systematic-debugging
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- COMMAND_EXECUTION (SAFE): The skill includes a shell script (
find-polluter.sh) and diagnostic bash snippets inSKILL.md. These commands are designed to help with debugging (e.g., running tests, checking environment variables, and inspecting macOS keychain identities) and are standard for technical troubleshooting tools. - PROMPT_INJECTION (SAFE): The skill employs strong imperative language (e.g., 'The Iron Law', 'ALWAYS find root cause') and scenario-based tests to ensure the agent follows a specific methodology. This framing is benign instructional language used for methodology enforcement rather than a malicious safety bypass.
- DATA_EXPOSURE (SAFE): Diagnostic examples in
SKILL.mdsuggest inspecting environment variables (IDENTITY) and security identities (security list-keychains). These are used for local verification within a debugging context and no logic for external exfiltration is present. - INDIRECT_PROMPT_INJECTION (SAFE): The skill instructs the agent to analyze external data like error messages and logs. While this represents a data ingestion surface, the process emphasizes systematic verification and evidence gathering, which naturally mitigates risks associated with untrusted log content.
Audit Metadata