[New Plugin] Add FangcunGuard guardrail

[zhiyuanzeng99]

[New Plugin] FangcunGuard guardrail

Closes #1736

Adds FangcunGuard as a guardrail plugin. FangcunGuard (by FangcunLeap) is a content-safety classifier that scores text into 10 safety categories — crimes, hate speech, self-harm, cybersecurity, extremism, and more — with a compact, low-latency, Chinese-optimized model.

Changes

• plugins/fangcunguard/manifest.json — plugin definition (apiKey credential, optional customHost)
• plugins/fangcunguard/scan.ts — handler supporting beforeRequestHook (scan prompt) and afterRequestHook (scan response)
• plugins/fangcunguard/globals.ts — calls POST {baseURL}/guard/context with a Bearer token
• plugins/fangcunguard/fangcunguard.test.ts — tests
• Registered in plugins/index.ts

Behavior

The gateway verdict maps directly from the model's is_safe field (verdict = is_safe): safe content passes, unsafe content is flagged. The full classification (label, unsafe_score, confidence, model) is returned in data. Empty text passes without an API call.

Testing

npx jest plugins/fangcunguard/fangcunguard.test.ts — 5 passed against the live API:

✓ should fail (verdict false) when the api key is invalid
✓ should pass (verdict true) when the prompt is safe
✓ should fail (verdict false) when the prompt is unsafe
✓ should scan the model response on afterRequestHook
✓ should pass (verdict true) on empty text

prettier --check and tsc --noEmit both pass.

Getting a key

Sign up at https://www.fangcunleap.com or contact info@fangcunleap.com.

🤖 Generated with Claude Code