The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands to manage state and verify project integrity.
Uses git status, git diff, and git ls-files to snapshot the initial state and generate a net report of changes.
Invokes project-specific automation tools like just (e.g., just test, just lint) to verify the success of the iterations.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface by design, as it ingests and executes instructions from external SKILL.md files.
Ingestion points: The agent resolves and reads SKILL.md files from paths or names provided in the user's invocation arguments.
Boundary markers: The skill does not explicitly define sanitization or boundary markers for the instructions retrieved from the target skill.
Capability inventory: The skill possesses the capability to inspect the repository (git) and execute project recipes (just).
Sanitization: Instructions from the target skill are followed "inline," relying on the agent's core safety guardrails when processing the sub-skill.

loop-skill