kit-audit-skills
Skill6 starsChanged 27 days agoMITView on GitHub
Audit this repo’s skills and rules for overlap, contradiction, stale names, catalog drift, and employer/proprietary leakage. Use after adding, renaming, or removing a skill or rule, or when the kit feels too large.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- rules/AGENTS.mdAGENTS.md
- kit-audit-metricsSkill
- kit-capture-lessonSkill
- kit-challenge-designSkill
- kit-check-blastSkill
- kit-create-prSkill
- kit-debate-decision-treeSkill
- kit-debug-prodSkill
- kit-decisionsSkill
- kit-evolveSkill
- kit-explainSkill
- kit-learn-repoSkill
- kit-minimizeSkill
- kit-optimize-contextSkill
- kit-planSkill
- kit-proveSkill
- kit-reviewSkill
- kit-strip-diffSkill
- kit-triage-commentsSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 4 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 9
Treat the kit as a product.
- Very specific: line 15
Not token/host context audit (kit-optimize-context)
- Very specific: line 16
Not a code-diff review (kit-review)
- Distinctive: line 17
Not rewriting skill bodies for style unless a contradiction requires a one-line fix after they say go
- Very specific: line 48
Do not omit -l. A run without -l installs into the agent.
- Distinctive: line 100
Bar: if a stranger at another company would need that name to use the skill, it is a leak.
- Distinctive: line 115
Do not recommend merge and keep as separate without saying which name to type.
- Very specific: line 157
Follow AGENTS.md: delete dirs, update README catalogs + Usage, fix remaining kit-* refs, scrub employer/proprietary leaks, keep names …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Audit Skills
Treat the kit as a product. Every extra skill is a name the user must remember. Prefer fewer, sharper entries over a fat catalog.
Read-only unless the user asks to apply cuts. Recommend first; delete/merge only after approval.
What this is NOT
- Not token/host context audit (
kit-optimize-context) - Not a code-diff review (
kit-review) - Not rewriting skill bodies for style unless a contradiction requires a one-line fix after they say go
Inventory
Scan on disk (do not trust memory or git status alone):
skills/*/SKILL.md— directory name, YAMLname,description,disable-model-invocationrules/*.mdc— filename,alwaysApply/globs, description, full rule bodyrules/AGENTS.md— short index only (no forked bodies)README.md— Skills Catalog, Rules Catalog, Usage slash list, consumption matrixAGENTS.md— consumer install/use plus add/remove/rename instructions vs what exists
Build a table: on disk / in catalog / in Usage / leftover alias dir.
Checks
1. Catalog drift
- Skill dir exists but missing from README catalog or Usage
- Catalog/Usage names a skill whose dir is gone
- YAML
name≠ directory name - Old renamed folders still on disk next to the new name
2. Skills CLI parse
From the repo root, list skills without installing:
npx skills add . -lDo not omit -l. A run without -l installs into the agent.
Pass when the command exits 0 and the listed names match every skills/*/SKILL.md directory on disk (same count, same names). Fail when the command errors, or a skill on disk is missing from the list.
3. Leftover aliases
Old renamed folders still on disk (kit-self-review next to kit-review, etc.) count as duplicates, not compatibility. Flag for delete unless the user explicitly wants a stub.
4. Same job, two names
Cluster by user intent, not title:
| Cluster | Smell |
| Think-before-code | Plan / reframe / grill / minimize overlapping “What this is NOT” |
| Fat diff | Strip extras (kit-strip-diff) vs smallest path before code (kit-minimize) vs bug review (kit-review) |
| Prove it | Smoke vs runtime vs blast-radius with the same burden of proof |
| Review | Author vs reviewer as two skills instead of modes |
| Ship | Tiny git recipes that are four commands |
| Always-on habits | Comment style, delegate, “always ask” — these belong in rules/, not slash skills |
A skill is redundant if a user who forgets its name would still get the outcome from another skill.
5. Contradictions
- A says “use B first”; B says “not A” or points at a deleted name
- Two skills claim the same trigger in
description - A rule and a skill teach the same behavior (pick one surface)
What this is NOTpoints at a skill that no longer exists- Description “Replaces X and X” / self-replacements / duplicate paths
Grep all kit-* mentions across skills/, rules/, README.md, AGENTS.md. Every hit must resolve to a live skill or live rule (examples like kit-new-skill in AGENTS.md are fine).
6. Skill vs rule
Move to rule when the guidance must fire without a slash (comments, git hygiene, delegate). Keep a skill when it is a paced workflow the user opts into (plan, review, prove, create-pr).
7. Size bar
Slash-invoke kits go stale around ~12–14 skills. Count live SKILL.md dirs. If over that, propose a cut list — do not only say “it’s a lot.”
8. Employer / proprietary leakage
Public kit must stay generic (see AGENTS.md — Keep skills generic).
Scan skills/, rules/, README.md (and examples under **/references/) for:
- Company or employer product brands
- Work emails, private hostnames, internal VPN/SaaS URLs
- Employer-only tool paths presented as the only way
- Comments that cite an employer as provenance (“from X’s rules”)
Bar: if a stranger at another company would need that name to use the skill, it is a leak.
Hits → fix (scrub or move to the consuming repo’s wiring file).
Policy lines that forbid employer details (e.g. “no employer hostnames”) are fine; concrete brand names are not.
Verdicts (required)
For each live skill/rule, one of:
- keep — unique job, links are clean
- merge into X — same job; say which name survives
- move to rule — always-on habit
- remove — too small, unused, or leftover alias
Do not recommend merge and keep as separate without saying which name to type.
Output
## Kit audit
**Counts:** N skills, M rules (catalog says …)
### Drift
- …
### CLI parse
- …
### Overlaps
- {A} ∩ {B} — same job because …
- Recommend: merge / keep split (why)
### Contradictions / stale refs
- `file` → `kit-dead-name`
### Skill vs rule
- …
### Leakage (employer / proprietary)
- …
### Proposed catalog
- keep: …
- merge: …
- rules: …
- remove: …
### Apply?
Stop. Ask which verdicts to execute.Cite paths. Do not invent invocation history — if you do not know what they used last month, say so and cut on overlap, not on guessed usage.
After they approve cuts
Follow AGENTS.md: delete dirs, update README catalogs + Usage, fix remaining kit-* refs, scrub employer/proprietary leaks, keep names generic.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

