@tangle-network/agent-eval 0.6.0 → 0.7.1

package/README.md

@@ -1,57 +1,32 @@
 # @tangle-network/agent-eval
 
-Domain-agnostic evaluation framework for Tangle agent apps. Multi-turn scenario execution, multi-judge scoring, agent-driver meta-testing, convergence tracking. Every agent (tax, legal, film, gtm) imports this to get a reproducible quality harness.
+Trace-first evaluation framework for Tangle agents. Core (spans, pipelines, sandbox harness, OTLP export), trust (dataset, red-team, calibration, behavior DSL), builder-of-builders (three-layer eval, resumable sessions, meta-runtime correlation), and frontier (meta-eval correlation study, Process Reward Modeling, bisector).
 
 ## Install
 
 ```bash
-npm install @tangle-network/agent-eval
+pnpm add @tangle-network/agent-eval
 ```
 
 ## Usage
 
-```ts
-import { BenchmarkRunner, ProductClient, defaultJudges } from '@tangle-network/agent-eval'
-
-const client = new ProductClient({
-  baseUrl: 'https://my-agent.tangle.tools',
-  routes: {
-    signup: '/api/auth/sign-up/email',
-    chat: '/api/chat',
-    // ...
-  },
-})
-
-const runner = new BenchmarkRunner(client, {
-  scenarios: myScenarios,
-  judges: defaultJudges('film production'),
-  systemPrompt: MY_SYSTEM_PROMPT,
-})
-
-const report = await runner.run()
-```
-
-## What's in the box
+**→ [`.claude/skills/agent-eval/SKILL.md`](./.claude/skills/agent-eval/SKILL.md)** — single source of truth for every usage pattern. Covers: minimal builder-of-builders path, the seven muffled-gate footguns paid for in shipped bugs, the three-layer eval contract, regression tests worth writing, and "when to use what" for the 100+ exports.
 
-- **ProductClient** — configurable HTTP client (routes are config, not code)
-- **ScenarioRegistry** — auto-discovery + filtering
-- **executeScenario** — multi-turn executor with artifact collection
-- **BenchmarkRunner** — orchestrates scenarios + judges + scoring
-- **AgentDriver** — meta-agent that plays personas against a real product
-- **MetricsCollector** — per-turn product state metrics
-- **ConvergenceTracker** — completion% over turns
-- **Reporter** — markdown + console output
-- **Judges** — 4 built-in (domain expert, code execution, coherence, adversarial) + `createCustomJudge` factory
+If you're an LLM or agent reading this, load the skill file before writing integration code — it encodes 10+ incident-driven directives that will save you from rediscovering them.
 
-## Tier
+## Dev
 
-Marketplace tier of the [agent-builder](https://github.com/drewstone/tangle-agent-builder) three-tier architecture. Uses [`@tangle-network/tcloud`](https://github.com/tangle-network/tcloud) for judge LLM calls.
+```bash
+pnpm build        # tsup
+pnpm test         # vitest
+pnpm typecheck    # tsc --noEmit
+```
 
 ## Related
 
-- [`@tangle-network/agent-gateway`](https://github.com/tangle-network/agent-gateway) — the gateway agents published through
-- [`@tangle-network/agent-client`](https://github.com/tangle-network/agent-client) — consumer SDK for those endpoints
-- [`@tangle-network/tcloud`](https://github.com/tangle-network/tcloud) — platform SDK (used internally by judges)
+- [`@tangle-network/agent-gateway`](https://github.com/tangle-network/agent-gateway)
+- [`@tangle-network/agent-client`](https://github.com/tangle-network/agent-client)
+- [`@tangle-network/tcloud`](https://github.com/tangle-network/tcloud)
 
 ## License