[New Plugin] NeuralTrust TrustGuard

[albertbausili]

Summary

• Adds the neuraltrust.evaluate guardrail (issue [Feature] NeuralTrust TrustGuard #1771) so Portkey can send chat, completion, and Anthropic Messages traffic to TrustGuard POST /v1/evaluate.
• Maps TrustGuard allow / block / report / transform onto Portkey verdicts. Transform writes back messages, completion prompt, or Anthropic system + tool_use. Block with hook deny: true is HTTP 446.
• Fails closed by default (verdict: false, error: null) so an unreachable TrustGuard cannot be forgiven. unreachableFallback: fail_open restores house style for connect / timeout / 502 / 504 only.
• Registers the handler in plugins/index.ts and lists neuraltrust in plugins_enabled. Credentials stay on the check (apiKey, optional collectorKey / apiBase).

Description:

• New partner plugin: plugins/neuraltrust/ (manifest, handler, payload codec, tests, README)
• Additive registration in plugins/index.ts
• neuraltrust added to conf.json / conf.example.json plugins_enabled

Tests Run/Test cases added:

• npx jest plugins/neuraltrust — 60 mocked tests (payload, transforms, fail-closed matrix, HTTP errors) plus optional live suite if .creds.json is present
• npx prettier --check on the plugin files

Type of Change:

• New feature (non-breaking change which adds functionality)

Test plan

• npx jest plugins/neuraltrust (60/60, including the optional live TrustGuard call)
• Before-request deny: true + TrustGuard block returns HTTP 446 and does not call the provider
• allow / report still reach the model; transform redacts chat messages and completion prompt
• Anthropic system and tool_use are preserved on extract/apply (unit coverage; not a live Anthropic provider run)
• Missing apiKey and unreachable TrustGuard fail closed unless unreachableFallback is fail_open