bughunters

Pass

Audited by Gen Agent Trust Hub on Sep 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from application user interfaces, network logs, and console output, which could contain malicious instructions.
  • Ingestion points: The Explorer agent reads application screens, console errors, and failed network requests from the app under test (described in SKILL.md).
  • Boundary markers: The skill lacks explicit delimiters to distinguish application data from instructions; it uses {{NAME}} placeholders for secrets but not for content boundaries.
  • Capability inventory: The Fixer agent has the capability to write code to the file system, and the GitHub agent can open Pull Requests (SKILL.md).
  • Sanitization: Secret values are masked in logs, but there is no evidence of sanitization or filtering for instructions embedded in the processed application data.
  • [COMMAND_EXECUTION]: The skill executes numerous shell commands to manage the environment and run the testing suite.
  • Evidence: The skill uses npx bughunters@latest, git, curl, gh (GitHub CLI), maestro, xcrun (simctl), and adb (SKILL.md). It also executes user-provided app.setup and agents.fixer.verify commands.
  • [EXTERNAL_DOWNLOADS]: The skill downloads and executes external tools and binaries during setup.
  • Evidence: Fetches the bughunters CLI from npm and installs Chromium via Playwright (SKILL.md). These resources originate from the skill's vendor and Microsoft respectively.
  • [DYNAMIC_EXECUTION]: The Fixer agent generates source code that is subsequently executed during the retest and verification phases.
  • Evidence: The skill runs verification scripts (e.g., npm run typecheck && npm test) on code generated and committed by the Fixer agent (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 27, 2026, 07:39 PM