diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions direct the agent to build and execute feedback loops using various CLI tools, HTTP scripts, and specialized testing harnesses to reproduce bugs.
- [DYNAMIC_EXECUTION]: The agent is encouraged to generate and run throwaway harnesses and custom scripts (e.g., modifying
scripts/hitl-loop.template.sh) as part of the phase-based diagnosis process. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted data such as network traces, log dumps, and HAR files which could contain malicious instructions.
- Ingestion points: Phase 1 and the fallback loops ingest external event logs, network payloads, and captured artifacts (e.g., log dumps, core dumps).
- Boundary markers: The instructions do not define specific delimiters for separating artifact content from the agent's logic, though they emphasize manual review of redacted output.
- Capability inventory: The agent has the capability to write and execute bash scripts, run CLI tools, and perform network requests (via curl/HTTP scripts).
- Sanitization: The skill mandates the redaction of secrets (API keys, auth headers) from output artifacts but does not explicitly describe sanitization of the input artifacts before processing.
Audit Metadata