The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various Git operations (git switch, git add, git commit, git push) and host-specific CLI tools (gh or glab) to manage the shipment of code. These actions are strictly gated by a prerequisite checklist and are restricted to non-default branches.
[PROMPT_INJECTION]: The skill contains a surface for indirect prompt injection within its automated drafting logic.
Ingestion points: It reads metadata, state information, and logs from local files (.refactor-chain/state.json and .refactor-chain/audit-log.jsonl) to populate templates.
Boundary markers: The ingested data is interpolated into markdown and commit templates without explicit delimiters or escaping to prevent the execution of embedded instructions.
Capability inventory: The skill has the capability to stage changes (git add -A), commit them, and push them to a remote server.
Sanitization: The primary mitigation is a mandatory human-in-the-loop checkpoint where the agent must show all drafted messages to the user for review and approval before execution, effectively neutralizing malicious payload delivery via the state files.

refactor-ship