The Agent Skills Directory

Local Command Execution: The skill executes several bundled shell scripts (scripts/bump-version.sh, scripts/list-changes-since-tag.sh, scripts/verify-lint-mirror.sh) and standard developer tools such as git, gh, uv, ruff, and pylint. These operations are used to manage the project's versioning and code quality as part of the intended release process.
Indirect Prompt Injection Surface: The skill ingests data from external sources by reading Pull Request titles and bodies via the GitHub CLI to populate the changelog. This represents a security consideration where untrusted input from PR contributors could attempt to influence the agent's behavior. The skill mitigates this through a structured sanitization rubric in assets/entry-sanitizer.md and by requiring human confirmation before finalizing changes.
Network and Credential Access: The skill utilizes the gh (GitHub) CLI tool, which may access the environment's GITHUB_TOKEN to interact with repository metadata and open pull requests. This access is scoped to standard GitHub API operations required for the release workflow.
Human-in-the-Loop Guardrails: The procedure includes three explicit checkpoints where the operator must review and confirm the version bump, the sanitized changelog diff, and any linting failures. These gates ensure that automated actions remain under human supervision before being committed or pushed.

cut-release