Add Hlido Agent Trust guardrail (closes #1689)

[ankitkapur1992-hlido]

Hlido Agent Trust guardrail

Implements the guardrail requested in #1689 — gate LLM/agent requests on Hlido's independent, evidence-backed trust scores for shipped AI agent products.

What it does

A beforeRequestHook guardrail (hlido.agentTrust) that, before a request runs, checks the target agent/model's Hlido trust score (0–100) against a configurable minScore (default 70) and fails the guardrail when the agent scores below it — or, optionally, when it has no independent review at all.

Why it's a little different

No credentials. Hlido's review data is public (CORS-open, no auth), so unlike the other guardrail providers this plugin needs zero API keys to configure — it just works.

Files

• plugins/hlido/manifest.json — plugin + agentTrust function metadata (slug, minScore, failOnUnreviewed)
• plugins/hlido/globals.ts — public scorecard fetch (/data/scorecards/{slug}.json, 404 = unreviewed)
• plugins/hlido/agentTrust.ts — the handler (mirrors the existing promptfoo/guard shape)
• plugins/hlido/hlido.test.ts — tests against the live public API
• plugins/index.ts — registered hlido: { agentTrust }

Verified

Tested against the live API: a reviewed agent (aider, score 92 / VITAL) passes at minScore ≤ 92 and is blocked above it; an unknown slug returns 404 → passes by default, blocks under failOnUnreviewed.

Disclosure

Opened by the Hlido team (maintainers of hlido.eu) in response to our own feature request #1689. Happy to adjust naming, parameters, or the scan shape to match your conventions — just point me at the preference.