github-repo-audit
Skill0 starsChanged 32 days agoMITView on GitHub
Checks a repository before it is shared or made public, then audits how it presents itself and returns a scored checklist with ranked fixes: README, license, community health files, description, topics, social preview, releases, CI and security signals, and root layout, plus a pre-public secrets and history check. Read-only. Make sure to use this skill whenever the user asks what is missing from a repository, wants a repo reviewed or audited, asks whether a project is ready to share, launch, open source, or make public (including a private repository about to go public, where it checks for leaked secrets first), or asks how to improve a repository without naming one specific file. Not for writing the fixes: each fix names the skill that writes it.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- techysy/yangyu-hermes-skillstheir own version, first seen Oct 13, 2026
- dancolta/github-repo-audittheir own version, first seen May 18, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- techysy/yangyu-hermes-skills★ 8same name
- dancolta/github-repo-audit★ 2same name
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 repository9 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- github-community-filesSkill
- github-discoverabilitySkill
- github-profile-optimizerSkill
- github-readme-writerSkill
- github-release-writerSkill
- github-repo-familySkill
- github-repo-hardeningSkill
- github-upstream-contributorSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 4 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Before a launch, a release announcement, or a job application that links the repository
- Distinctive: line 30
Use the format in references/examples.md.
- Very specific: line 49
2. Collect remote facts: Run the bundled script. It makes read-only gh calls and prints one JSON object:
- Distinctive: line 71
Read-only: Do not create, edit, or delete files, and do not run any remote change, even if a fix is one line.
- Distinctive: line 74
Never print secret values: If a secret seems to be tracked, report the file path and the kind of secret only.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
GitHub repository audit
Score a repository against what visitors, contributors, and GitHub itself look for, then hand back a short, ranked list of fixes. Each fix points to the skill in this package that writes it. This skill never changes anything.
When to use
- "Audit my repo", "review this repository", "what is missing from my GitHub project"
- "Is this ready to make public?" or "I am about to open source this" (use
pre-publicmode) - Before a launch, a release announcement, or a job application that links the repository
- As the first step when the user asks to "improve" a repository without saying what to improve
Input
- Target (
$ARGUMENTSor ask): a local path (default: the current directory) orowner/repo. - Mode:
standard(default) orpre-public. Usepre-publicwhen the user mentions going public or open sourcing, even if they did not name the mode. - Goal (optional): who the repository is for, such as users of a library, contributors, employers, or researchers. It changes which checks are critical.
Output
- A summary line:
N of M applicable checks pass. Critical: X open.If GitHub reports a community health percentage, show it too, labeled as GitHub's number. - A results table grouped by area, with Pass, Fix, or N/A, and the evidence for each verdict.
- Top fixes, ranked: at most seven, critical first. Each names its sibling skill and the exact first action.
- In
pre-publicmode: a separate Before going public block with blocking findings.
Use the format in references/examples.md.
Workflow
Copy this checklist into your response and tick items off as you go:
Audit progress:
- [ ] 1. Resolve the target and data tier
- [ ] 2. Collect remote facts
- [ ] 3. Scan local files
- [ ] 4. Score every check
- [ ] 5. Rank the fixes
- [ ] 6. Pre-public checks (pre-public mode only)
- [ ] 7. Report1. Resolve the target. For a local path, read git remote get-url origin to find owner/repo. If there is no remote, audit local files only and mark remote-only checks N/A (no remote). Check gh auth status once and pick the data tier from operating-rules.md.
2. Collect remote facts. Run the bundled script. It makes read-only gh calls and prints one JSON object:
bash "${CLAUDE_SKILL_DIR}/scripts/collect_repo_facts.sh" OWNER/REPOA null field means the call failed or needs more permission (branch rules, for example). Report it as Unknown, never as Fix. If the script exits with code 3, gh is unavailable. Continue with local files and, if they are available, the GitHub connector tools (data sources), and say which checks you could not run. If it exits with code 4, the repository was not found or is not visible to the signed-in account. Confirm the name with the user instead of reporting every check as Unknown.
3. Scan local files. Use Glob and Read on the working tree. Look for the README in .github/, root, and docs/, community files in their supported locations, .github/workflows/, manifests, .gitignore, and large or generated files. Run git ls-files | head -200 to see what is tracked. Read only the parts of large files you need.
4. Score every check in references/checklist.md. For each check, record the verdict and one line of evidence: a path, a field value, or a command result. Mark a check N/A when it does not fit the project type. For example, CITATION.cff applies only to research software and datasets. Adjust severity to the stated goal as the checklist describes.
5. Rank the fixes. Order by severity, then by impact on a first-time visitor, then by effort, smallest first. Merge fixes one skill handles in one pass. For example, a missing CONTRIBUTING and a missing SECURITY become one github-community-files item.
6. Pre-public checks. In pre-public mode, run the "Before going public" checks in the checklist. Anything that could leak a secret or personal data blocks publishing until it is resolved. When gitleaks or trufflehog is installed, suggest running it and show the command. Do not install either tool.
7. Report in the output shape from operating-rules.md. This skill has no commands awaiting approval. When checks are Unknown because the data tier cannot see them, close the Result part (before Next step) with a How to check the Unknowns list that gives the user one exact read-only command or settings path for each. For example: rulesets with gh api repos/OWNER/REPO/rules/branches/BRANCH, security features with gh api repos/OWNER/REPO --jq .security_and_analysis (or Settings → Advanced Security), and the social preview in Settings → General → Social preview.
Hard rules
Shared rules: operating-rules.md. Additional rules for this skill:
- Read-only. Do not create, edit, or delete files, and do not run any remote change, even if a fix is one line. Offer the sibling skill instead.
- Evidence or Unknown. Every verdict cites what you saw. If you could not see it, the verdict is Unknown, not Fix, and it stays out of Top fixes: list it under How to check the Unknowns instead. A pasted file list or summary is not proof that something is absent unless the user says it is complete, or explicitly says the thing is missing.
- No invented scores. Report counts of checks, not made-up percentages or grades. GitHub's
health_percentagemay be quoted as GitHub's figure. - Never print secret values. If a secret seems to be tracked, report the file path and the kind of secret only.
Example
Input: /github-repo-audit acme-labs/tinyqueue
Output (abridged):
**tinyqueue audit:** 16 of 30 applicable checks pass. Critical: 4 open (A1, B3, B4, D1). (GitHub community health: 57%.)
| Area | Check | Verdict | Evidence |
|---|---|---|---|
| Front door | A1 Description | Fix | Empty `description` |
| README | B3 Install steps | Fix (Critical) | README has no install command; `pyproject.toml` defines package `tinyqueue` |
| Community | C1 License | Pass | `LICENSE` detected as MIT |
| Releases | D1 Published releases | Fix (Critical) | 4 tags (`v0.1.0` to `v0.4.0`), 0 releases |
**Top fixes**
1. Critical: add install and quick-start sections. Run `github-readme-writer` (improve mode).
2. Critical: publish GitHub Releases for the existing tags. Run `github-release-writer`.
3. Critical: write a description and 8 to 12 topics. Run `github-discoverability`.The complete example, including pre-public mode, is in references/examples.md.
Reference files
- references/checklist.md: every check, with its severity, how to verify it, and the skill that fixes it
- references/examples.md: two complete audits, a standard one and a pre-public one
scripts/collect_repo_facts.sh: the read-onlyghcollector. Run it; do not read it into context.- references/shared/github-facts.md: platform limits cited in verdicts
Related skills
github-readme-writer: README fixes (area B)github-discoverability: description, topics, homepage, social preview (area A)github-community-files: LICENSE, CONTRIBUTING, SECURITY, templates (area C)github-release-writer: releases and changelog (area D)github-repo-hardening: rulesets, security features, and Actions hardening (area E)github-repo-family: when the same fixes repeat across many repositories, use account-wide defaults instead
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.

