The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses bash commands to manage the lifecycle of a loop, including file system checks and system service management.
Evidence: The script sources local helper libraries (launchd-lib.sh, state-lib.sh, and registry-lib.sh) located within the vendor's plugin directory to perform unregistration and cleanup.
Evidence: It uses rm -rf to delete the loop's state directory after the stop operation, which includes a safety check mentioned in the documentation to ensure it only operates within the user's home directory.
Evidence: It generates and suggests a git commit command to document the termination of the loop contract.
[PROMPT_INJECTION]: The skill exhibits a standard surface for indirect prompt injection by processing external data.
Ingestion points: The skill reads from LOOP_CONTRACT.md (Step 1 and Step 3) to generate a final state summary.
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potentially malicious content within the contract file.
Capability inventory: The skill has access to bash, file editing, and tool searching capabilities.
Sanitization: No explicit sanitization or validation of the contract's content is described before interpolation into the DONE section or git commit message.

stop