model-routing

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes the hermes CLI tool using Node.js's spawn with an array of arguments. This is a secure execution pattern that mitigates shell injection risks.
  • [CREDENTIALS_UNSAFE]: The routing scripts in scripts/hermes-routing.mjs and scripts/profile-store.mjs include explicit validation logic to detect and prevent the storage of API keys, tokens, or other sensitive credentials within routing profiles.
  • [DATA_EXPOSURE]: Configuration data is managed within a scoped directory (~/.agents/model-routing). The skill implements strict validation for filenames and paths using the assertName and deriveScopeKey functions to prevent directory traversal attacks.
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions emphasize 'minimum context packaging' and the use of bounded contracts for delegated tasks. By advising against forwarding full conversation histories to sub-agents, the skill limits the attack surface for indirect prompt injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 10:03 PM
Security Audit — agent-trust-hub — model-routing