Skip to content
SkillModel-led4,780 tokens in use

Agentic Actions Auditor

Audits GitHub Actions workflows that run AI coding agents and traces attacker-controlled input from issues and pull requests into agent prompts.

8,319 stars on the repository CC-BY-SA-4.0trailofbits/skills · plugins/agentic-actions-auditor/skills/agentic-actions-auditor
/plugin marketplace add trailofbits/skills
View source

Install command as written in the repository's README

Skill label

measured from its files

What installing Agentic Actions Auditor costs

106tokens idle
4,780tokens in use

Idle is what sits in context while the skill is installed. In use is its instructions, loaded when it runs. Counts are approximate, made offline with cl100k_base.

Model dependence Model-led

Script-ledMixedModel-led

Mostly instructions, so how well it works depends on the model that follows them.

This describes the skill's structure, how much of it is scripts. It does not say how well it works.

Touches

shelldeclared and detectednetworkdetected

Declared tags come from the skill's frontmatter; detected ones from its scripts, its code blocks and the commands it names in inline code. This is detected, not audited.

How it scores

Quality

How well the skill is written, counted from its own files: its frontmatter, its steps, examples, stated output and constraints, its size and whether every file it ships is one it names.How we score it →
Good
  • Frontmatter12/15
  • Structure25/25
  • Examples12/15
  • Output10/15
  • Size10/15
  • Resources0/15

Security

Starts at 100 and loses points for each thing in the skill's files that a reader should look at first.How we score it →
Excellent
  • Reaches the network−10

    does this on the machine, and the frontmatter does not say so

    SKILL.md, line 72

Repository health

Pre-computed score based on development activity, maintenance, community, maturity, and trend momentum.How we score it →
Good
  • Development activity100
  • Maintenance52
  • Community64
  • Maturity32
  • Momentum40

Agentic Actions Auditor is a static analysis guide for CI/CD pipelines that invoke AI agents: Claude Code Action, Gemini CLI, OpenAI Codex and GitHub AI Inference. It finds the workflow files, picks out the AI steps, records their configuration and checks nine attack vectors.

The result is a findings report with a severity, the file and step, an evidence snippet, a numbered data flow from the attacker to the agent, and a fix tied to the specific action. It works on a local repository or on a remote one by owner and name.

What You Get

  • A findings report with severity, file, job and step, impact, evidence and remediation for each issue
  • A numbered data flow trace from the attacker-controlled GitHub event to the agent action
  • A summary table with one row per workflow and its highest severity
  • A clean-repo report that lists the workflows scanned and the AI actions found when nothing is wrong
  • Remote analysis output with links to each workflow file on GitHub

Common Use Cases

  • Reviewing a repository before enabling Claude Code Action on pull requests
  • Checking whether issue or PR text can reach an agent prompt through an env block
  • Auditing sandbox settings and user allowlists on Codex or Gemini workflow steps
  • Assessing an open-source project’s CI for prompt injection risk

What Makes It Unique

Most Actions linters look for direct expression injection in run steps. This skill is built around AI agents and covers the env var intermediary pattern, where event data reaches a prompt through an env block with no visible expression, along with CLI data fetches, subshell expansion in allowed tools and eval of agent output. It also has a list of rationalizations the auditor must reject, and a rule that sandbox and allowlist weaknesses only amplify an injection path.

Under The Hood

The skill is instructions plus reference files, one per attack vector and one per action profile. It declares Read, Grep, Glob and Bash as allowed tools. Bash is limited by the skill’s own rules to GitHub API calls through the gh command that fetch workflow listings and file contents from a remote repository, and the skill tells the agent never to pipe fetched YAML into a shell or interpreter. Locally it only reads files, and it does not modify workflows.

Join founders building with open source

Opinionated takes, migration guides, cost-saving tips, and insights from the open source ecosystem.

Subscribe on SubstackJoin 750+ subscribers