The Agent Skills Directory

Indirect Prompt Injection (LOW): The agent fetches task descriptions and mentions from a local API, which serves as a potential vector for malicious instructions.
Ingestion points: lead-HEARTBEAT.md (Steps 1-4) and worker-HEARTBEAT.md (Steps 1-2) retrieve data using curl.
Boundary markers: Absent. The agent is instructed to "Review deliverables" or "Do the work described in the task" without explicit delimiters or safety warnings for the retrieved content.
Capability inventory: curl is used for network operations to localhost:8080 (GET, POST, PATCH).
Sanitization: None visible; the agent directly processes the output of the API calls.
Command Execution (SAFE): The skill uses curl to interact with http://localhost:8080. This is the primary intended purpose of the skill for task synchronization and does not involve executing remote scripts or accessing sensitive local files.

openclaw-mission-control