chain (multi-root): 3 issues — Phase 0.4 diagnosis-detection precision sweep from #53 verify (#114)

[kiki830621]

Refs #59 #64 #65

Summary

Multi-root chain (N=3 roots: #59 #64 #65) — Phase 0.4 diagnosis-detection precision sweep from #53's verify follow-up family. 3 sister fixes touching scripts/check-diagnosis-readiness.sh + skills/idd-all/SKILL.md.

Out of scope this PR: #61 (shell test fixture infra) — diagnosed in same batch as Plan-tier with framework-choice surface (bats-core vs plain shell). Deferred per feedback_lead_minimal discipline. Stays diagnosed for separate iteration.

Cluster overview

#	root_id	Spawn source	Phase	PR commit
#59	59	root	implemented	9d58660
#64	64	root	implemented	e6c41ff
#65	65	root	implemented	e6c41ff

Per-issue details

#59 — idd-all Python detection sync (refactor)

2 substring sites in skills/idd-all/SKILL.md (commit 9d58660) swapped from '## Diagnosis' in c['body'] (substring match) → line-anchored regex re.search(r'(?m)^## Diagnosis', c['body']), matching the check-diagnosis-readiness.sh canonical convention shipped in #53 / PR #58:

• line 450 — Phase 3 complexity readback Python helper
• line 533 — Phase 3b Spectra context capture inline Python

Cited sister sites in #59 issue body:

• idd-list/SKILL.md:115 — narrative prose describing detection strategy (not executable code)
• idd-update/SKILL.md:120 — narrative prose describing scan rules (not executable code)
• idd-close/SKILL.md:416 — uses startswith("## Diagnosis") (jq) which is already line-1-anchored (stronger than substring)

Only idd-all had substring patterns in real executable code. Scope narrowed accordingly.

#64 — CommonMark indent tolerance (enhancement)

scripts/check-diagnosis-readiness.sh regex (commit e6c41ff) widened:

- '[.comments[] | select(.body | test("(?m)^## Diagnosis"))] | length'
+ '[.comments[] | select(.body | test("(?m)^[ ]{0,3}## Diagnosis"))] | length'

CommonMark spec allows 1-3 space leading indent for ATX headings. The strict ^ anchor false-negatived user-pasted comment bodies with legal indent. [ ]{0,3} correctly:

• Matches canonical IDD col-0 comments (col-0 = [ ]{0})
• Matches 1-3 space indented ATX headings (CommonMark-compliant)
• Excludes 4+ space indent (per spec that's a code block, not a heading)
• Excludes tab indent U+0009 (not a valid ATX heading indent)

#65 — Fenced-code limitation doc (documentation, Approach A)

NEW comment block in scripts/check-diagnosis-readiness.sh (commit e6c41ff) documenting that detection is line-based and doesn't track fenced code block state. A comment body that quotes ## Diagnosis inside a fenced code block ( ```) for documentation purposes will false-positive as 'diagnosed'.

Approach A (docs-acknowledge) selected per diagnosis decision point + feedback_lead_minimal. Mitigation: chain Phase 0.4 AskUserQuestion lets the user override the auto-detect verdict (run /idd-diagnose first / proceed anyway / cancel). Simpler heuristic + user-override safety net preferred over Approach B (full markdown state-machine parser) — the parser itself would contain its own edge-case bugs in fence nesting / language hints / indentation rules.

Review status

• Diagnose ✓ for all 3 issues (per-issue Diagnosis comments)
• Implement ✓ (3 commits + CHANGELOG fixup)
• Verify ⏳ (per-issue 6-AI ensemble — running)
• Verify-gated: per-issue verify PASS — cluster ready to merge → /idd-close #N per issue after merge

---

🤖 Generated by /idd-all-chain. Do NOT add GitHub close trailers (Closes/Fixes/Resolves) — IDD discipline requires manual /idd-close per issue after merge to enforce checklist gate + per-issue closing summary.

[kiki830621]

/idd-verify --pr 114 — cluster verify report

Phase: verified (verify-gated PASS)
Mode: PR mode, cluster (3 issues: #59 #64 #65)
Diff: 108 lines, 4 files (scripts/check-diagnosis-readiness.sh + skills/idd-all/SKILL.md + CHANGELOG.md + plugin.json)
Reviewers: 3 general-purpose Agents (requirements / logic / devil's advocate) + Codex gpt-5.5 xhigh (background)

Why lean 3-AI ensemble: 108-line diff, doc-only precision tweaks + 4-line regex change. Security + regression depth not warranted (no new code paths, no new attack surface, no cross-skill interactions worth dedicated reviewer). 3-AI ensemble + Codex preserves cross-model independent check.

Aggregate verdict: PASS

Reviewer	Verdict
Requirements	PASS — all 10 deliverables correctly addressed
Logic	PASS — Python import re confirmed in scope; jq Oniguruma supports {0,3}; col-0 still matches; tab + 4-space stay excluded
Devil's Advocate	PASS — empirical 31-fixture battery across both regex engines confirms Unicode whitespace exclusion; idd-close startswith() verified line-1-anchored; no cross-issue regression between #64 widening + #65 fenced-code limitation
Codex (gpt-5.5 xhigh)	Process gap — background task at 0 bytes when this report compiled. Recorded per v2.59.0+ convention (precedent: PR #110, PR #113)

Findings

0 blocking findings. 3 non-blocking observations (all FYI, not blocking merge):

1. (requirements) Asymmetric regex precision between Python (^## Diagnosis) vs jq (^[ ]{0,3}## Diagnosis) — both correct for canonical IDD shape; future cross-reader might find surprising. Optional follow-up: one-liner clarifying that Python sites scan AI-generated content (always col-0) vs helper script scans potentially user-pasted content (legal CommonMark indent)
2. (DA F1) Project rule .claude/rules/attribute-assessment.md about Likert scoring applies to AI-generated assessment, not hand-written P3 labels — no violation in this PR's ### Vagueness Pre-check per-issue audit blocks
3. (DA F2/F3) CR-only line delimiter would false-negative in both regex engines, but GitHub API normalizes to LF — operationally impossible edge case

Per-issue verdict

#59 — idd-all Python substring → regex: PASS

• skills/idd-all/SKILL.md:450 substring → re.search(r'(?m)^## Diagnosis', c['body']) ✓
• skills/idd-all/SKILL.md:533 same change ✓
• import re in scope at both edit sites (line 450 inline python3 -c block already imports re; line 533 short script needs re — verified added)
• Scope discipline confirmed: cited idd-list:115 / idd-update:120 sites are narrative prose, not executable code; idd-close:416 uses startswith() (DA empirically confirmed line-1-anchored)

#64 — CommonMark indent tolerance: PASS

• scripts/check-diagnosis-readiness.sh:65 regex widened to (?m)^[ ]{0,3}## Diagnosis ✓
• jq Oniguruma engine supports {0,3} quantifier (verified empirically by DA's 6-fixture jq battery)
• Edge cases verified:
  • col-0 ([ ]{0}) still matches ✓
  • 1-3 space indent ([ ]{1,3}) now matches (CommonMark-compliant) ✓
  • 4+ space indent (code block per spec) stays excluded ✓
  • Tab character (U+0009) stays excluded — [ ] is literal ASCII space ✓
  • Unicode whitespace (NBSP / U+1680 / U+2000-200A / U+202F / U+205F / U+3000) stays excluded — [ ] is byte-level ASCII ✓ (DA empirical)

#65 — Fenced-code limitation doc (Approach A): PASS

• NEW comment block at scripts/check-diagnosis-readiness.sh:21-30 documenting line-based detection's fenced-code false-positive
• Comment placement clean (doesn't break bash syntax)
• Mitigation cross-reference to chain Phase 0.4 AskUserQuestion (DA verified location at idd-all-chain/SKILL.md:201,207-209)
• Trade-off rationale documented: simpler heuristic + user-override safety net > full state-machine parser

Shared

• plugin.json 2.67.0 → 2.68.0 (minor — justified by additive precision tweaks)
• CHANGELOG.md [2.68.0] entry covers all 3 issues + explicit [test] Add shell test fixture infra for diagnosis detection logic (verify follow-up from #53) #61 deferral note

Process gap

Codex gpt-5.5 xhigh background task at 0 bytes output when this report compiled (~7 min after spawn). 3-AI Claude ensemble carried successfully; cross-model independent verification not achieved this round. Recorded transparently per v2.59.0+ convention. This is the 3rd consecutive cluster verify in this session where Codex hung at 12+ min — pattern recorded for potential investigation. Not hidden in aggregate.

Next: merge ready

Cluster ready to merge:

1. Merge chain (multi-root): 3 issues — Phase 0.4 diagnosis-detection precision sweep from #53 verify (#114) #114 (squash recommended — single review surface)
2. /idd-close #59 #64 #65 per issue (per-issue closing summary required)
3. Distribution sync v2.67.0 → v2.68.0 via /plugin-tools:plugin-update issue-driven-dev

---

Verify: 3-AI ensemble PASS (Codex process gap recorded). 0 blocking findings, 3 non-blocking observations all FYI.