R76.H2 Audit — colibri-verification SKILL.md donor-ref inventory

Task: R76.H2 — rewrite .agents/skills/colibri-verification/SKILL.md to eliminate donor-era tool references outside heritage blocks.

Base commit: d5f6a1ff (main, Phase 0 SEALED at 28/28).

Scope: .agents/skills/colibri-verification/SKILL.md (297 lines, canon) + .claude/skills/colibri-verification/SKILL.md (mirror, currently stale per CLAUDE.md §9.2 drift backlog).

Method

Ran rg -n against the canonical SKILL.md for every donor-era pattern the task prompt requires striking, plus the broader donor memory/context/gsd/roadmap families memory lists as carry-over.

Canon vs. mirror divergence (pre-rewrite)

diff .agents/skills/colibri-verification/SKILL.md .claude/skills/colibri-verification/SKILL.md reports:

1. Frontmatter description: is bare in mirror; quoted in canon.
2. Mirror is missing the Phase 0 reality stamp (> blockquote) at line 9.
3. Mirror’s PLAN table (line 178-182 in canon, 177-181 in mirror) still lists task_create_batch, task_link_roadmap, task_deps, gsd_workflow_create, gsd_agent_pool_create as clean live tools (no heritage note).
4. Mirror’s CLOSE table (line 223-227 in canon, 222-226 in mirror) still lists memory_pack and audit_session_end as clean live tools.

Net: mirror is the less-reality-stamped ancestor that canon partially attempted to annotate in Wave H. Wave H did not cascade the fix to mirror. R76.H2 will rewrite canon clean and resync mirror from canon.

Donor-ref hit list (canonical file only)

All hits are outside any > HERITAGE: block. Line numbers reference the current canon.

Line	Family	Donor token(s)	Substitute
9	reality stamp	Lists all donor names under back-ticks as part of the Wave H explanatory stamp	Keep (reality stamp) — replace with a cleaner R76.H2 stamp that lists the 14 shipped tools positively instead of listing the donor tokens
35	roadmap_*	roadmap_list	Drop bullet; replace with “Target task identified in docs/guides/implementation/task-breakdown.md”
36	roadmap_*	roadmap_validate	Drop bullet; replace with a docs-check reference to colibri-docs-check skill
37	roadmap_*	roadmap_nodes	Drop bullet; node enumeration lives in task-breakdown.md
38	roadmap_*	roadmap_get	Drop bullet; content lookup is a file read
44	unified_*	unified_set_project	Drop — no project-switching surface in Phase 0 (single project colibri)
45	task edge	task_deps	Rewrite as task_get / task_list (check status and description manually; no dep tool in Phase 0)
52	unified_*	unified_init	Drop — server is auto-initialized on MCP handshake
53	unified_*	unified_vitals	Replace with server_health
59	audit_session_set_context	audit_session_set_context	Replace with note: context arg is merged into audit_session_start
84	context_*	context_create	Drop — no context surface in Phase 0; use thought_record to snapshot reasoning at checkpoints
89	unified_* + context_*	unified_backup / context_create	Drop — no backup surface in Phase 0; rollback is git reset + worktree discard
121	memory_pack	memory_pack	Drop — no memory surface in Phase 0; reasoning is anchored in thought_record + Merkle root
129	merkle_attest	merkle_attest	Replace with merkle_root (Phase 0 attestation = root hash)
147	roadmap_*	roadmap_list	Drop row; roadmap lives in docs/guides/implementation/task-breakdown.md
148	roadmap_*	roadmap_nodes	Drop row (same reason)
150	context_*	context_create	Replace with thought_record (Phase 0 reasoning capture)
162	task edge	task_eisenhower	Drop — no prioritizer tool in Phase 0; priority is a field on task_create
163	task edge	task_deps	Drop — deps are not a Phase 0 MCP surface; noted via task description
164	roadmap_*	roadmap_prerequisites_check	Drop — prerequisites live in task-breakdown.md narrative
165	analysis_*	analysis_recommend	Drop — no analysis surface in Phase 0
166	thought edge	thought_plan	Replace with thought_record (single Phase 0 reasoning-capture tool)
167	thought edge	thought_decide	Replace with thought_record
168	gsd_*	gsd_plan	Drop — gsd surface deferred to Phase 1.5 per ADR-005; plan is the 5-step chain packet
179	task_link_roadmap	task_link_roadmap	Drop — row already noted Phase 0 absence; rewrite row to reference task_create description field only
180	task_deps / task_depends_on	task_deps / task_depends_on	Drop — row rewrite for “dependency narrative lives in description” only
181	gsd_workflow_create	gsd_workflow_create	Drop row — Phase 1.5 defer; not in Phase 0 exit gate for PLAN
182	gsd_agent_pool_create	gsd_agent_pool_create	Drop row — Phase 1.5 defer per ADR-005
196	context_*	context_create	Replace with thought_record
208	roadmap_*	roadmap_validate	Drop row — colibri-docs-check skill handles doc-integrity
209	gsd_*	gsd_verify_comprehensive	Drop row — no verification omnibus tool in Phase 0; npm run build && npm run lint && npm test is the gate
211	context_*	context_verify	Drop row — no context surface
227	audit_session_end	Inline mention “no separate audit_session_end”	Keep cell narrative but adjust: “Session seals when merkle_finalize commits the final root” (wording-only polish)
243	context_*	context_verify	Remove from JS chain
246	roadmap_*	roadmap_validate	Remove from JS chain
249-255	gsd_*	gsd_verify_comprehensive	Remove from JS chain — replace with audit_verify_chain + thought_record + merkle_finalize + merkle_root sequence per CLAUDE.md §7
270	roadmap_*	roadmap_repair	Drop row — nothing to repair outside docs; handled by colibri-docs-check
273	context_*	context_create	Drop row — no context surface

Total donor-ref hits to rewrite: 36 discrete mentions across 33 lines.

Quick-reference table audit (line 15-25)

The Quick Reference table at line 15 is pure phase-name prose — no tool names — so it requires no rewrite.

Common-failures table audit (line 268-275)

Three rows cite donor tools (roadmap_repair, context_create, memory_pack). All three rows lose their donor cells per the rewrite plan. Two other rows (Task not found, Audit chain broken) use only Phase 0 shipped tools — task_list, audit_session_start — and stay intact. The Merkle finalize fails row stays. The Memory pack empty row is struck entirely.

JS example audit (line 236-262)

The verifyCompletion chain currently calls:

1. audit_verify_chain ✓ (Phase 0 shipped)
2. context_verify ✗ (strike)
3. roadmap_validate ✗ (strike)
4. gsd_verify_comprehensive ✗ (strike)
5. merkle_finalize ✓ (Phase 0 shipped)

Rewrite target (from task prompt §Canonical verification chain):

audit_session_start(...)
  ... work ...
audit_verify_chain(...)
thought_record(...)       ← MUST precede merkle_finalize
merkle_finalize(...)
merkle_root(...)

The rewritten JS block will implement exactly that 6-call chain.

No heritage blocks currently in the file

The file contains zero > HERITAGE: blockquote markers. All donor refs are on the live path and all need rewriting. The existing “Phase 0 reality stamp” line-9 blockquote is not a heritage block (the rest of the file is still live process guidance); it is a reality stamp that R76.H2 will replace with a cleaner affirmative stamp.

Out-of-scope confirmations

• No other skills touched.
• No src/ changes.
• No frontmatter redesign beyond the description cleanup forced by YAML quoting safety.
• Skill description keeps the Greek labels η + π (correct — verification sits on the η proof store + π verifier axes).

Exit gate

This audit establishes:

• 36 donor-ref hits across 33 lines
• Zero heritage blocks present
• JS example chain fully rewritable to the CLAUDE.md §7 6-call sequence
• Mirror is stale; resync from canon at Step 4

Ready for Step 2 (Contract).