skill-audit
Skill0 starsChanged 2 months agoMITView on GitHub
Governs writing, editing, compressing, merging, splitting, or deleting skill files in this library, including inside skill-creator runs. Trigger on "audit my skills", "fix this skill", "compress this skill", "should these be merged", "write a new skill", "why didn't that skill fire", "this skill isn't helping", or when Nick reports bloated or over-templated output he can't trace to a source. Not for using a skill to do its job.
64 other projects use a version of this.
Adopted by
Most recently taken up by
- GravionLabs/skillstheir own version, first seen Nov 15, 2026
- qiuyiwu1989-star/openclaw-skill-opstheir own version, first seen Nov 15, 2026
- DancinGhost/agent-metaskillstheir own version, first seen Nov 15, 2026
- vlad-ryzhkov/ai-context-engineering-for-frontendtheir own version, first seen Nov 15, 2026
- macintog/skillstheir own version, first seen Nov 15, 2026
- automagik-si/genietheir own version, first seen Nov 15, 2026
Used with
Projects with it also use root-cause, write-a-skill and 8 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
Ingramradical235Skillroot-cause
Structured root cause analysis using WHERE (MECE decomposition) then WHY (causal structure diagrams) pipeline. Includes coin-flip detection, logical-leap detection, and vicious cycle intervention. Use when buffer enters YELLOW/RED, when metrics stall, or when Actions execute but don't produce results.
Also in 3 projects that use this
simo97Skillwrite-a-skill
Create new agent skills with proper structure, progressive disclosure, and bundled resources. Use when user wants to create, write, or build a new skill.
Also in 4 projects that use this
Microsoft AzureSkillreflect
Learning capture system that extracts HIGH/MED/LOW confidence patterns from conversations to prevent repeating mistakes. Use after user corrections ("no", "wrong"), praise ("perfect", "exactly"), or when discovering edge cases. Complements .squad/agents/{agent}/history.md and .squad/decisions.md.
Also in 5 projects that use this
sunnyclinches44Skillexplain
Explains code using real-world analogies and ASCII diagrams
Also in 4 projects that use this
eherrerosjSkillhumanize
Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive 'Signs of AI writing' guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases. Triggers on: 'humanize', 'make this sound human', 'remove AI writing', 'de-AI', 'sound more natural', 'less robotic', 'too AI-sounding', or when reviewing text that reads like AI output.
Also in 3 projects that use this
MicrosoftSkillreport
Turn an exploration (threads, findings, charts) into a single Markdown report — note, blog post, executive summary, KPI dashboard, slide brief, or multi-section analytical report, with embedded charts.
Also in 3 projects that use this
Other versions64 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 64
- nyldn/claude-octopus★ 4.4ksame name
- automagik-dev/genie★ 353same name
- automagik-si/genie★ 353same name
- namastexlabs/automagik-genie★ 351same name
- Rtur2003/Claude-Code-Promts-Skills★ 53same name
- fimoklei/pm-ai-playbook★ 42same name
- cth9191/skill-audit★ 23same name
- anysiteio/agent-skills★ 20same name
- okjpg/skill-audit★ 15same name
- macintog/codex-spine★ 12same name
54 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository15 other files it gives its agents.
- blog-cold-readSkill
- content-proportion-checkSkill
- defensibility-checkSkill
- giving-feedbackSkill
- hamming-taste-principlesSkill
- idea-to-briefSkill
- learning-mediumSkill
- linear-write-cleanupSkill
- portable-handoffSkill
- saddles-positioningSkill
- session-stateSkill
- strategy-decisionsSkill
- transcript-auditSkill
- uncertainty-killerSkill
- writing-craftSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 56 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 36
Never fix a body-layer problem by growing a description.
“Common” is measured: kept by at least 6 in 10 of the 56 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill Audit
The library was built against an older baseline. The binding constraint has moved from capability to salience: I can do the work, but competing directives, prescribed shapes, and expository prose degrade it.
The test for every line: could I have produced this without being told? If yes, it costs attention rather than adding it. Skills earn tokens by carrying what only Nick knows — decisions, constraints, sources of truth, taste, exemplars — not by teaching me how to work.
Answer this before editing anything
Does this skill exist to change my output, or to change Nick's thinking?
Some files are notebooks Nick wrote to think with, that I read as instruction sets
(hamming-taste-principles is explicitly a research log). Compressing one of those and demoting its
reasoning to references/ optimizes it for me and destroys it for its actual primary reader. The two
purposes want opposite treatment, so resolve this first. If it serves both, split it.
Two layers, very different costs
| Layer | When it loads | Discipline |
| Description | Every turn, every skill, always | The expensive layer. Trigger signals plus one boundary clause. Budget ~400 chars. No rationale, no advocacy, no restating the body. |
| SKILL.md body | Only when the skill fires | Cheap by comparison. Compressing here saves tokens that were only spent when the skill was relevant. |
Never fix a body-layer problem by growing a description. Precedence rules belong in the body of whichever skill owns the trigger, not in the always-resident layer.
Classify — wrong class means opposite prescription
| Class | Purpose | Prescription |
A — Rails (gtm-rules, product-source-of-truth, deal-update) | Stop me doing something plausible but wrong | Imperative, absolute, short. Verify every tool name against the live toolset — a rail naming a dead tool has absolute authority and sends me nowhere. Stale rails are worse than missing ones. |
B — Taste (content-proportion-check, writing-craft, linear-write-cleanup) | Transfer Nick's standard | Decisions and discriminating tests in SKILL.md; reasoning in references/. Compress after benchmarking, never on the strength of an argument. |
C — Producers (case-study, cx-demo-brief, morning) | Reliably build an artifact | Keep format prescription — a real consumer needs it. Deterministic steps → scripts/. |
D — Coaching (strategy-decisions, learning-medium) | Change Nick's thinking, not the output | State explicitly: the output is a question, not a document. Otherwise co-firing Class B converts it into a report about coaching. |
E — Meta (skill-creator, schedule, setup-cowork) | Operate the system | Leave alone unless asked. |
Mixed-class files get split — they get swallowed by whichever half fires louder.
Rules
- Verify tool names. Any skill naming a tool gets its references checked against the live toolset. Tool names drift; skills don't notice. This is the highest-severity defect class in the library.
- Verify, don't flag. An instruction to "flag this for Nick" that a live tool could close is a tax
charged on every fire.
search_callsandget_call_transcriptcan source a practitioner claim; links can be fetched. Convert flag-and-report into verify-then-report. - One trigger, one owner. Grep the library before adding trigger phrases. On collision, prefer deduplication and narrowing over merging — a merged skill inherits the union of trigger surfaces and fires on a superset of signals. Merge only when both skills would return the same verdict. If you claim a precedence order, write it out; naming the fix is not the fix.
- Templates: measure, don't assume. Prescribed sections cut both ways — they invite filler when I have nothing real for a slot, and they prevent omission by forcing enumeration. For a skill whose job is catching what's missing, omission is the costlier error. Keep format where a downstream consumer needs it (a HubSpot field, a .docx layout, a doc read cold). Elsewhere, benchmark before removing.
- Provenance is one line. Name the source so we share vocabulary; don't re-teach the framework.
- Write in the register you want back. Skill prose primes output prose. A file about concision written expansively argues against itself.
- Earn "always." Proactive-override language only where genuinely unconditional. When a third of the library claims priority and none states rank, the word does nothing.
- One verification line. "Before returning, check X." Cheapest quality gain available.
- Verify adversarially. For any substantial skill edit or audit claim, spawn a fresh-context agent to
fact-check it against the actual files. Self-review does not catch this class of error — a
well-evidenced-looking audit can be wrong in most of its specifics and look right from the inside.
blog-cold-read's zero-context grader is the pattern.
A longer rubric with worked examples stays in the private library; the classes above cover most edits.
Measure instead of arguing
skill-creator ships the harness: scripts/run_eval.py, scripts/aggregate_benchmark.py,
agents/grader.md, eval-viewer/generate_review.py. It runs with-skill against baseline in parallel and
reports pass rate, tokens, and time.
Nothing here has been baseline-tested against the current model, and baseline is what moved. A skill whose with-skill result matches baseline is pure cost. Do not set line-count targets by aesthetics; the non-derivable content sets the floor, and measurement finds it.
Before returning any skill edit
State the class, the description-length delta and body-length delta separately, what was cut and where it went, and which trigger phrases now collide with which other skills. If a cut removed something non-derivable, say so and put it back.
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.

