The Agent Skills Directory

[SAFE]: Secret Management: The instructions explicitly prohibit the model from printing or storing sensitive tokens (like PYPI_API_TOKEN or UV_PUBLISH_TOKEN) in chat or code, directing users to secure platform features like GitHub Secrets.
[SAFE]: CI/CD Security: The skill promotes best practices by recommending PyPI Trusted Publishing (OIDC) and requiring that external GitHub Actions be pinned to reviewed full commit SHAs.
[SAFE]: Operational Integrity: Comprehensive pre-flight checks ensure the Git environment is clean and synchronized before release, and a mandatory 'Release Plan and Approval' step ensures no mutations occur without explicit human verification.
[INDIRECT_PROMPT_INJECTION]: The skill ingests commit messages to generate changelog entries (SKILL.md), which represents a surface for indirect instructions. This is mitigated by a mandatory 'Release plan and approval' boundary marker and a verification step requiring the agent to re-read and validate the changelog content. Capabilities including network publication and file writes are guarded by these approval checkpoints.

make-release