The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill configures a CI/CD pipeline that processes external, untrusted data such as Git commit messages and Pull Request metadata to automate release notes and versioning. This creates an attack surface where a contributor could attempt to influence the automated output via specially crafted commit messages.\n
Ingestion points: Git commit messages and PR data ingested by semantic-release in the templates/release-on-demand.yml workflow.\n
Boundary markers: Relies on the Conventional Commits specification to distinguish release instructions from prose.\n
Capability inventory: The workflow requires contents: write, issues: write, and pull-requests: write permissions to push tags and create GitHub Releases.\n
Sanitization: Handled by the internal logic of the semantic-release plugins.\n- [EXTERNAL_DOWNLOADS]: The GitHub Actions workflow templates reference official GitHub Actions (actions/checkout, actions/setup-node) for repository checkout and environment setup. These are well-known, trusted services and are pinned to specific commit hashes for integrity validation.

pooled-release