elizaos--eliza/packages/benchmarks/woobench/CLAUDE.md

WooBench — Agent Guide

Mystical reading conversation and revenue benchmark. Evaluates an agent's ability to conduct tarot, I Ching, and astrology readings across 10 persona archetypes (skeptic, true believer, emotional crisis, scammer, etc.) while correctly handling payment conversion, crisis support, and scam resistance. Registered as woobench.

Run

# Direct, from packages/benchmarks/
python -m benchmarks.woobench --model gpt-5 --output benchmark_results/

# Filter by divination system
python -m benchmarks.woobench --system tarot --model gpt-5

# Filter by persona archetype
python -m benchmarks.woobench --persona skeptic --model gpt-5

# Run a single scenario
python -m benchmarks.woobench --scenario skeptic_tarot_01 --model gpt-5

# Through the suite orchestrator
python -m benchmarks.orchestrator run --benchmarks woobench --provider <p> --model <m>

Smoke test (no API keys)

# Deterministic dummy agent + heuristic evaluator — no credentials needed
python -m benchmarks.woobench --agent dummy --evaluator heuristic --model dummy

# Dry run — lists scenarios that would be executed, no agent calls
python -m benchmarks.woobench --dry-run

# dummy-charge smoke: exercises the payment action path with a mock payment URL
python -m benchmarks.woobench --agent dummy-charge --evaluator heuristic \
    --payment-mock-url http://localhost:9999 --model dummy

Test the harness

pytest packages/benchmarks/woobench/tests/ -v

Layout

Path	Role
__main__.py	CLI entrypoint (python -m benchmarks.woobench)
runner.py	Orchestration loop (concurrency, result aggregation)
evaluator.py	Per-scenario turn driver and payment detection
scorer.py	Aggregates scenario results into BenchmarkResult
types.py	Dataclasses: Scenario, ScenarioResult, BenchmarkResult, RevenueResult
payment_actions.py	Payment action parsing and dispatch
payment_mock.py	MockPaymentClient for harness tests
personas/	One module per persona archetype
scenarios/	Tarot, I Ching, and astrology scenario definitions
tests/	pytest suite (scorer unit tests + payment mock integration)

Notes

• Results write to benchmark_results/woobench_<model>_<timestamp>.json (gitignored).
• Scored by _score_from_woobench_json in registry/scores.py; overall_score (0–100) is normalized to 0–1.
• Supported agents: eliza (default, elizaOS TS bridge), hermes, openclaw, smithers, dummy, dummy-charge.
• Evaluator modes: llm (OpenAI-compatible judge, default) and heuristic (deterministic, no credentials).
• Payment flow tested via --payment-mock-url pointing at a mock payments service; see payment_mock.py.
• Full background: README.md.

⛔ NON-NEGOTIABLE — evidence, trajectories & real end-to-end tests

The binding, repo-wide standard is AGENTS.md. Read it. Nothing in this package is done until it is proven done — a reviewer must confirm it works without reading the code, from the artifacts you attach. This applies to every feature, fix, refactor, and chore here. "Tests pass" is not proof; "CI is green" is not proof.

• Record AND read model trajectories. Capture the actual inputs and outputs of the model from a live LLM — not the deterministic proxy, not a mock: the prompt, the providers/context, the raw model output, every tool/action call, and the result. Then open the trajectory and review it by hand. A captured-but-unread trajectory is not evidence (packages/scenario-runner/bin/eliza-scenarios run <scenario> --report <out>).
• Real, full-featured E2E — no larp. Every feature ships detailed end-to-end tests that drive the real path end to end. Not the happy "front door" only: cover error paths, edge/empty/invalid input, concurrency, roles/permissions, and adversarial input. A test that asserts against a mock/stub/fixture standing in for the thing under test does not count. If the real model/device/chain/connector/account is hard to reach, make it reachable — that is the work, not an excuse to mock. If the existing tests here are shallow or mocked, fixing them is part of your change.
• Screenshots + logs at every phase, plus a complete walkthrough video/run-through of the entire feature or view, start to finish (bun run test:e2e:record).
• Manually review every artifact the change touches — never just the green check: client logs (console + network), server logs ([ClassName] …), the model trajectories in and out, before/after full-page screenshots, and the domain artifacts listed below for this package.
• No residuals. No shortcuts. The goal is not "done" — it is everything done. Clear every blocker by the hard path: build the real architecture, stand up the real model/device/service, actually test it. Never leave a TODO, a stub, a stepping-stone, or a "follow-up." When unsure, research thoroughly, weigh the options, and ship the best, highest-effort, production-ready version. Keep going until every possibility is exhausted.

Artifacts → attached inline in the PR (MP4 video, JPG screenshots, logs in <details>); attach each evidence type or explicitly mark it N/A with a reason — never leave it blank. If develop moved and changed behavior, re-capture evidence; stale proof is worse than none.

Capture & manually review for this package — benchmark / eval suite:

• A real-model run (not the mock/smoke fixture) producing the score-report JSON, with the numbers inspected and the provider/model recorded.
• The per-item trajectories the harness captured, spot-reviewed for correctness — a green harness run over mock fixtures is not a result.
• The provider matrix actually exercised, and the scoring math validated against a known case.
• Failure / timeout / partial-output handling in the harness itself.