lfd-design
Warn
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: Phase 0 directs the agent to inventory the environment for sensitive information, specifically searching for the presence of API keys within .env files. Accessing such sensitive file paths is a significant security concern.\n- [COMMAND_EXECUTION]: The skill requires the agent to generate multiple shell scripts (score.sh, probe.sh, status.sh, lint.sh) and then execute them (Phase 6) to calibrate the harness and verify its integrity. This dynamic generation and execution of code is a high-risk capability.\n- [EXTERNAL_DOWNLOADS]: Phase 3 involves collecting data from public reference artifacts and external URLs to populate the evaluation environment. This interaction with unverified remote sources introduces potential for malicious data ingestion.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it instructs the agent to reverse-engineer external artifacts (Phase 2) to generate system specifications and test cases. Malicious instructions embedded in these external resources could influence the design of the harness.\n
- Ingestion points: Phase 3 (Collection of data from public external artifacts).\n
- Boundary markers: The skill mentions 'blinding' for the eval set, but lacks explicit markers to prevent the initial design phase from following instructions in the source data.\n
- Capability inventory: Subprocess execution for harness verification (Phase 6).\n
- Sanitization: No sanitization or validation of external artifact content prior to processing is mentioned.
Audit Metadata