architecture-brainstorming

Warn

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to recommend or perform installations from an untrusted third-party source using npx skills@latest add mattpocock/skills. This involves fetching external code from a source not recognized as a trusted vendor.
  • [REMOTE_CODE_EXECUTION]: The instruction to use npx to add new skills involves downloading and executing packages from the NPM registry at runtime, which is a vector for remote code execution.
  • [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and process potentially untrusted data from the user's repository, such as guidance documents, contracts, and tests, which could be used to manipulate the agent's architectural reasoning or proposals.
  • Ingestion points: SKILL.md specifies reading repository guidance, entry points, contracts, domain vocabulary, and tests during the 'Ground the session' phase.
  • Boundary markers: None identified in the skill instructions to separate repository content from the agent's system instructions.
  • Capability inventory: The skill generates architectural designs, HTTP contracts, and flow sketches, and can recommend the installation of further external tools based on its analysis.
  • Sanitization: There are no instructions for sanitizing or validating the content read from repository files before it is processed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 6, 2026, 05:08 PM