opencli-oneshot

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The prompt includes examples that concatenate and inject live credentials into requests (e.g., reading cookies/localStorage tokens into Authorization headers) and even contains a literal-looking Bearer token in the TS Header template, meaning an LLM could be instructed to emit secrets verbatim into generated code/commands.

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

• Third-party content exposure detected (high risk: 1.00). The skill's Step 1–3 workflow explicitly opens arbitrary target URLs (opencli browser open ), captures and inspects network responses (opencli browser network, page.getInterceptedRequests) and executes page.evaluate()/fetch() against those third‑party pages/APIs, so untrusted public content is read and used to influence adapter generation and subsequent tool behavior.