diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data sources, including HAR files, network traces, and log dumps, to diagnose bugs. It also captures arbitrary user input through a terminal-based interaction script (scripts/hitl-loop.template.sh).
- Ingestion points: Processes log dumps, core dumps, network artifacts, and direct user input via terminal prompts.
- Boundary markers: The skill mandates the redaction of secrets () but does not specify delimiters to prevent the agent from following instructions embedded within the debug data.
- Capability inventory: The agent is authorized to use curl, execute CLI commands, drive headless browsers, and use debuggers/REPLs to test hypotheses.
- Sanitization: The instructions explicitly require redacting authentication headers and secrets from captured artifacts.
- [DYNAMIC_EXECUTION]: The skill encourages the agent to create and run various types of feedback loops, such as unit tests, HTTP scripts, and browser automation.
- Evidence: Phase 1 and 6 involve writing new test code or modifying template scripts (scripts/hitl-loop.template.sh) and then executing them via shell commands (bash, node, python).
Audit Metadata