caveman-setup

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill receives the CAVE_API_KEY in its prompt and instructs the agent to place that key into the repo's env file and to perform a real verification request with it — even though it advises using env vars and not printing the key, the workflow requires the secret value to be handled verbatim (written into files or used in outgoing requests), which creates exfiltration risk.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs making real runtime LLM requests through the Caveman gateway (example base URL https://gateway.caveman.so and the concrete example https://gateway.caveman.so/w/support-bot/openai/v1), so that external endpoint will be contacted at runtime and will control the model requests/prompts.