The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection through the ingestion of untrusted construction data.\n
Ingestion points: The map_field and map_schema methods in SKILL.md ingest strings (field names and values) from user-provided construction project schemas.\n
Boundary markers: The skill logic does not utilize explicit delimiters or 'ignore embedded instructions' warnings when processing data during ontology mapping.\n
Capability inventory: According to claw.json, the skill is granted filesystem permissions, which allow for reading and writing data on the host system.\n
Sanitization: While instructions.md directs the agent to 'Validate inputs before processing', the reference implementation in SKILL.md lacks programmatic sanitization or escaping mechanisms for the interpolated data strings.

ontology-mapper