The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process construction project data from various file formats (CSV, Excel, JSON). This ingestion of untrusted external content represents a surface for indirect prompt injection where instructions could be hidden inside data fields.
Ingestion points: External data files and user parameters specified in instructions.md.
Boundary markers: The instructions do not define specific delimiters to separate user data from system instructions, though it recommends industry standards.
Capability inventory: The skill requests filesystem permissions in claw.json and uses the python3 binary for processing logic.
Sanitization: The instructions.md file explicitly directs the agent to 'Validate inputs before processing', which is a necessary security measure for skills handling external data.

compliance-tracker