convex-component-adoption-planner

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

• Third-party content exposure detected (high risk: 0.85). The helper’s runtime workflow fetches public Convex documentation via curl -s https://www.convex.dev/components/components.md and curl -s https://www.convex.dev/components/<slug>/<slug>.md, which are outsider-authored free text that can be ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The helper and prompt stubs explicitly fetch live Convex docs at runtime (e.g., https://www.convex.dev/components/components.md and https://www.convex.dev/components//.md), and those remote markdown pages are injected into the CODEX_FULL_PROMPT/PLAN workflow so the fetched content directly controls agent instructions.