The Agent Skills Directory

Environment Security: The skill includes a dedicated helper function to scrub sensitive credentials, such as OPENAI_API_KEY and GITHUB_TOKEN, from the environment before executing shell commands. This prevents accidental exposure of secrets to build tools or scripts.
Controlled Command Execution: It uses subprocess.run to manage Git operations and build tasks. These commands are executed with structured argument lists rather than shell strings, and inputs like version numbers and commit hashes are validated against strict regex patterns to prevent command injection.
Secure Dependency Resolution: The skill forces uv and make commands to use the official PyPI registry (https://pypi.org/simple) via the UV_DEFAULT_INDEX environment variable, which mitigates risks associated with dependency confusion from malicious registries.
Workflow Isolation: Release preparation occurs within a dedicated, detached Git worktree. This design ensures that the user's primary workspace remains clean and protects the main branch from accidental mutations during the release process.
Least Privilege and Scope: The skill's instructions explicitly restrict it from performing network-mutating actions, such as pushing commits or opening pull requests on GitHub, ensuring the agent stays within its intended local-only preparation scope.

release-candidate-prep