasons.", record.sections["Context"], "Why maps to Context"
+      assert_equal "medium", record.frontmatter["migration_confidence"],
+                   "heuristic heading mapping is flagged"
+    end
+  end
+
+  def test_missing_core_fields_get_not_stated_markers_and_low_confidence
+    project("bare.md" => BARE_SOURCE) do |_root, repo, migrator|
+      migrator.migrate!
+      record = repo.entries.first.record
+      assert_equal Adr::Migrator::NOT_STATED_MARKER, record.sections["Decision"]
+      assert_equal Adr::Migrator::NOT_STATED_MARKER, record.sections["Context"]
+      assert_equal "low", record.frontmatter["migration_confidence"]
+    end
+  end
+
+  def test_apply_fills_only_touches_pending_fields
+    project("partial.md" => PARTIAL_SOURCE, "bare.md" => BARE_SOURCE) do |root, repo, migrator|
+      manifest = migrator.migrate!
+      pending_file = manifest["migrated"].find { |m| m["source"].include?("partial") }["file"]
+      bare_file = manifest["migrated"].find { |m| m["source"].include?("bare") }["file"]
+
+      migrator.apply_fills!(
+        pending_file => { "consequences" => "Fewer lint debates.",
+                          "alternatives" => "Per-repo configs — rejected, drift." },
+        bare_file => { "decision" => "INVENTED", "context" => "INVENTED" }
+      )
+
+      filled = Adr::Record.parse(File.read(File.join(root, pending_file)))
+      assert_equal "Fewer lint debates.", filled.sections["Consequences"]
+      assert_equal "medium", filled.frontmatter["migration_confidence"], "LLM-filled stays flagged"
+
+      bare = Adr::Record.parse(File.read(File.join(root, bare_file)))
+      assert_equal Adr::Migrator::NOT_STATED_MARKER, bare.sections["Decision"],
+                   "core fields are never LLM-inventable"
+      assert_equal "low", bare.frontmatter["migration_confidence"]
+    end
+  end
+
+  def test_report_lists_files_confidence_sources_and_flag_rate
+    project("0001-full.md" => FULL_SOURCE, "bare.md" => BARE_SOURCE) do |root, _repo, migrator|
+      manifest = migrator.migrate!
+      report = File.read(File.join(root, "docs", "adr", "migration-report.md"))
+      assert_includes report, "docs/decisions/0001-full.md"
+      assert_includes report, "| low |"
+      assert_includes report, "| high |"
+      assert_includes report, "flag rate: 50.0%"
+      assert_equal 0.5, manifest["flag_rate"]
+    end
+  end
+
+  def test_unrecognized_content_is_reported_never_converted
+    project("notes.md" => "# Notes\n\nchatter\n") do |root, repo, migrator|
+      manifest = migrator.migrate!
+      assert_empty repo.entries
+      assert_equal "docs/decisions/notes.md", manifest["unrecognized"].first["source"]
+      assert_includes File.read(File.join(root, "docs", "adr", "migration-report.md")),
+                      "docs/decisions/notes.md"
+    end
+  end
+
+  def test_migration_is_idempotent_across_reruns
+    project("0001-full.md" => FULL_SOURCE) do |_root, repo, migrator|
+      migrator.migrate!
+      migrator.migrate!
+      assert_equal 1, repo.entries.size
+    end
+  end
+
+  def test_index_is_regenerated_with_migrated_adrs
+    project("0001-full.md" => FULL_SOURCE) do |_root, repo, migrator|
+      migrator.migrate!
+      assert_includes File.read(repo.index_path), "PocketBase backend"
+    end
+  end
+end
diff --git a/plugins/os-adr/tests/record_test.rb b/plugins/os-adr/tests/record_test.rb
new file mode 100644
index 0000000..93e73f9
--- /dev/null
+++ b/plugins/os-adr/tests/record_test.rb
@@ -0,0 +1,49 @@
+require_relative "test_helper"
+
+class RecordTest < Minitest::Test
+  include AdrTestHelpers
+
+  def test_parses_frontmatter_and_sections
+    record = Adr::Record.parse(sample_adr)
+    assert_equal "0001", record.id
+    assert_equal "Accepted", record.status
+    assert_equal Date.new(2026, 7, 3), record.date
+    assert_equal "Use Postgres", record.title
+    assert_equal ["db/", "lib/storage"], record.frontmatter["affected-paths"]
+    assert_equal "We need a database.", record.sections["Context"]
+    assert_equal "SQLite — no concurrent writers.", record.sections["Alternatives rejected"]
+  end
+
+  def test_id_with_leading_zeros_stays_a_string
+    record = Adr::Record.parse(sample_adr(id: "0007"))
+    assert_equal "0007", record.id
+  end
+
+  def test_round_trip_preserves_content
+    record = Adr::Record.parse(sample_adr)
+    reparsed = Adr::Record.parse(record.to_markdown)
+    assert_equal record.frontmatter, reparsed.frontmatter
+    assert_equal record.title, reparsed.title
+    assert_equal record.sections, reparsed.sections
+  end
+
+  def test_supersede_with_sets_both_fields
+    record = Adr::Record.parse(sample_adr)
+    record.supersede_with("0009")
+    assert_equal "0009", record.frontmatter["superseded-by"]
+    assert_equal "Superseded", record.status
+    reparsed = Adr::Record.parse(record.to_markdown)
+    assert_equal "0009", reparsed.frontmatter["superseded-by"]
+    assert_equal "Superseded", reparsed.status
+  end
+
+  def test_raises_on_missing_frontmatter
+    assert_raises(Adr::Record::ParseError) { Adr::Record.parse("# just a heading\n") }
+  end
+
+  def test_preserves_unknown_frontmatter_fields
+    record = Adr::Record.parse(sample_adr(extra_frontmatter: "migration_confidence: low\n"))
+    assert_equal "low", record.frontmatter["migration_confidence"]
+    assert_includes record.to_markdown, "migration_confidence: low"
+  end
+end
diff --git a/plugins/os-adr/tests/repository_test.rb b/plugins/os-adr/tests/repository_test.rb
new file mode 100644
index 0000000..ff0fb9b
--- /dev/null
+++ b/plugins/os-adr/tests/repository_test.rb
@@ -0,0 +1,62 @@
+require_relative "test_helper"
+
+class RepositoryTest < Minitest::Test
+  include AdrTestHelpers
+
+  def test_next_id_on_empty_dir_is_0001
+    with_project do |root|
+      assert_equal "0001", Adr::Repository.new(root: root).next_id
+    end
+  end
+
+  def test_next_id_is_max_plus_one
+    with_project("0001-a.md" => sample_adr(id: "0001"),
+                 "0002-b.md" => sample_adr(id: "0002")) do |root|
+      assert_equal "0003", Adr::Repository.new(root: root).next_id
+    end
+  end
+
+  def test_next_id_ignores_missing_index
+    with_project("0007-x.md" => sample_adr(id: "0007")) do |root|
+      repo = Adr::Repository.new(root: root)
+      refute File.exist?(repo.index_path)
+      assert_equal "0008", repo.next_id
+    end
+  end
+
+  def test_next_id_ignores_stale_index
+    with_project("0007-x.md" => sample_adr(id: "0007")) do |root|
+      repo = Adr::Repository.new(root: root)
+      File.write(repo.index_path, "| 0042 | stale row |")
+      assert_equal "0008", repo.next_id
+    end
+  end
+
+  def test_exists_requires_dir_and_index
+    with_project do |root|
+      repo = Adr::Repository.new(root: root)
+      refute repo.exists?
+      File.write(repo.index_path, "index")
+      assert repo.exists?
+    end
+    Dir.mktmpdir do |root|
+      refute Adr::Repository.new(root: root).exists?
+    end
+  end
+
+  def test_find_by_id
+    with_project("0002-b.md" => sample_adr(id: "0002", title: "B")) do |root|
+      entry = Adr::Repository.new(root: root).find("0002")
+      assert_equal "B", entry.record.title
+      assert entry.path.end_with?("0002-b.md")
+    end
+  end
+
+  def test_non_adr_files_are_ignored
+    with_project("0001-a.md" => sample_adr, "notes.md" => "# notes") do |root|
+      repo = Adr::Repository.new(root: root)
+      assert_equal 1, repo.entries.size
+      assert_equal "0002", repo.next_id
+    end
+  end
+end
diff --git a/plugins/os-adr/tests/test_helper.rb b/plugins/os-adr/tests/test_helper.rb
new file mode 100644
index 0000000..d3af48f
--- /dev/null
+++ b/plugins/os-adr/tests/test_helper.rb
@@ -0,0 +1,52 @@
+require "minitest/autorun"
+require "tmpdir"
+require "fileutils"
+require_relative "../lib/adr"
+
+module AdrTestHelpers
+  PLUGIN_ROOT = File.expand_path("..", __dir__)
+  FIXTURES = File.join(__dir__, "fixtures")
+
+  # Yields a tmp project root containing an initialized docs/adr/ populated
+  # with the given ADR fixture contents (id => full file body).
+  def with_project(adrs = {})
+    Dir.mktmpdir do |root|
+      dir = File.join(root, "docs", "adr")
+      FileUtils.mkdir_p(dir)
+      adrs.each { |filename, body| File.write(File.join(dir, filename), body) }
+      yield root
+    end
+  end
+
+  def sample_adr(id: "0001", title: "Use Postgres", status: "Accepted", extra_frontmatter: "")
+    <<~MD
+      ---
+      id: "#{id}"
+      date: 2026-07-03
+      status: #{status}
+      supersedes:
+      superseded-by:
+      affected-paths: [db/, lib/storage]
+      affected-components: [storage]
+      #{extra_frontmatter}---
+
+      # #{id} — #{title}
+
+      ## Context
+
+      We need a database.
+
+      ## Decision
+
+      Use Postgres.
+
+      ## Consequences
+
+      Ops burden.
+
+      ## Alternatives rejected
+
+      SQLite — no concurrent writers.
+    MD
+  end
+end