The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several system and development commands.
Manages macOS services using launchctl kickstart to restart the 'system-bus-worker' service.
Performs file system cleanup using rm to delete agent-generated test files in the __tests__/ directory.
Uses Git commands (git commit, git worktree remove, git branch -D) for repository maintenance.
Runs the joelclaw CLI tool (a vendor resource from 'joelhooks') for status monitoring and diagnosis.
[DATA_EXPOSURE]: The skill reads logs and execution output files from the /tmp/agent-loop/ directory. This is used to triage failures and monitor the progress of automated stories.
[INDIRECT_PROMPT_INJECTION]: The skill processes output from external files which are generated by other agent runs, creating a potential surface for indirect instruction injection.
Ingestion points: Reads log data from /tmp/agent-loop/<LOOP_ID>/<STORY_ID>-<ATTEMPT>.out.
Boundary markers: Absent; the instructions direct the agent to read the tail of the files without specifying delimiters or safety warnings for the content.
Capability inventory: Includes file deletion (rm), service management (launchctl), and repository modification (git).
Sanitization: No sanitization or validation of the log content is described before the agent is expected to act on the information.

loop-nanny