jared/cc-os

Watch

Fork 0

ADR-0060 amendment proposal: description-formula quality + delegation-mechanism statement as Tier-2 catalog candidates #185

New issue

Open

opened 2026-07-31 12:24:27 +00:00 by jared · 2 comments

jared commented

2026-07-31 12:24:27 +00:00

Owner

Human decision gate — do not action autonomously. ADR-0060's catalog is closed by design; per plugins/os-aidd-lint/skills/mine-blindspots/SKILL.md, a new Tier-2 smell "routes to an ADR-0060 amendment proposal, never a routine ticket" and never a direct catalog edit. This issue is the proposal request, not the change.

Consolidated ticket — merges #176 (closed as duplicate 2026-08-13). All distinct Tier-2 charter candidates from both tickets are listed below; none dropped.

Context

/os-aidd-lint:check ran clean on plugins/cc-architect/skills/cc-architect/SKILL.md on 2026-07-31 while a manual Opus review against plugins/cc-architect/references/layout/ found 10 substantive issues (docs/cc-architect-skill-layout-findings.md). Most misses are deterministic and filed as Tier-1 cop tickets. The judgment-dense remainder are Tier-2 charter candidates, anticipated by ADR-0084 (2026-07-30) but never promoted: "Judgment-dense layout claims (context: fork fit, description formula quality) are deliberately excluded from Tier 1; they are Tier 2 charter candidates."

Candidate list

description-formula quality — source: #185 finding 2 + #176 (cc-architect/references/layout/descriptions.md). Definition: description must follow [Action verb] [what]. [Unambiguous trigger].; words naming specific operations must appear in the description itself (before truncation), not only in when_to_use. Evidence: single-file/single-instance (cc-architect SKILL.md), no blind-validation study. Split proposed: mechanical half (starts with action verb + has a second sentence) is Tier-1-able via verb list + sentence-boundary regex; only the qualitative half (is the trigger clause unambiguous / non-redundant with when_to_use) needs a Tier-2 judge. Decide split vs. straight-to-Tier-2, and whether the Tier-1 half alone makes the remaining Tier-2 candidate too thin, before drafting an amendment.
delegation-mechanism unstated — source: #185 finding 4 (plugins/cc-architect/references/layout/skill-frontmatter.md). Definition: skill must state which delegation mechanism owns it (Agent tool vs. context: fork); detection signal is absence-of-required-statement + presence of dispatch language. Distinct from ambiguous-actor (ADR-0063, driver/subagent ambiguity in prose) — that entry did not fire on this instance. Evidence: single-file/single-instance, no validation study.
context: fork narrow-fit — source: #176 (cc-architect/references/layout/). No validation study run yet.
disable-model-invocation decision criteria — source: #176 (cc-architect/references/layout/). No validation study run yet.
injection graceful-degradation prose — source: #176 (cc-architect/references/layout/). No validation study run yet.
workflow model-tier fit — source: #176 (cc-architect/references/layout/). No validation study run yet.

Precedent chain for the amendment

The catalog has been amended three times since ADR-0060 closed it at three entries; it now carries six names (rationale-in-the-imperative-position, misplaced-pre-emption, duplicated-authority, decision-suppressing-status-language, speculative-hardening, ambiguous-actor).

ADR-0062 — CLAUDE.md as a second artifact class; amends ADR-0060.
ADR-0063 — multi-agent actor charter as a supplementary catalog; amends ADR-0060.
ADR-0068 — decision-suppressing-status-language admitted as a fourth main-catalog entry; amends ADR-0060. Closest structural precedent: a single new main-catalog name, justified by evidence.

ADR-0060 established status: provisional for entries admitted on thin evidence — the natural landing spot if any candidate above proceeds on a one-file sample.

What each candidate needs before it can be decided

A validation study in the ADR-0060 shape: blind critique of files unrelated to the origin case, reconciled against the proposed entry, checking both positive instances AND false-positive-free exclusion boundaries. One-file samples are not enough to admit a catalog name (applies to all 6 candidates — none has this yet).
Candidate 1's split decision (Tier-1 mechanical + narrower Tier-2, vs. straight to Tier 2).
Charter text for each admitted entry in the shipped format (definition, exemplar, counter-example, prescribed fix) — see plugins/os-aidd-lint/references/tier2-charter.md.
An ADR-0060 amendment per admitted entry (ADR-0062/0063/0068-style), never a direct catalog edit.

Caveat on the evidence (candidates 1–2)

The subject file (cc-architect/skills/cc-architect/SKILL.md) was edited to satisfy the then-new Tier-1 cops in commit ef9b3ff (2026-07-30), one day before the clean 2026-07-31 scan — a clean result on a file tuned to the linter is weak evidence of health. Corroboration: the shipped charter's speculative-hardening exemplar cites SKILL.md:64, a line since deleted by ef9b3ff. That stale exemplar needs fixing regardless of this proposal's outcome (a wording fix, not a ticket, per the charter's own change control).

Provenance

Gap analysis of the clean 2026-07-31 Tier-1 + Tier-2 run. Ground truth: docs/cc-architect-skill-layout-findings.md findings 2 and 4 (candidates 1–2); cc-architect/references/layout/ (candidates 3–6). Primary evidence: plugins/os-aidd-lint/tmp/aidd-tier2-20260731-32235-uxgqyp/. Governing ADRs: 0060, 0084; precedent chain 0062/0063/0068.

**Human decision gate — do not action autonomously.** ADR-0060's catalog is closed by design; per plugins/os-aidd-lint/skills/mine-blindspots/SKILL.md, a new Tier-2 smell "routes to an ADR-0060 amendment proposal, never a routine ticket" and never a direct catalog edit. This issue is the proposal request, not the change. **Consolidated ticket** — merges #176 (closed as duplicate 2026-08-13). All distinct Tier-2 charter candidates from both tickets are listed below; none dropped. ## Context /os-aidd-lint:check ran clean on plugins/cc-architect/skills/cc-architect/SKILL.md on 2026-07-31 while a manual Opus review against plugins/cc-architect/references/layout/ found 10 substantive issues (docs/cc-architect-skill-layout-findings.md). Most misses are deterministic and filed as Tier-1 cop tickets. The judgment-dense remainder are Tier-2 charter candidates, anticipated by ADR-0084 (2026-07-30) but never promoted: "Judgment-dense layout claims (context: fork fit, description formula quality) are deliberately excluded from Tier 1; they are Tier 2 charter candidates." ## Candidate list 1. **description-formula quality** — source: #185 finding 2 + #176 (cc-architect/references/layout/descriptions.md). Definition: description must follow `[Action verb] [what]. [Unambiguous trigger].`; words naming specific operations must appear in the description itself (before truncation), not only in when_to_use. Evidence: single-file/single-instance (cc-architect SKILL.md), no blind-validation study. **Split proposed**: mechanical half (starts with action verb + has a second sentence) is Tier-1-able via verb list + sentence-boundary regex; only the qualitative half (is the trigger clause unambiguous / non-redundant with when_to_use) needs a Tier-2 judge. Decide split vs. straight-to-Tier-2, and whether the Tier-1 half alone makes the remaining Tier-2 candidate too thin, before drafting an amendment. 2. **delegation-mechanism unstated** — source: #185 finding 4 (plugins/cc-architect/references/layout/skill-frontmatter.md). Definition: skill must state which delegation mechanism owns it (Agent tool vs. `context: fork`); detection signal is absence-of-required-statement + presence of dispatch language. Distinct from ambiguous-actor (ADR-0063, driver/subagent ambiguity in prose) — that entry did not fire on this instance. Evidence: single-file/single-instance, no validation study. 3. **context: fork narrow-fit** — source: #176 (cc-architect/references/layout/). No validation study run yet. 4. **disable-model-invocation decision criteria** — source: #176 (cc-architect/references/layout/). No validation study run yet. 5. **injection graceful-degradation prose** — source: #176 (cc-architect/references/layout/). No validation study run yet. 6. **workflow model-tier fit** — source: #176 (cc-architect/references/layout/). No validation study run yet. ## Precedent chain for the amendment The catalog has been amended three times since ADR-0060 closed it at three entries; it now carries six names (rationale-in-the-imperative-position, misplaced-pre-emption, duplicated-authority, decision-suppressing-status-language, speculative-hardening, ambiguous-actor). - **ADR-0062** — CLAUDE.md as a second artifact class; amends ADR-0060. - **ADR-0063** — multi-agent actor charter as a supplementary catalog; amends ADR-0060. - **ADR-0068** — decision-suppressing-status-language admitted as a fourth main-catalog entry; amends ADR-0060. Closest structural precedent: a single new main-catalog name, justified by evidence. ADR-0060 established `status: provisional` for entries admitted on thin evidence — the natural landing spot if any candidate above proceeds on a one-file sample. ## What each candidate needs before it can be decided - A validation study in the ADR-0060 shape: blind critique of files unrelated to the origin case, reconciled against the proposed entry, checking both positive instances AND false-positive-free exclusion boundaries. One-file samples are not enough to admit a catalog name (applies to all 6 candidates — none has this yet). - Candidate 1's split decision (Tier-1 mechanical + narrower Tier-2, vs. straight to Tier 2). - Charter text for each admitted entry in the shipped format (definition, exemplar, counter-example, prescribed fix) — see plugins/os-aidd-lint/references/tier2-charter.md. - An ADR-0060 amendment per admitted entry (ADR-0062/0063/0068-style), never a direct catalog edit. ## Caveat on the evidence (candidates 1–2) The subject file (cc-architect/skills/cc-architect/SKILL.md) was edited to satisfy the then-new Tier-1 cops in commit ef9b3ff (2026-07-30), one day before the clean 2026-07-31 scan — a clean result on a file tuned to the linter is weak evidence of health. Corroboration: the shipped charter's speculative-hardening exemplar cites SKILL.md:64, a line since deleted by ef9b3ff. That stale exemplar needs fixing regardless of this proposal's outcome (a wording fix, not a ticket, per the charter's own change control). ## Provenance Gap analysis of the clean 2026-07-31 Tier-1 + Tier-2 run. Ground truth: docs/cc-architect-skill-layout-findings.md findings 2 and 4 (candidates 1–2); cc-architect/references/layout/ (candidates 3–6). Primary evidence: plugins/os-aidd-lint/tmp/aidd-tier2-20260731-32235-uxgqyp/. Governing ADRs: 0060, 0084; precedent chain 0062/0063/0068.

jared added the

ready-for-human

label

2026-07-31 12:24:27 +00:00

jared added the

create

enhancement

lint-rule

labels

2026-08-01 16:38:47 +00:00

jared referenced this issue

2026-08-01 16:45:01 +00:00

os-aidd-lint: promote judgment-dense layout-doc claims to Tier 2 charter candidates #176

jared commented

2026-08-01 16:45:02 +00:00

Author

Owner

This was generated by AI during triage.

Proposed: 2 Tier-2 catalog candidates from cc-architect SKILL.md gap analysis — description-formula quality and delegation-mechanism-unstated. Both anticipated by ADR-0084 as Tier-2-bound but never promoted.

Evidence so far: single-file, single-instance each (cc-architect/skills/cc-architect/SKILL.md); no blind-validation study yet; evidence noted as possibly stale (file tuned to linter one day before the clean scan).

Maintainer must decide: whether to split Candidate 1 into a Tier-1 mechanical check (verb + second-sentence regex) plus a narrower Tier-2 judgment call, or take it straight to Tier 2; then whether to commission the validation study before drafting the ADR amendment.

Recommendation: take the split on Candidate 1 (cheaper, reduces catalog growth) and run the validation study for both candidates before drafting the amendment — one-file evidence is not enough per ADR-0060's own bar.

> *This was generated by AI during triage.* Proposed: 2 Tier-2 catalog candidates from cc-architect SKILL.md gap analysis — description-formula quality and delegation-mechanism-unstated. Both anticipated by ADR-0084 as Tier-2-bound but never promoted. Evidence so far: single-file, single-instance each (cc-architect/skills/cc-architect/SKILL.md); no blind-validation study yet; evidence noted as possibly stale (file tuned to linter one day before the clean scan). Maintainer must decide: whether to split Candidate 1 into a Tier-1 mechanical check (verb + second-sentence regex) plus a narrower Tier-2 judgment call, or take it straight to Tier 2; then whether to commission the validation study before drafting the ADR amendment. Recommendation: take the split on Candidate 1 (cheaper, reduces catalog growth) and run the validation study for both candidates before drafting the amendment — one-file evidence is not enough per ADR-0060's own bar.

jared added the

plugin/os-aidd-lint

worklist/lint-rule

labels

2026-08-13 18:13:46 +00:00

jared referenced this issue

2026-08-13 18:34:21 +00:00

os-aidd-lint: promote judgment-dense layout-doc claims to Tier 2 charter candidates #176

jared added the

frozen

label

2026-08-16 14:30:54 +00:00