Files
dev-workflows/skills/safedeps/SKILL.md
Evandro Fonseca Junior bb3fe21766 fix(skills): vendor the support files the vendored skills point at
The vendored SKILL.md files were byte-identical to upstream, but the
playbooks, templates and schemas they reference were left behind in the
original repos, so a loading skill pointed at a file that does not exist
anywhere in this repository.

Vendored 1:1 into each skill directory (so they survive
`cp -r skills/* ~/.hermes/skills/`):

- plays/ for api-security-review, code-review-security, sca-audit,
  secrets-scan, web-security-review (OWASP/secure-agent-playbook)
- templates/finding.md for the four OWASP skills that cite it
- schemas/finding.schema.json for cve-triage, patch-prioritization,
  dependency-scanning (UnitOneAI/SecuritySkills)
- docs/sarif-output.md, docs/fixer-policy.md and
  docs/normalized-json-output.md for dependency-scanning

Path fixes, the only edit to vendored text:
- the three UnitOneAI skills referenced ../../../schemas/ and ../../../docs/
  which assumed upstream's skills/<category>/<skill>/ depth; this repo is
  flat, so the links now resolve inside the skill directory
- safedeps told the agent to run <skill_dir>/../../check_deps.py while the
  script is vendored next to SKILL.md; corrected to <skill_dir>/check_deps.py
  and documented the script's exit codes (1 = finding, not a crash)
- NOTICE.md records the new files, the path edits, and the remaining known
  gaps (hermaguard's kensei-pipeline references, the ASVS dataset citations,
  and hermaguard being behind upstream)

Verified: every backticked path in every SKILL.md now resolves;
check_deps.py runs from the installed location and returns live OSV.dev
data ([email protected] -> 2 advisories); hermaguard's 46 tests still pass.
2026-09-14 07:40:21 -03:00

2.9 KiB

name, description
name description
safedeps Grounds dependency decisions in live OSV.dev vulnerability data instead of a model's frozen training-cutoff knowledge. Trigger whenever you are about to add a dependency, bump/upgrade a package version, choose between package versions, answer "is <package> safe / vulnerable", or review a manifest or lockfile (requirements.txt, package.json, package-lock.json, go.mod) for security issues. Also trigger on requests like "check my dependencies for vulnerabilities" or "what version of X should I use".

safedeps

Your training data has a cutoff. CVEs published after it are invisible to you, and a vulnerable package version doesn't error at runtime -- it just runs, quietly exploitable. Don't guess. Ask OSV.dev.

When to run this

Before you write a version number into a manifest, or when asked to review one, run the checker and use its real output instead of your own recollection of which versions are safe.

python3 <skill_dir>/check_deps.py <package>@<version>:<ecosystem>
python3 <skill_dir>/check_deps.py path/to/requirements.txt

(<skill_dir> is this file's directory; the script is vendored next to this file, as check_deps.py. Ecosystems: npm, PyPI, Go, crates.io, Maven, etc. -- omit :ecosystem and it defaults to npm.)

The script prints one of two things per package, straight from OSV.dev:

OK    npm:[email protected] -- no known advisories
VULN  npm:[email protected] -- 2 advisory(ies):
  - CVE-2026-4800 [HIGH] lodash vulnerable to Code Injection via `_.template` imports key names
      fix: upgrade to 4.18.0

What to do with the output

Exit code is 0 when every package is clean, 1 when at least one advisory was found, and 2 when OSV.dev was unreachable -- 1 is a finding, not a crash. (The manifest mode only accepts files named requirements.txt, package.json, package-lock.json or go.mod.)

  1. If a package you're about to add/bump shows VULN, do not use that version. Use the version named in fix: instead (or newer).
  2. If no fixed version is listed, tell the user there's an open advisory with no patch yet -- let them decide.
  3. Always surface the advisory IDs and severities to the user; don't silently swap versions without saying why.
  4. If the script reports OSV.dev is unreachable, say so plainly and fall back to your own knowledge with a caveat that it may be stale -- don't pretend you checked.

Examples

Adding a dependency: user asks to add flask to a Python project. Run check_deps.py [email protected]:PyPI (the version you were about to pick) before writing it to requirements.txt. If it's clean, proceed. If not, check the fix: version and use that instead.

Reviewing a project: user asks "are my dependencies safe?" Run check_deps.py requirements.txt (or package.json) in their project root and report every VULN line with its fix version.