mr-wolv/skill_playground_bakr101_2026/catalog-consistency-auditor
Claude Code and agent skill by mr-wolv/skill_playground_bakr101_2026
- community
- ext:[{"type":"hook","description":"Pre/post lifecycle hooks"}]
Checking that the SKILL.md file is still in the repository.
Audit skill catalogs and repository documentation for drift against the actual skill set. Use when counts look stale, indexes disagree, mirrored stores may be out of sync, or catalog artifacts need one-pass reconciliation.
Install
npx vskill@latest install mr-wolv/skill_playground_bakr101_2026/catalog-consistency-auditorSemi-extensibleMentions customization, but needs skill-specific knowledge. Read the extensible skills standard
SKILL.md
The same SKILL.md was published earlier as mr-wolv/skill_playground_bakr101_2026/catalog-consistency-auditor.
Catalog Consistency Auditor
Audit a skill repository for structural drift between the actual skill folders and the documents that describe them.
This skill is for catalog integrity, not general code quality. Use it when the problem is that repository artifacts, counts, indexes, or mirrored skill stores may no longer agree.
When to use
- A README, catalog, or manifest mentions counts that may be stale
- The workspace
skills/directory and a mirrored global skill store may have diverged - A skill was added, removed, renamed, merged, or split and the surrounding docs may be inconsistent
- You want a one-pass reconciliation of catalog artifacts after changing the skill set
- You suspect documentation drift across skill indexes, manifests, and navigation docs
Scope discipline (HARD RULE)
When the user asks you to "fix everything" / "close all findings" / "make it converge", treat every defect as in-scope — including ones that pre-existed this session or live outside the files you originally touched. "It was already broken" is NOT an exemption and must never be offered as a reason to leave a defect standing. If a convergence/completeness audit surfaces a gap, fix the gap. Pre-existing state is just more of the audit's scope, not a permission to stop. The deliverable is a repo where the stated completion claim actually holds, not one where known issues were rationalized away.
Boundary
Use this skill for repository catalog coherence: source-of-truth discovery, count verification, mirror parity checks, and reconciliation of catalog-bearing documents and manifests.
Do not use this skill for historical reconstruction of why a repository evolved the way it did. For that, use repository-archaeology.
Do not use this skill for general codebase health, tech debt, dead code, or dependency audits. Use the repository-intelligence specialist skill that matches that problem.
What this skill owns
This skill focuses on:
- structural source of truth discovery
- count verification
- mirror parity checks
- documentation drift detection
- index and manifest reconciliation
- explicit reporting of unresolved ambiguities
This skill does not decide whether a new capability should exist. Use capability-architecture or planning skills for that question.
Workflow
1. Establish the source of truth
Treat the actual skill folders as the structural source of truth.
At minimum, verify:
- repository skill directory set
- mirrored/global skill directory set, if applicable
If documentation disagrees with the folder set, the folder set wins.
2. Compute the verified state
Record the facts before editing docs:
- total skill directory count
- repo-only entries
- global-only entries
- exact parity status
- any renamed/suspicious near-duplicates discovered during review
Do not copy prior totals from docs.
3. Audit catalog-bearing artifacts
Check all repository artifacts that can drift from the source of truth, including:
README.md- root
SKILL.md SKILL-CATALOG.mdSKILL-CATALOG-DOMAIN.mdSDLC-PHRASE-CHEATSHEET.mdskills.jsondocs/index.md- governance or maintenance docs
AGENTS.mdif it references catalog behavior or sync rules
Look for:
- stale counts
- conflicting totals
- outdated category claims
- references to missing skills
- missing references to newly added skills
- inaccurate sync statements
- obsolete maintenance instructions
4. Reconcile in one pass
When you fix drift, update all affected artifacts in the same change.
Prefer:
- verified counts over inherited counts
- operational language over marketing language
- explicit uncertainty where classification is ambiguous
- narrow factual claims that are easy to keep true
If an exact breakdown cannot be verified confidently, avoid inventing one. State the verified fact and note what remains unclassified.
5. Report unresolved issues explicitly
If you encounter ambiguity, record it clearly instead of guessing. Examples:
- a skill appears workflow-oriented and category placement is unclear
- the repo and global stores differ and intentionality is unknown
- legacy docs use a classification scheme that no longer maps cleanly to the current catalog
6. Automate derived counts — never hand-edit the sweep (USER PREFERENCE)
Hard rule from the user: "automate everything of a hardcoded kind; deep automation for anything of such nuisance is needed so I'm not actively pained by it." When a catalog carries counts in many docs (README.md, SKILL.md, SKILL-CATALOG.md, SKILL-CATALOG-DOMAIN.md, docs/index.md, docs/catalog-governance.md, skills.json), hand-editing all of them on every skill add/remove is the brittle failure mode. The resilient pattern is to derive the counts from the filesystem and regenerate:
- total = len([p for p in skills_dir.glob('*/') if (p/'SKILL.md').exists()])
- custom = union of all skills.json categories arrays
- community = total - custom (community_skill_names enumerates them exactly)
A small generator (e.g. scripts/refresh_derived_catalog.py with --check/--apply) rewrites the manifests + prose in one idempotent pass. Wire it into the PRE-COMMIT HOOK so a diverged count self-heals and stages before the gate — the commit can never be blocked (or silently wrong) by a stale literal. Freeze-tests that assert counts must DERIVE expected values from skills.json at runtime, never hardcode 240/241/64/65.
Exclude historical report files (VERSION-1.0-*.md, AUDIT-COMPLETENESS.md) from regeneration — they record point-in-time numbers and must not be rewritten. Scripted recipe + the exact file map: references/automate-derived-counts.md.
PITFALL — sync_global_to_repo.py --force is a SILENT NO-OP without --apply. The apply block is nested inside if args.apply, so --force alone prints a plan and writes nothing (you burn a run). Correct convergence invocation under merged (B==C) topology: python scripts/sync_global_to_repo.py --apply --force (takes a timestamped backup at skills.backup-<ts> first). It only overwrites SKILL DIRS whose content differs — root docs (README.md/SKILL.md) are NOT touched, so count governance stays with refresh_derived_catalog.py. Because the store IS the truth post-merge, --apply --force is the SAFE way to converge a lagging export to the truth; do NOT hand-re-edit drifted literals after a sync, re-run refresh --apply. The older "restore from git, the store may be stale" warning was written pre-merge when B≠C — in merged topology the store cannot be stale relative to itself.
Convergence audit (the "many truths → one truth" pass)
When the task is not a one-off reconciliation but a convergence audit — declare a catalog "complete", "coherent", or "Version 1.0" — produce a report, not just a fix-list. The deliverable is a findings table where every finding carries impact / recommended action / priority / architectural justification, plus a go/no-go verdict on the completion claim. Framing: the goal is to find inconsistencies, ambiguity, duplication, drift, architectural debt, simplification opportunities — not to add capabilities.
Run the repo's own QC gate first and report its result as ground truth. A green gate.py / make verify is strong evidence the mechanical foundation is converged even when prose docs contradict each other. Convergence findings are almost always in the documents, not the validator.
reusable techniques that caught real drift (cheap to re-run; see references/convergence-audit-recipe.md for the scripted versions and references/merge-b-c-one-store.md for the root-cause B<->C oscillation fix). Recovery recipe when an in-place mutation test corrupts the working tree: references/test-suite-hygiene.md. The *.backup-* junk the syncs leave and the background-commit + gh run watch CI tail of a finalization pass: references/finalization-operational-pitfalls.md.
- Source-of-truth contradiction check. Grep every governance/doc for "source of
truth" and confirm they agree on WHICH directory is authoritative. This session found docs/catalog-governance.md calling skills/ the source of truth while AGENTS.md/README.md/docs/index.md called ~/.agents/skills the source of truth — a direct "many similar truths" violation and the single highest-value finding. Check it first.
- Prose-count lint. Docs hard-code totals that drift from the folder set (a catalog
summary said "62 custom / 176 community = 238" while the validator confirmed 240/64/176). Cross-reference every literal count in markdown against len(os.listdir('skills')) and the json total_skills.
- Cross-artifact token cross-reference. Parse every backtick-coded skill token out of
each catalog/doc (skipping fenced code blocks), then diff against the folder set: orphans (token with no folder), missing (folder never mentioned), duplicates (token listed >1×). Diff skills.json community_skill_names vs folders to derive the implied custom set; compare to any custom markers in the catalogs.
- Placeholder-description finder. Regex the catalog table for empty/truncated cells
(| > |, | >- |, | ~ |) — unfinished entries in a "complete" deliverable.
- Coverage gaps. List folders in NO catalog row and NO cheatsheet trigger phrase;
prioritize surfacing high-value meta/orchestration skills that are undiscoverable.
Operational pitfalls (learned the hard way)
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/catalog-consistency-auditor/SKILL.md
Popularity
- 7-day trend score
- 5
Not refreshed yet.
Extension points
hookPre/post lifecycle hooks
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/mr-wolv/skill_playground_bakr101_2026/catalog-consistency-auditor)