claude-code-headless

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

• Third-party content exposure detected (medium risk: 0.30). In references/integration-patterns.md, the Express and FastAPI web endpoints accept caller-controlled prompt/tools from req.body and pass request.prompt directly as the claude -p argument, so outsider-authored free text can be ingested at runtime into the agent.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.70). The skill exposes a --permission-mode bypassPermissions option and an acceptEdits mode (auto-accept file changes), which enable skipping permission prompts and could allow an agent to modify files on the host without interactive approval.