dyno-skill-fit

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

• Third-party content exposure detected (high risk: 0.70). The required workflow reads outsider-authored free text from a public web page at runtime (it asserts command.called("curl", { args: ["-fsSL","https://docs.dynobox.xyz/config-authoring.md"] })), which can be ingested into the agent/LLM context via the fetched documentation.

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

• Potentially malicious external URL detected (high risk: 1.00). The dyno scenario asserts the agent will run curl at runtime to fetch https://docs.dynobox.xyz/config-authoring.md (references/assertion-guidance.md shows the curl command and dyno/dyno-skill-fit.dyno.mjs asserts it must be called), so remote content fetched at runtime can directly influence the agent's instructions.