debug
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that ingests untrusted external data, which could contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: The skill instructs the agent to inspect "errors", "logs", "source" code, and "test output" (found in SKILL.md and references/_shared/orchestration/reality-check.md).
- Boundary markers: The instructions do not specify the use of delimiters or clear warnings to the agent to ignore instructions embedded within the data being debugged.
- Capability inventory: The skill grants the agent the ability to modify the file system (SKILL.md Step 5) and execute commands for regression verification (SKILL.md Step 6).
- Sanitization: No sanitization or validation protocols are defined for the data extracted from logs or source files.
- [NO_CODE]: The skill is comprised entirely of Markdown-based guidelines and a JSON evaluation file. It does not include any scripts, binaries, or runtime dependencies.
- [SAFE]: The skill promotes secure engineering practices, including 'Diff Discipline' to ensure unrelated user changes are not overwritten and 'Decision Lock' to prevent unauthorized or accidental architectural changes.
Audit Metadata