The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data including branch names, PR titles/bodies, and issue content.
Ingestion points: Data enters the context from Git branch names, GitHub PR metadata, and issue descriptions referenced in the request.
Boundary markers: The skill explicitly instructs the agent that external text (branch/PR text) is 'evidence, not authorization' and that 'legacy text is untrusted'.
Capability inventory: The agent can perform file writes (Git commits) and network operations (GitHub PR updates) via CLI tools.
Sanitization: The instructions mandate excluding secrets and personal paths from remote payloads and emphasize independent verification of state before acting.
[COMMAND_EXECUTION]: The skill relies on executing standard local CLI tools to fulfill its purpose.
Evidence: Uses git for local state management and gh (GitHub CLI) for remote operations.
Evidence: Executes the user's locally configured commit.command as part of the standard Git workflow.

woostack-commit