investigate-bug
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform diagnostic probes and run validation scripts (e.g.,
python3 docs/validation/investigate-bug/check.py) to confirm reported defects. These actions are constrained by a requirement for explicit authorization and are limited to diagnostic-write scopes or isolated scratch copies. - [INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data from bug reports and target instructions (
SKILL.md, Step 1). While no specific boundary markers are mandated, the instructions mitigate this risk by requiring the agent to clarify missing intent with the report owner and compare actual behavior against documented expectations rather than following report-provided triggers blindly. Sanitization is addressed through the mandatory redaction of secrets and private data from all diagnostic evidence. - [DYNAMIC_EXECUTION]: The procedure involves generating and executing reproduction scripts (probes) to establish the cause of a bug. This capability is managed through a documented lifecycle that includes recording changes, using isolated environments, and performing a mandatory cleanup of probes to ensure no unapproved product fixes or temporary changes remain.
Audit Metadata