pr-test

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly reads user-generated GitHub PR content (e.g., "gh pr view {N} --json body" in Step 1 and Step 8) and uses that PR description to build test plans, drive test actions, and decide/submit reviews, so untrusted third-party input can materially influence tool behavior.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill includes a runtime fallback to use OpenRouter as the copilot LLM (CHAT_BASE_URL=https://openrouter.ai/api/v1), which the backend will call at runtime and whose model outputs directly control agent prompts/responses, so this external URL is a runtime dependency that can control agent behavior.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). This skill instructs the agent to perform many direct state-changing operations (docker compose up/down, docker exec into Redis and DB, psql queries, writing .env files, copying host OAuth credentials from keychain, and optionally installing packages), which modify system state and access sensitive host secrets — so it should be flagged as high-risk even though it doesn't explicitly request sudo.