The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of GitHub Actions workflow files and configuration metadata from GitHub repositories using the gh api tool. This functionality is restricted to well-known services (GitHub) and is necessary for the skill's primary purpose of CI/CD security auditing.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted third-party workflow YAML files which could contain adversarial content. To mitigate this risk, the methodology includes a mandatory 'Bash Safety Rules' section that strictly prohibits the agent from executing, sourcing, or piping any fetched content into shell environments or language interpreters, treating all external content purely as static data.
Ingestion points: Workflow files and composite action configurations are fetched via the GitHub Contents API in SKILL.md Step 0 and Step 2.
Boundary markers: The 'Bash Safety Rules' in SKILL.md Step 0 serve as internal boundary markers for the agent.
Capability inventory: The skill uses Read, Grep, Glob, and Bash (specifically for gh api calls).
Sanitization: The skill methodology explicitly forbids passing external content to execution sinks like bash, python, or eval.

agentic-actions-auditor