diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external artifacts such as logs, HAR files, and core dumps which could contain malicious instructions designed to influence agent behavior.
- Ingestion points: Analyzes debug data, log dumps, and project-specific documentation (CONTEXT.md).
- Boundary markers: The skill advises manual redaction of tokens but does not specify strict boundary markers for the agent to separate instructions from analyzed data.
- Capability inventory: The workflow empowers the agent to execute shell commands, generate scripts, and make network requests via curl.
- Sanitization: No automated sanitization is specified beyond a reminder to the user to redact sensitive tokens before sharing output.
- [DYNAMIC_EXECUTION]: The skill's primary focus is the creation and execution of dynamic code, including test harnesses, CLI scripts, and headless browser scripts, to establish reproduction loops for software defects.
Audit Metadata