diagnose-fix-loop
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources to drive its logic. \n
- Ingestion points: The agent reads user-provided targets (Step 1) and external artifacts processed by the diagnose skill (Step 2). \n
- Boundary markers: None explicitly mentioned in the instructions to separate external data from system prompts. \n
- Capability inventory: The skill is designed to edit local files (Step 4) and execute validation commands (Step 5). \n
- Sanitization: No explicit sanitization or filtering of external input is described. \n- [COMMAND_EXECUTION]: The skill involves running reproduction signals or measurement signals to verify fixes (Step 5). This typically requires executing local test scripts or development tools. \n- [EXTERNAL_DOWNLOADS]: The instructions suggest the user install the 'diagnose' skill from the author's repository using the npx command if it is not already present.
Audit Metadata