mcpbeat

Action Audit

bitwarden/action-audit

> Audit GitHub Actions action usage across an org. Searches for a specific action (incident mode) or sweeps all workflow files for non-compliant action references (audit mode). Produces a read-only report of findings with compliance status and resolved SHAs. Does not modify any files. <example> </example> <example> </example>

2k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
129
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/bitwarden/ai-plugins --skill action-audit

What it tells the agent to use

found in the instruction text
Bash runs shell commands — read the instruction before connecting

The instruction itself

8 sections, as written by the author

Rules

  • This skill is strictly read-only. Do not modify, create, or delete any files.
  • No mutating API calls. gh api GET requests are allowed freely. Do not use -X POST, -X PUT, -X PATCH, or -X DELETE.
  • Flag uncertainty. If a finding is ambiguous, note it in the report rather than guessing.

Pin Compliance Rules

Before classifying any action reference, read ${CLAUDE_PLUGIN_ROOT}/skills/bitwarden-workflow-linter-rules/SKILL.md and apply the step_pinned rule as the compliance definition for all steps below. That skill is the single source of truth for what is and is not compliant.

Modes

  • incident (default): Targeted search for a specific action — used when an action is compromised or deprecated.
  • audit: Sweep all workflow files org-wide for any non-compliant action references.

Step 1: Parse Context

Determine the mode from the user's request:

  • If the user names a specific action (e.g., tj-actions/changed-files), use incident mode.
  • If the user asks for a general sweep of unpinned actions, use audit mode.
  • If a replacement action is mentioned, note it for the remediation step (handled separately by the action-remediate skill).

Step 2: Search Org-Wide

Incident mode — search for the specific action:

gh search code "uses: <action-name>" --owner <org> --path .github/workflows/ --limit 100

Also search without the uses: prefix to catch indirect references:

gh search code "<action-name>" --owner <org> --path .github/workflows/ --limit 100

Audit mode — find all workflow files and extract uses: references:

gh search code "uses:" --owner <org> --path .github/workflows/ --limit 100

Then apply the step_pinned compliance filter from ${CLAUDE_PLUGIN_ROOT}/skills/bitwarden-workflow-linter-rules/SKILL.md to each reference.

> Note: GitHub code search indexes can lag by minutes to hours after a recent push. Results may not reflect the very latest commits. Flag this caveat in the output.

Step 3: Parse and Display Results

For each uses: reference (excluding local ./ paths), determine:

  • Repo and file path
  • Current uses: value (full line)
  • Action type:
  • internal — starts with bitwarden/
  • third-party — all others (excluding local)
  • Pin status:
  • hash — pinned to a full 40-char SHA
  • tag — pinned to a version tag (e.g., @v3, @v1.2.3)
  • branch — pointing to a named branch (e.g., @main, @master)
  • none — no ref at all
  • Compliant: Apply the step_pinned rule from ${CLAUDE_PLUGIN_ROOT}/skills/bitwarden-workflow-linter-rules/SKILL.md — ✅ if compliant, ❌ otherwise.

Display a table:

| Repo | File | Current Reference | Type | Pin Status | Compliant |

| ---- | ---- | ----------------- | ---- | ---------- | --------- |

| ... | ... | ... | ... | ... | ... |

In incident mode, include all rows. In audit mode, omit compliant (✅) rows.

If there are no non-compliant findings, inform the user and stop.

Step 4: Resolve Remediation Targets

Apply the correct fix approach based on action type and mode. Do not treat all non-compliant references the same way.

Incident mode — replacement action provided:

If the user mentioned a replacement action in Step 1, do not resolve a SHA for the compromised action. Instead, resolve the SHA for the replacement action:

gh api repos/<owner>/<repo>/commits/<ref> --jq '.sha'

Present the resolved replacement SHA and a verification link (https://github.com/<owner>/<repo>/commit/<sha>) to the user. Ask for confirmation before finalizing.

Internal actions (bitwarden/):

  • The expected fix is to change the ref to @main. No SHA resolution needed.
  • If the action is currently on a SHA, do not automatically treat this as non-compliant — a SHA pin is more restrictive than @main and may be intentional (e.g., frozen during a security incident or pinned for reproducibility). Inform the user and ask whether to change it to @main before including it in the remediation list.

Third-party actions:

  • Resolve the current SHA for each unique non-compliant action:
gh api repos/<owner>/<repo>/commits/<ref> --jq '.sha'

Where <owner>/<repo> is the action's repo and <ref> is the target tag or main.

Present to the user:

  • Resolved SHA
  • Verification link: https://github.com/<owner>/<repo>/commit/<sha>

Ask: "Does this SHA look correct? Type yes to confirm, or provide a different SHA."

Wait for confirmation before finalizing the report.

> In audit mode, group unique third-party actions and resolve each once rather than per-occurrence.

Step 5: Summary Report

Output a final summary:

| Repo | File | Current Reference | Type | Compliant | Remediation |

| ---- | ---- | ----------------- | ---- | --------- | ----------- |

| ... | ... | ... | ... | ... | ... |

The Remediation column should contain:

  • For internal actions: change ref to @main
  • For third-party actions: the resolved 40-char SHA + inline comment to add (e.g., @abc123...def456 # v4.1.1)

Inform the user that they can use the action-remediate skill to apply fixes based on these findings.

How to use it

Copy the folder

Take bitwarden/action-audit from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.