12-factor-agent-design-review

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSNO_CODE
Full Analysis
  • [SAFE]: The skill acts as a methodological framework and educational resource for engineering reliable AI agents. It does not perform sensitive operations or request high-privilege tool access.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted third-party code and architecture documents. While the instructions lack explicit data boundary markers (e.g., XML tags for user input), the methodology mandates a 'Validation loop' that requires citations and evidence for every finding, significantly reducing the likelihood that instructions embedded in the analyzed data could deceive the agent.
  • [EXTERNAL_DOWNLOADS]: The skill documentation refers to external repositories and discussions from HumanLayer, a recognized organization in the LLM reliability space. These references are used solely for methodological context and do not involve runtime script execution or untrusted downloads.
  • [NO_CODE]: The core skill consists of Markdown-based instructions and templates. The TypeScript files located in the evals/ directory are static evaluation fixtures designed to test the agent's ability to detect architectural flaws; they are not intended to be executed as part of the skill's primary functionality.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 06:53 AM
Security Audit — agent-trust-hub — 12-factor-agent-design-review