bug-diagnosis

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is designed to guide the agent through executing shell commands and scripts, such as running test suites (e.g., pytest) and automation tools like git bisect, as part of the standard debugging workflow.
  • [DYNAMIC_EXECUTION]: The skill provides a shell script template (assets/repro-script-template.sh) intended for the agent to customize and execute. This runtime generation and execution of scripts is a core feature for bug reproduction but constitutes a dynamic execution pattern.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from debug logs and structured diagnosis logs (JSON), which could theoretically contain malicious instructions aimed at influencing the agent's behavior during automated loops.
  • Ingestion points: Processes user-provided JSON diagnosis logs and plain-text debug logs via scripts like hypothesis_log.py and redact_scan.py.
  • Boundary markers: The skill relies on standard file structures and does not implement specialized delimiters or 'ignore instructions' markers for the contents of the logs being analyzed.
  • Capability inventory: The agent has the capability to execute arbitrary shell commands and scripts as defined in the debugging workflows.
  • Sanitization: The skill includes a proactive redaction script (redact_scan.py) that filters for credentials and PII, but it does not perform content sanitization to prevent the interpretation of natural language instructions (prompt injection) within the data.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 03:14 PM