systematic-debugging
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill mandates processing untrusted data from error messages and logs, creating an indirect injection surface. Ingestion points: Error messages, stack traces, and component logs in Phase 1 (Steps 1, 4, 5). Boundary markers: Absent. Capability inventory: File system access and shell execution (via find-polluter.sh and Phase 1 diagnostics). Sanitization: Absent.
- [DATA_EXPOSURE]: Diagnostic templates in SKILL.md Phase 1 Step 4 encourage logging environment variables and security identities. Evidence: 'env | grep IDENTITY' and 'security list-keychains'. This pattern can inadvertently leak credentials or configuration secrets into the agent's output.
- [COMMAND_EXECUTION]: The utility script 'find-polluter.sh' executes 'npm test' on files located by the 'find' command. Evidence: find-polluter.sh line 49. In an untrusted environment, this could trigger execution of malicious test code.
Audit Metadata