evolver

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the @evomap/evolver package from the official npm registry. As this is the official package of the vendor (evomap), this download is considered a standard operational dependency.
  • [COMMAND_EXECUTION]: The skill executes local commands for environment diagnostics (node --version, git --version) and tool operation (evolver). It also utilizes npm install -g for setup, but includes a specific security warning to avoid using sudo for global installations, suggesting user-level prefixes instead.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect injection surface as it ingests evolution assets (Genes, Capsules, and Recipes) from the EvoMap Hub and local workspace memory.
  • Ingestion points: Untrusted or external data enters the agent context through the evolver_recipe_search, evolver_search_assets, and evolver_poll MCP tools.
  • Boundary markers: The skill provides explicit instructions to the agent to treat output as prompts only and mandates that Evolver output should not be applied as source changes without explicit user approval.
  • Capability inventory: The agent can perform network operations via a local proxy (127.0.0.1:19820), install npm packages, and execute the evolver CLI tool.
  • Sanitization: The instructions contain specific prohibitions against printing sensitive configuration data, including bearer tokens, node secrets, and API keys from ~/.evolver/settings.json and .env files.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:07 AM