The Agent Skills Directory

[Environment Variable Sanitization]: The skill implements a security best practice by explicitly unsetting sensitive environment variables like OPENAI_API_KEY, GITHUB_TOKEN, and GH_TOKEN before executing child processes. This prevents accidental credential leakage during the release preparation process.
[Strict Input Validation]: The release version supplied by the user is validated against a rigorous regular expression (VERSION_PATTERN) in the helper script. This ensures that only well-formed semantic versioning strings are processed, mitigating risks associated with malformed or malicious input.
[Command Execution via Subprocess]: The skill uses subprocess.run to automate Git operations and build tasks (e.g., make sync, uv run). The commands are executed with shell=False (implicit in the list-based argument passing), which prevents shell injection vulnerabilities by ensuring arguments are not interpreted by a command shell.
[Isolated Worktree Materialization]: Release candidate materialization occurs within a dedicated, detached Git worktree. This isolation strategy protects the main source checkout from unintended modifications and ensures that all readiness gates are performed on a clean, reviewed state.
[Restricted Network Operations]: Network activity is restricted to fetching from origin and accessing the official Python Package Index (pypi.org) for dependency synchronization. The skill explicitly forbids mutating GitHub state (e.g., pushing branches or creating PRs), maintaining a read-only posture towards remote repositories.

release-candidate-prep