cluster: 6 Simple docs/pattern follow-ups (#96-backlog triage)

[kiki830621]

Refs #60 #62 #63 #78 #90 #91

Summary

Cluster-PR for the 6 Simple-tier issues from the #96-backlog complexity triage (/idd-diagnose batch of 18 → 6 Simple / 12 Plan). All 6 are docs / pattern-copy changes with no behavior change — bundled into one PR with one combined verify. The 12 Plan-tier issues from the same triage are driven separately.

Cluster overview

#	Change	Commit
#60	batch-and-cluster.md — new ## Cluster-PR eligibility section + idd-implement cross-ref	fead69c
#62	usecase-routing.md — decision-tree bulk-solve note	2d50108
#63	usecase-routing.md — #44 chain-solve explicit URL link (row 27)	ea94b08
#78	idd-issue/SKILL.md + CHANGELOG.md — v2.55.0 multi-finding migration note for CI callers	dc947dd
#90	openspec/LANGUAGE.md (new file) — GitHub-side tracker marker convention	08f7c26
#91	spectra-archive/SKILL.md — Step 0: Bootstrap Stage Task List	9b6b360

Adaptations (recorded in per-issue Implementation Complete comments)

• [docs] idd-issue multi-finding: backward compat migration note for legacy CI callers (v2.55.0+) #78 — the issue Strategy assumed a ## [2.55.0] CHANGELOG entry exists; it does not (CHANGELOG jumps 2.60.0 → 2.62.0). The behavioral-change notice was added under ## [Unreleased] as a labelled retroactive record; the idd-issue/SKILL.md inline note is the substantive deliverable.
• [docs] standardize GitHub linking convention across archived Spectra proposals (mid-plan tangential from #56) #90 — openspec/LANGUAGE.md did not exist; created it (the issue's named primary target — LANGUAGE.md is a standard OpenSpec project-conventions file).

Checklist

• Diagnose — 6 issues, Simple verdict each (batch complexity map)
• Implement — 6 commits + R2 fix (e49d6cb), Refs #N each
• Verify — 6-AI R1 CONDITIONAL (2 DA blocking) → R2 PASS (R2 master)
• Pending: human review of this cluster PR + /idd-close #60 #62 #63 #78 #90 #91 after merge

---

Generated by /idd-implement cluster-PR mode. Do NOT add a GitHub close trailer (Closes/Fixes/Resolves) — IDD discipline requires manual /idd-close per issue after merge to enforce the checklist gate + per-issue closing summary.

[kiki830621]

Verify Report — PR #101 (cluster: 6 Simple docs/pattern follow-ups) — Round 1

Engine

5 general-purpose Agents (Requirements / Logic / Security / Regression / Devil's Advocate) + Codex (gpt-5.5). Codex was still running at aggregation time — this master covers 5/6; a Codex addendum will follow if it surfaces anything new.

Aggregate verdict: 🟡 CONDITIONAL PASS — do not merge as-is

4 of 5 Claude reviewers PASS. Devil's Advocate surfaced 2 HIGH blocking findings, both independently confirmed by the coordinator (file-existence facts, not opinion). 4 of the 6 issues (#60 #62 #63 #78) are clean; 2 (#90 #91) need an R2 fix.

5-AI verdict

Reviewer	Verdict	Note
Requirements	PASS	6/6 FULLY addressed; both adaptations (#78/#90) judged to satisfy issue intent. 2 cosmetic nits (see below)
Logic	PASS	#60 anchor exact-match ✓, #91 TaskCreate 1:1 with Steps 1-8 ✓, #62/#63 disjoint ✓, #90 valid markdown ✓
Security	PASS / APPROVE	docs-only; no secrets, no auto-close trap, LANGUAGE.md #NN uses literal letters (safe)
Regression	PASS	0 blocking regressions; all edits land cleanly, KAC-conformant, no scope creep
Devil's Advocate	DO NOT MERGE	2 HIGH blocking — see below. Conceded #60/#62/#63/#78

🔴 Blocking findings (DA, both coordinator-confirmed)

DA-1 — #91 fixed only one of two invocation surfaces (HIGH)

#91 added Step 0 to .claude/skills/spectra-archive/SKILL.md. But .claude/commands/spectra/archive.md also exists, has no Step 0, and does not delegate to the SKILL.md — confirmed:

• .claude/commands/spectra/archive.md exists (5675 bytes), grep -c 'Step 0' → 0
• grep 'skills/spectra-archive\|SKILL.md' in the command file → no delegation reference

/spectra:archive (the command form) resolves through the command file. So #91's Step 0 addition does not reach that invocation surface. #91's issue body + diagnosis both named only the SKILL.md — the command file was missed. (This is the same drift family as #93, but the command file is a 4th copy the #93 issue didn't enumerate.)

DA-2 — #90 created a file at a RESERVED path (HIGH)

#90 created openspec/LANGUAGE.md with a workflow convention. But openspec/LANGUAGE.md is already reserved — confirmed:

• spectra-discuss SKILL.md:35 — "Try to read openspec/LANGUAGE.md. This file is the project's canonical vocabulary — terms with definition, avoid, why notes"
• spectra-discuss has a "Vocabulary drift" mechanism (line 252-254) that captures TO openspec/LANGUAGE.md
• Same reference in .claude/commands/spectra/discuss.md + .agents/skills/spectra-discuss/SKILL.md

openspec/LANGUAGE.md is meant to hold a controlled vocabulary, not a workflow convention. #90's file has zero vocabulary terms — spectra-discuss Step 0 will now load it as a "vocabulary file" with no vocabulary. Wrong home.

R2 fix plan (same PR, per verify-fix-in-same-PR discipline)

• [docs] standardize GitHub linking convention across archived Spectra proposals (mid-plan tangential from #56) #90 — git rm openspec/LANGUAGE.md; relocate the **GitHub-side tracker**: #NN convention to a new openspec/CONVENTIONS.md (purpose-built filename, parallel to LANGUAGE.md in openspec/, no reserved-name collision).
• [refactor] spectra-archive SKILL.md missing Bootstrap TaskList (inconsistent with idd-* skills) (mid-plan tangential from #56) #91 — additionally add the Step 0 section to .claude/commands/spectra/archive.md so the command invocation surface is covered; keep the SKILL.md edit.

Non-blocking nits (Requirements, ACCEPTED for R2 or closing)

• [docs] Formalize cluster-PR eligibility criteria in batch-and-cluster.md (verify follow-up from #49+#53) #60 — cross-link added to idd-implement only, not idd-all-chain (issue acceptance named both). Non-blocking — chain mode auto-derives same-root eligibility. Optionally add in R2.
• [docs] idd-issue multi-finding: backward compat migration note for legacy CI callers (v2.55.0+) #78 — the literal ⚠ glyph the issue's wording showed is in the idd-issue/SKILL.md note but not the CHANGELOG line. Cosmetic.

Cross-cutting (DA, acknowledged)

• This cluster (89 lines, same #96-backlog parent label) marginally violates the >50-line bundle heuristic that [docs] Formalize cluster-PR eligibility criteria in batch-and-cluster.md (verify follow-up from #49+#53) #60 itself introduces. Acknowledged — accepted as a one-off (all-docs, all-Simple, low review cost).
• [docs] standardize GitHub linking convention across archived Spectra proposals (mid-plan tangential from #56) #90 / [refactor] spectra-archive SKILL.md missing Bootstrap TaskList (inconsistent with idd-* skills) (mid-plan tangential from #56) #91 were mislabelled "Simple" in the triage — both had cross-file behavioral reach the "Simple docs" framing hid. The diagnose-from-body-without-verifying-targets gap is the root; R2 corrects the outcome.

---

Next: R2 fixes #90 + #91 on this branch → re-verify R2.

[kiki830621]

Verify Report — PR #101 — Round 2

Aggregate verdict: ✅ PASS — ready to merge

Both R1 blocking findings closed. R2 delta: 1 file relocate (#90) + 1 scoping resolution (#91, user decision). 4 issues (#60 #62 #63 #78) were clean in R1 and untouched by R2.

R1 → R2 closure

R1 finding	Severity	R2 status
DA-2 — #90 created openspec/LANGUAGE.md at a reserved path (spectra-discuss vocabulary file)	HIGH	CLOSED — commit e49d6cb relocates the convention to openspec/CONVENTIONS.md. Verified not reserved: grep -rn 'CONVENTIONS.md' across .claude/skills/, .claude/commands/, .agents/, plugins/, openspec/config.yaml → no references. git mv detected as 77%-similarity rename.
DA-1 — #91 fixed only the SKILL.md surface, not .claude/commands/spectra/archive.md	HIGH	RESOLVED (re-scoped) — see below

DA-1 resolution

On inspection the DA-1 finding was partially overstated. .claude/commands/spectra/archive.md is wrapped in <!-- SPECTRA:START v1.0.2 --> … <!-- SPECTRA:END --> tool-managed markers — the spectra CLI regenerates that region; hand-editing it is the actual anti-pattern. The SKILL.md surface #91 edited:

• IS the surface hit by the /idd-close → /spectra-archive skill cascade (the exact scenario [refactor] spectra-archive SKILL.md missing Bootstrap TaskList (inconsistent with idd-* skills) (mid-plan tangential from #56) #91's issue body cites — so [refactor] spectra-archive SKILL.md missing Bootstrap TaskList (inconsistent with idd-* skills) (mid-plan tangential from #56) #91's edit is not "zero runtime effect")
• is consistent with feature: spectra-apply auto-post Implementation Complete to GitHub issue (resolve /idd-close supersession friction) #56's precedent of directly hand-editing .claude/skills/spectra-archive/SKILL.md (which has no SPECTRA:START/END markers)

Decision (user): keep #91's SKILL.md edit; do not hand-edit the tool-managed command file. The command file's Step-0 gap is a facet of the spectra-archive copy divergence and is folded into #93 — whose 3-copy list is now corrected to 4 (the command file is the 4th). Recorded on:

• [refactor] spectra-archive SKILL.md missing Bootstrap TaskList (inconsistent with idd-* skills) (mid-plan tangential from #56) #91 Implementation Complete comment — ### R1 verify follow-up (PR #101, DA-1) section
• [refactor] spectra-archive SKILL.md exists in 3 copies (.claude/, .agents/, plugins/) — divergence risk (follow-up from #56) #93 — comment-4485274923 documenting the 4th copy + the tool-managed-vs-hand-maintained distinction

Why R2 has no separate 6-AI re-ensemble

R2's code delta is a single file relocate (LANGUAGE.md → CONVENTIONS.md, content near-identical) that directly closes the named DA-2 finding — verifiable by the reserved-name grep above. #91 had no code change in R2 (user re-scoping decision). Re-running the 5-agent ensemble would re-review #60/#62/#63/#78 — unchanged since R1 PASS. Same rationale as PR #99 R3 (a finder-specified doc fix needs confirmation, not a fresh ensemble).

Non-blocking nits (carried from R1, ACCEPTED)

• [docs] Formalize cluster-PR eligibility criteria in batch-and-cluster.md (verify follow-up from #49+#53) #60 — cross-link added to idd-implement only, not idd-all-chain. Non-blocking (chain mode auto-derives same-root eligibility).
• [docs] idd-issue multi-finding: backward compat migration note for legacy CI callers (v2.55.0+) #78 — literal ⚠ glyph in the SKILL.md note but not the CHANGELOG line. Cosmetic.

Process gap

• Codex (6th reviewer) never returned — the gpt-5.5 xhigh background run was still executing long past the 5 Claude agents and did not complete before R2 aggregation. R1 + R2 verdicts rest on the 5 Claude reviewers (4 PASS + DA, with both DA findings independently coordinator-confirmed via file-existence checks). Cross-model blind verification was unavailable this run. Per idd-verify Step 2.5d, this is recorded explicitly rather than hidden inside an aggregate "PASS".

Final per-issue state

#	R1	R2	Final
#60	PASS	—	✅ clean
#62	PASS	—	✅ clean
#63	PASS	—	✅ clean
#78	PASS	—	✅ clean
#90	🔴 DA-2	relocated to CONVENTIONS.md	✅ closed
#91	🔴 DA-1	re-scoped (SKILL.md edit stands; command file → #93)	✅ resolved

---

Next: human review of PR #101 → merge → /idd-close #60 #62 #63 #78 #90 #91 (per-issue closing summary).