zj-diagnose
Pass
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a local bash script
scripts/hitl-loop.template.shto facilitate human-in-the-loop debugging by prompting the user for input and status updates. - [PROMPT_INJECTION]: The skill ingests untrusted external data such as logs and traces which creates a surface for indirect prompt injection.
- Ingestion points: User-provided artifacts (HAR files, log dumps, traces) and captured terminal input (SKILL.md, scripts/hitl-loop.template.sh).
- Boundary markers: None identified.
- Capability inventory: File system access, CLI tool execution (curl, git, playwright/puppeteer), and local script execution.
- Sanitization: No explicit sanitization or filtering of external content is documented.
Audit Metadata