The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the git and gh (GitHub CLI) tools to interact with repositories. This is the primary purpose of the skill and is handled via structured workflows that include safety checks.
[DYNAMIC_EXECUTION]: The skill dynamically identifies and executes repository-specific build and test scripts (e.g., npm test, pytest, cargo test) based on the project's environment. It also executes a local validation script, scripts/validate_pr.py, which is part of the skill's own codebase.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as git commit logs, file diffs, and repository templates. This creates a potential surface for indirect prompt injection.
Ingestion points: Data enters the context via git log, git diff, gh pr view, and various project manifest files (e.g., package.json).
Boundary markers: The skill explicitly requires the use of --body-file rather than inline strings to prevent shell injection via the PR body. It does not use explicit delimiters for diff content but relies on human review.
Capability inventory: The skill can execute shell commands, file system writes, and repository-defined test suites.
Sanitization: The included scripts/validate_pr.py performs regex-based scanning for secrets (AWS keys, GitHub tokens, Slack tokens, JWTs, etc.) and unresolved placeholders (TODO, FIXME) before any PR is created.
[CREDENTIALS_UNSAFE]: While the skill interacts with sensitive file paths (e.g., .env, .ssh/id_rsa), it does so exclusively to scan for and prevent the accidental inclusion of secrets in pull requests. It includes explicit instructions to redact sensitive values and requires human confirmation for changes affecting authentication or permissions.

pr