metergraph-investigate

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The skill instructs the agent to install an npm package (npx --yes metergraph-cli@latest) from an external registry using an arbitrary version identifier, which constitutes an unverifiable runtime external dependency under weak provenance.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (medium risk: 0.30). The skill documents operational instructions for investigating LLM traces, usage spikes, and costs via the metergraph CLI and MCP tools. It mentions that metergraph_replay_trace ("agent:replay") sends a captured trace to a model provider and can spend money (SKILL.md:214-215). This represents a direct financial operation (trades/swaps/purchases/settlement authorization via replay calls). However, the skill documents qualifying independent access control and approval mechanisms, including explicit user consent/approval requirements and CLI flags (--yes) enforced by the tool execution layer (SKILL.md:47-51, SKILL.md:211-216). Therefore, the severity is medium.