triage-ml-task
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill uses a mapping table to translate user natural language intent into skill-loading actions, creating a surface for indirect prompt injection.\n- Ingestion points: The skill ingests user task descriptions (intent) from the prompt and project status data from the output of the
python -m skore_skills statuscommand.\n- Boundary markers: The skill uses a structured mapping table to constrain routing, but lacks explicit instructions to the agent to ignore or delimit potentially malicious instructions embedded within the user's task description.\n- Capability inventory: The skill executes a vendor-specific CLI tool (python -m skore_skills status) and can trigger the loading of any other installed skill in the catalog.\n- Sanitization: There is no mention of input validation, escaping, or sanitization for the user's intent string before it is processed for routing.
Audit Metadata