khe-brain-benji
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes a local Node.js script (
scripts/brain-check.mjs) that the agent is instructed to run to verify the accuracy, formatting, and citation compliance of its own responses. This script functions as a local linter and does not execute untrusted code.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to refresh its internal knowledge base by fetching and reading content from official sources, specificallyhttps://benji.org/andhttps://github.com/benjitaylor/skills. These are targeted at the well-known sources relevant to the skill's primary purpose.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes text from a provided corpus, presenting a surface for indirect prompt injection.\n - Ingestion points: The agent reads and processes markdown files within the
references/directory.\n - Boundary markers: The skill does not implement explicit boundary delimiters for the ingested text.\n
- Capability inventory: The agent has the capability to execute a local Node.js script (
scripts/brain-check.mjs) on files it generates.\n - Sanitization: The skill enforces strict sanitization through hard rules requiring paraphrasing, limited quotes, and mandatory citations, which are verified by the automated script.\n- [SAFE]: The skill demonstrates defensive design, prioritizing citation accuracy and silence when information is unavailable. It explicitly forbids building code or making speculative claims, directing users to other specialized tools for those tasks.
Audit Metadata