The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download the coco-loops utility from the vendor's official GitHub repository at https://github.com/coco-research/coco-loops and install it locally via pip.
[COMMAND_EXECUTION]: The skill utilizes several CLI commands (coco-loops charter, coco-loops run, coco-loops status) to manage the autonomous loop lifecycle, including charter validation and monitoring.
[INDIRECT_PROMPT_INJECTION]: The skill contains an attack surface for processing untrusted data via its task management system.
Ingestion points: The loop reads instructions from a JSON task queue (queue.json) seeded by the user in Step 5.
Boundary markers: Employs a structured JSON schema for charters and requires validation against an immutable constitution floor using coco-loops charter --schema.
Capability inventory: The skill uses coco-loops run to invoke a maker agent (e.g., cursor-agent) to propose file system changes within a git worktree.
Sanitization: Enforces a propose-only level (L1) by default, requiring a separate verifier to judge proposals and preventing automatic commits or destructive operations.
[DYNAMIC_EXECUTION]: The skill facilitates an autonomous loop that dynamically generates and evaluates code proposals; however, these are strictly limited to a propose-only state requiring human review before application.

coco-loop