dogfood-verification

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill provides hardcoded default credentials (admin@objectos.ai / admin123) intended for use with a local database seeded during the development and testing process.
  • [COMMAND_EXECUTION]: The skill instructs the agent to perform multiple shell operations for environment setup and application management, including port checking with lsof, starting the development server with pnpm, and performing version control operations with git and gh (GitHub CLI).
  • [COMMAND_EXECUTION]: Provides a JavaScript snippet designed to be executed within the browser (via preview_eval) to override native Event.prototype methods (preventDefault and returnValue) and the window.onbeforeunload handler. This is used to bypass browser navigation locks during automated verification.
  • [EXTERNAL_DOWNLOADS]: The skill references the use of pnpm to build workspace packages, which involves fetching dependencies from the official NPM registry, a well-known and trusted service.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it instructs the agent to process data from a running web application (via preview_eval DOM queries and preview_screenshot).
  • Ingestion points: Browser DOM content and visual screenshots captured during verification (referenced in SKILL.md).
  • Boundary markers: None identified in the instructions to separate untrusted application data from agent instructions.
  • Capability inventory: The agent has capabilities to write files (git apply), execute shell commands (pnpm, git), and perform network requests (fetch, curl) as described in SKILL.md.
  • Sanitization: No specific sanitization or validation of the ingested browser content is mentioned before the agent acts upon it.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 10:00 AM