honcho

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill explicitly calls out Honcho tools (honcho_search returns raw excerpts and honcho_context runs queries on Honcho's backend) that fetch and inject user-generated/untrusted memory from the Honcho service (app.honcho.dev or a self-hosted baseUrl) into the agent's workflow, and those excerpts can materially influence decisions and tool use.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill uses the Honcho cloud endpoint (https://app.honcho.dev) at runtime to fetch memory/context and dialectic LLM outputs (via honcho tools like honcho_context) that are injected into/used to drive agent prompts, so external content can directly control the agent's instructions.