diagnose
Pass
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill frequently uses shell commands for environment discovery, tool version checking, and executing reproduction logic such as tests and automation scripts (e.g.,
command -v,npm view,argent --version).- [EXTERNAL_DOWNLOADS]: The instructions recommend installing external developer tools via npm, specifically Software Mansion's@swmansion/argentand theagent-deviceutility, to facilitate mobile and browser automation.- [REMOTE_CODE_EXECUTION]: By design, the skill executes external reproduction scripts, test harnesses (Playwright, Cypress, Vitest), and automation tools to verify bug symptoms and collect diagnostic data.- [PROMPT_INJECTION]: The skill ingests untrusted data from tool outputs and logs. \n - Ingestion points: logs from
argent,agent-device, Playwright, and instrumented code; \n - Boundary markers: Use of structured logging, stable prefixes, and debug IDs; \n
- Capability inventory: Subprocess execution for reproduction and file system writes for instrumentation; \n
- Sanitization: Focuses on 'behavior-neutral' instrumentation but lacks explicit input sanitization before ingestion.
Audit Metadata