| name | tooling-supply-chain-audit |
| description | Audit third-party packages, libraries, plugins, MCP servers, or upstream repos before pulling them in. Confirms maintainers real, code not malicious, project maintained. |
| allowed-tools | Bash Read Grep WebFetch |
Supply chain audit
Triggers
audit this dep, vet this crate, is this safe, supply chain audit, trust check.
Vet a third-party package before it ships in your code. The output is a trust verdict (green / yellow / red) plus a short writeup. Don't recommend "allow" without running the checks.
When to run
Run this skill BEFORE:
- Adding a new direct dependency to
Cargo.toml, package.json, pyproject.toml, requirements.txt, go.mod, Pipfile, Brewfile, or any other manifest.
- Installing a new MCP server (
claude mcp add / codex mcp add, ~/.claude.json / ~/.codex/config.toml, or a plugin marketplace install).
- Adding a new GitHub Action (
uses: org/action@vX) or workflow that pulls a third-party container.
- Pulling a new brew tap or formula.
- Cloning and running an upstream repo the user found in a search result, blog post, or social link.
- The harness denies a
cargo fetch / npm install / pip install / similar action with a "guessed external dependency" or "untrusted external code" error.
- the user says "audit this," "vet this," "deep scan this," "is this real," or "should I trust this."
Skip this skill for:
- Bumping an existing pinned version of a dep that is already trusted (handled by dependabot review, not a fresh audit).
- Pulling from a known/trusted repo or registry.
- One-off
gh api / curl reads of public docs.
Verdict scale
- Green - Allow. Proceed normally. Ordinary maintenance risk (any dep can rot or get compromised later); use
cargo audit / npm audit / dependabot for ongoing watch.
- Yellow - Allow with caveats. Document the yellow flags in the audit writeup so future-you knows the soft spots. Examples: bus factor of 1, very young project, single small maintainer.
- Red - Stop. Do not add. Surface findings to the user. Examples: typosquat, code suggests data exfiltration, account hijack signals, abandoned with no path forward, license mismatch with project.
Audit checklist
The full multi-step checklist (org reality, maintainer signals, repo health, code review, downstream adoption, supply-chain risk patterns, build-script red flags, hijack patterns) lives in references/audit-checklist.md, with deeper-detail companions in references/maintainer-signals.md, references/build-script-redflags.md, and references/hijack-patterns.md.
More
