turnstile-spin

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill explicitly permits and instructs the user to "paste in chat" API tokens and also tells the agent to parse a returned "secret" from CLI/JSON output, so the LLM may receive and could be required to handle or reproduce secret values verbatim (high exfiltration risk).

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

• Potentially malicious external URL detected (high risk: 0.90). scripts/persist-skill.sh invokes npx degit to fetch the skill bundle from the git repo reference cloudflare/skills/skills/turnstile-spin at runtime, which pulls remote content (SKILL.md + scripts) that will control agent instructions and behavior.