#405 - os-sdlc: reviewer packet evidence gap — missing programmer/lint evidence, stale gate-log.txt - jared/cc-os - Forgejo: Beyond coding. We forge. | Jared Swanson

jared commented

2026-08-15 16:17:22 +00:00

Owner

Rescoped 2026-08-15 after investigation (evidence brief: .sdlc/tickets/405/evidence-brief.md).

Dropped from scope — not a bug: the missing programmer/lint evidence during #403's review is explained by usage: #403 never ran through the pipeline (no .sdlc/tickets/403/, no dispatch rows; hand-committed as 3f0a5a7), so those stages produced no evidence. Also: the original "stale gate-log.txt" wording conflated two artifacts — gate-log.txt is written only by the escalation-park flow and is never read by the packet builder; the packet's Gate log section comes from artifact.yaml's gate_log array.

Confirmed weakness (this ticket's scope): Artifact#disclosures and Artifact#gate_log (plugins/os-sdlc/lib/os_sdlc/artifact.rb:114,136) are append-only, per-ticket, with no reset on a new run. bin/os-sdlc review-packet renders the full arrays unconditionally (build_review_packet_sections, bin/os-sdlc ~891-940), so a reused artifact.yaml presents prior-run entries as current evidence.

Tasks

Add run-reset semantics: when a ticket starts a new run (intake, and the Resumer retry/skip ledger-reset path), clear the artifact's disclosures and gate_log arrays so a reused artifact.yaml cannot carry prior-run evidence into a review packet.
Rider: correct plugins/os-sdlc/reference/pipeline-stages.md (~lines 39-59) — it describes a "code-actor hook" that does not exist in hooks/hooks.json; the review packet is built by the manual CLI subcommand bin/os-sdlc review-packet.

Acceptance criteria

A review packet built after a fresh run start shows "(none)" for Disclosures and Gate log when those stages have not run in the current run, even if artifact.yaml previously held entries.
Append behavior within a single run is unchanged (existing tests stay green).
A test covers: artifact with prior-run gate_log/disclosure entries → run reset → packet renders "(none)".
pipeline-stages.md no longer claims a code-actor hook; it names the real CLI path.

In scope

Artifact run-reset semantics, review-packet correctness for reused artifacts, the pipeline-stages.md correction.

Out of scope

gate-log.txt / escalation-park flow (unrelated artifact); #403's own history (closed, out-of-pipeline by design); any hook registration work.

Blocking edges

None — the two tasks are independent.

VERDICT: CREATE — confirmed data-integrity defect (stale prior-run evidence rendered as current in review packets); skeptic hard floor overrides the depth-1 default.

Rescoped 2026-08-15 after investigation (evidence brief: .sdlc/tickets/405/evidence-brief.md). **Dropped from scope — not a bug:** the missing programmer/lint evidence during #403's review is explained by usage: #403 never ran through the pipeline (no .sdlc/tickets/403/, no dispatch rows; hand-committed as 3f0a5a7), so those stages produced no evidence. Also: the original "stale gate-log.txt" wording conflated two artifacts — gate-log.txt is written only by the escalation-park flow and is never read by the packet builder; the packet's Gate log section comes from artifact.yaml's gate_log array. **Confirmed weakness (this ticket's scope):** Artifact#disclosures and Artifact#gate_log (plugins/os-sdlc/lib/os_sdlc/artifact.rb:114,136) are append-only, per-ticket, with no reset on a new run. bin/os-sdlc review-packet renders the full arrays unconditionally (build_review_packet_sections, bin/os-sdlc ~891-940), so a reused artifact.yaml presents prior-run entries as current evidence. ## Tasks - [ ] Add run-reset semantics: when a ticket starts a new run (intake, and the Resumer retry/skip ledger-reset path), clear the artifact's disclosures and gate_log arrays so a reused artifact.yaml cannot carry prior-run evidence into a review packet. - [ ] Rider: correct plugins/os-sdlc/reference/pipeline-stages.md (~lines 39-59) — it describes a "code-actor hook" that does not exist in hooks/hooks.json; the review packet is built by the manual CLI subcommand bin/os-sdlc review-packet. ## Acceptance criteria - [ ] A review packet built after a fresh run start shows "(none)" for Disclosures and Gate log when those stages have not run in the current run, even if artifact.yaml previously held entries. - [ ] Append behavior within a single run is unchanged (existing tests stay green). - [ ] A test covers: artifact with prior-run gate_log/disclosure entries → run reset → packet renders "(none)". - [ ] pipeline-stages.md no longer claims a code-actor hook; it names the real CLI path. ## In scope Artifact run-reset semantics, review-packet correctness for reused artifacts, the pipeline-stages.md correction. ## Out of scope gate-log.txt / escalation-park flow (unrelated artifact); #403's own history (closed, out-of-pipeline by design); any hook registration work. ## Blocking edges None — the two tasks are independent. VERDICT: CREATE — confirmed data-integrity defect (stale prior-run evidence rendered as current in review packets); skeptic hard floor overrides the depth-1 default.

jared added the

plugin/os-sdlc

label

2026-08-15 16:17:22 +00:00

jared referenced this issue

2026-08-15 16:17:29 +00:00

os-sdlc: resume --action retry crashes on agentless states (NOT NULL dispatches.agent) #404

jared added the

ready-for-agent

label

2026-08-15 16:34:49 +00:00

jared referenced this issue

2026-08-15 17:22:50 +00:00

os-backlog: ticket-skeptic gauntlet — verdict agent gating AI follow-up capture and ticket-decompose #407

jared commented

2026-08-15 23:32:02 +00:00

Author

Owner

Work starting via os-sdlc pipeline on branch worktree-issue-405

jared commented

2026-08-15 23:58:04 +00:00

Author

Owner

Resolution

Done: Run-reset semantics implemented: Artifact.reset_run_at is the single owner of locating and resetting a ticket's artifact.yml; called from Engine#intake, Resumer retry!/skip! (never abort!), and WorkspaceTicketSeed's existing-artifact branch, so a reused artifact.yml cannot carry prior-run disclosures/gate_log into a review packet (renderer already emits (none) for empty arrays). pipeline-stages.md corrected: review packet is built by bin/os-sdlc review-packet, not a nonexistent code-actor hook.

Evidence: Commit b33a2c3 on branch worktree-issue-405; new test plugins/os-sdlc/tests/runner/artifact_run_reset_test.rb covers intake/retry/skip reset, abort negative, within-run append regression; suite green (925 runs, 0 failures); changed files rubocop-clean; reviewer round-2 verdict APPROVE (.sdlc/tickets/405/verdict.md) after round-1 REQUEST-CHANGES corrections.

Follow-ups: none — reviewer's non-blocking note (ArtifactFieldSet positional arrays) judged not worth a ticket; gate-log.txt/escalation-park flow explicitly out of scope per the ticket

## Resolution **Done:** Run-reset semantics implemented: Artifact.reset_run_at is the single owner of locating and resetting a ticket's artifact.yml; called from Engine#intake, Resumer retry!/skip! (never abort!), and WorkspaceTicketSeed's existing-artifact branch, so a reused artifact.yml cannot carry prior-run disclosures/gate_log into a review packet (renderer already emits (none) for empty arrays). pipeline-stages.md corrected: review packet is built by bin/os-sdlc review-packet, not a nonexistent code-actor hook. **Evidence:** Commit b33a2c3 on branch worktree-issue-405; new test plugins/os-sdlc/tests/runner/artifact_run_reset_test.rb covers intake/retry/skip reset, abort negative, within-run append regression; suite green (925 runs, 0 failures); changed files rubocop-clean; reviewer round-2 verdict APPROVE (.sdlc/tickets/405/verdict.md) after round-1 REQUEST-CHANGES corrections. **Follow-ups:** none — reviewer's non-blocking note (ArtifactFieldSet positional arrays) judged not worth a ticket; gate-log.txt/escalation-park flow explicitly out of scope per the ticket

jared closed this issue

2026-08-15 23:58:04 +00:00

jared referenced this issue

2026-08-16 00:28:28 +00:00

os-sdlc: BriefAssembler emits empty stub briefs (no ticket text) #365

jared referenced this issue

2026-08-16 00:28:28 +00:00

os-sdlc: resume --action retry crashes when exhausted_from is an automated state #364

jared referenced this issue

2026-08-16 00:28:29 +00:00

Resumer retry on agent-less states should re-run the gate #371

jared referenced this issue