The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to audit GitHub Actions workflows and configuration files, which are external data sources that could potentially contain malicious instructions intended to manipulate the agent during the auditing process. Use of this skill involves processing untrusted inputs such as pull request titles and branch names.
Ingestion points: Instructions specify reading repository workflow YAML files (.github/workflows/*.yml) and pull request metadata.
Boundary markers: The instructions mandate passing GitHub expressions through quoted environment variables to prevent shell injection and require explicit permission declarations.
Capability inventory: The agent is authorized to inspect, modify, and validate workflow configurations and CI logs.
Sanitization: Recommends external linting and security scanning tools like actionlint and zizmor to verify configuration safety.
[EXTERNAL_DOWNLOADS]: The skill references and includes a code snippet for the dataaxiom/ghcr-cleanup-action GitHub Action. Although the skill correctly recommends pinning the action to a specific commit SHA (d52806a0dc70b430571a37da1fde39733ffd640f) to provide supply chain security, it introduces a dependency on a third-party provider not included in the pre-verified trusted organizations list.

github-actions-maintainer