mb-verify
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project files that may contain instructions overriding agent behavior.
- Ingestion points: The agent reads task records (.task.json), acceptance criteria (.memory-bank/features/), and protocol plans (.protocols/plan.md).
- Capability inventory: The agent is instructed to execute shell commands (tests/CLI) and use browser automation (Playwright/CDP).
- Boundary markers: No explicit delimiters or 'ignore instructions' warnings are defined to isolate data from commands.
- Sanitization: The skill lacks content sanitization or verification of the logic inside ingested files.
- [COMMAND_EXECUTION]: The skill is explicitly designed to execute local shell commands and quality gates.
- Evidence: The instructions direct the agent to 'run the smallest meaningful check' and 'prefer deterministic checks (tests/CLI)'.
Audit Metadata