systematic-debugging
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to 'Read Error Messages Carefully' in Phase 1, establishing an ingestion point for potentially untrusted data from compiler outputs, runtime errors, or external service responses which may contain embedded instructions. * Ingestion points: Phase 1, Step 1 ('Read Error Messages Carefully') in SKILL.md. * Boundary markers: Not present. * Capability inventory: Shell command execution via test runners (npm, bun) and diagnostic tools (security, codesign). * Sanitization: Not present.
- [COMMAND_EXECUTION]: The framework suggests executing diagnostic commands in Phase 1, Step 4 to debug multi-component systems. Examples provided include dumping environment variables filtered by keyword ('env | grep IDENTITY') and querying system keychains ('security list-keychains', 'security find-identity'). These actions access sensitive environmental state and cryptographic metadata as part of the troubleshooting process.
Audit Metadata