github-action.md

GitHub Action

The Action runs the same CLI check in CI.

Workflow

After the first tagged release, use a pinned version:

name: agent-skill-lock
on:
  pull_request:
  push:
    branches: [main]

jobs:
  check-agent-capabilities:
    runs-on: ubuntu-latest
    permissions:
      contents: read
    steps:
      - uses: actions/checkout@v4
      - uses: harshmathurx/agent-skill-lock@v0.1.0

Do not use a floating branch such as main for a security-adjacent CI check.

Inputs

Input	Default	Meaning
root	.	Repository path to scan, relative to GITHUB_WORKSPACE

Example:

- uses: harshmathurx/agent-skill-lock@v0.1.0
  with:
    root: packages/web

Permissions

The Action needs only:

permissions:
  contents: read

It does not comment on PRs, write files, upload artifacts, or call a hosted service.

Root containment

The Action passes the root through an environment variable and enables workspace containment:

env:
  AGENT_SKILL_LOCK_ROOT: ${{ inputs.root }}
  AGENT_SKILL_LOCK_ENFORCE_WORKSPACE: "1"

That avoids direct expression interpolation into the shell command and prevents scanning outside GITHUB_WORKSPACE.

Failure behavior

A stale lockfile fails the job with exit code 1 and prints the drift.

The fix is deliberate:

agent-skill-lock update
git add agent-skills.lock.json
git commit -m "chore: update agent capability lock"

Review the lockfile diff before committing it. Updating the lockfile without understanding the change defeats the point.