dependency-vetting
Skill0 starsChanged 3 months agoApache-2.0View on GitHub
Verify a third-party package exists, is the intended one, and is safe to depend on before adding it. Use whenever adding or upgrading a dependency, reviewing a PR that changes a manifest or lockfile, or evaluating a suggested library. Covers hallucinated-package (slopsquatting) defense, typosquatting, license and maintenance checks.
- Installs packages
2 other projects use a version of this.
Adopted by
Most recently taken up by
- svyatov/supermatttheir own version, first seen Sep 23, 2026
- TheDecipherist/claude-code-mastery-project-starter-kittheir own version, first seen Jun 26, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- TheDecipherist/claude-code-mastery-project-starter-kit★ 337same name
- svyatov/supermatt★ 20same 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 repository18 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ai-impact-assessmentSkill
- artifact-intakeSkill
- backlog-prioritizationSkill
- blameless-postmortemSkill
- control-assessmentSkill
- control-verificationSkill
- eval-designSkill
- evidence-writingSkill
- exploratory-testingSkill
- red-teamingSkill
- retro-facilitationSkill
- scanner-triageSkill
- secure-code-reviewSkill
- story-splittingSkill
- test-strategySkill
- threat-modelingSkill
- writing-acceptance-criteriaSkill
The file
Files
Dependency vetting
Consumers: developer, /implement, /review, pr-governance.yml,
process/gates/g3-code-complete.md.
Why this exists as a distinct control
Models hallucinate plausible package names. Attackers monitor for the popular hallucinations and register them — slopsquatting. The suggested import looks reasonable, the install succeeds, and you have executed an attacker's code at build time with full CI privileges.
This is the supply-chain risk specific to AI-assisted development (AIC-7), and the check that prevents it costs about ninety seconds.
Before adding any package
1. Does it exist, and is it the one you mean?
- Publisher/owner — is it the org you expect?
requestsis published bypsf; arequests-libpublished by a three-week-old account is not the same thing. - Source repository — does the registry page link to a real repo, and does that repo's README match the package?
- Download history — a genuinely popular library has years of downloads. A package with 200 downloads and a name you recognize is a red flag, not a discovery.
- Age — first published last month, claiming to be a mature utility? No.
- Name distance — is it a near-miss of something popular? Common tricks: hyphen vs.
underscore (
python-dateutil/python_dateutil), added or dropped prefix (js-yaml/jsyaml), plural/singular, transposed characters,-js/-pysuffixes.
If the model suggested this package and you have not personally confirmed it exists, assume it does not.
2. Should you depend on it at all?
- Is it needed? Left-pad problems are real. A 6-line utility is not worth a supply-chain edge, a license obligation, and a permanent upgrade duty.
- Maintenance signal — recent commits, issues answered, a release in the last year. An unmaintained dependency is a scheduled vulnerability (SA-22).
- Transitive weight — how many packages does this actually pull in? Check before, not after.
- Alternatives — including the standard library, and including writing the six lines.
3. License
Compatible with your distribution model. Copyleft usually needs a decision rather than a blanket block — decide deliberately rather than inheriting a default from a scanner config. Record the decision.
4. Security posture
- Known vulnerabilities (SCA runs in CI, but check before you commit)
- Does it require elevated install-time execution — post-install scripts, build hooks?
- Does it phone home?
Recording it
The PR must carry the justification. pr-governance.yml labels manifest changes
dependency-change and asks for this explicitly:
| Package | Version | Why | License | Existence verified |
|---------|---------|-----|---------|--------------------|
| pydantic | 2.9.2 | Schema validation at the API boundary; replaces 200 lines of hand-rolled validation | MIT | github.com/pydantic/pydantic, 100M+ monthly, publisher `samuelcolvin` — confirmed |"Existence verified" is not a checkbox. Say what you checked.
Locking
- Commit lockfiles. Always.
- Install frozen —
npm ci,pip install --require-hashes,go mod verify,cargo --locked. Notnpm install. - Integrity hashes verified, not just versions pinned.
GitHub Actions specifically
Pin to a full commit SHA, never a tag.
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2 ✓
- uses: actions/checkout@v4 ✗A tag is mutable. A compromised action tag is a supply-chain compromise of every workflow that uses it — including the ones enforcing your other controls. Dependabot maintains the SHAs and the version comments for you.
Upgrades
- Patch/minor — Standard change. Grouped, automated, still CI-gated and still reviewed by a non-author.
- Major — Normal change. Gets a story, an impact assessment, and real review. Never an auto-merge path.
- Security patch — enters on the finding's SLA (Critical 7d / High 30d), ahead of feature work.
Removal
Removing a dependency is a legitimate, valuable change. Every one you drop is attack surface, upgrade toil, and license obligation gone permanently.
Controls
SR-3 (supply chain controls) · SR-4 (provenance) · SR-4(3) (validate as genuine and unaltered) · SR-5 · SR-11 (component authenticity) · SA-4 · SA-22 (unsupported components) · CM-8 (component inventory / SBOM) · RA-5 · AIC-7 (supply chain integrity of AI suggestions) · OWASP LLM05.
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.

