diagnose
Pass
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the AI agent to automate reproduction using various methods such as unit/integration tests, HTTP scripts (curl), CLI tools, and browser automation (Playwright/Puppeteer).
- [COMMAND_EXECUTION]: The skill references a provided shell script (
scripts/hitl-loop.template.sh) designed for interactive human-in-the-loop debugging sessions where the agent prompts the user for status updates. - [DATA_EXFILTRATION]: As part of the diagnostic process, the agent is instructed to request potentially sensitive artifacts from the user, including HTTP Archive (HAR) files, log dumps, or permission to add instrumentation to production environments.
- [PROMPT_INJECTION]: The skill involves processing external, potentially untrusted data such as application logs, user-pasted error messages, and network traces. This ingestion of external data constitutes an indirect prompt injection surface, though the risk is mitigated by the skill's structured diagnostic workflow.
Audit Metadata