earakely-scale/PinryEnv-REA

pinry

pinry

An AgentEnv bundle with its recorded runs, published with agentenv-hf.

Tables

• pinry_tasks: one row per task of the bundle, with its prompt, envs, verifiers and steps.
• pinry_episodes: one row per recorded run, with its status, model, reward and scores, the prompt and response, and the agent's transcript as chat messages when its trajectory format is one agentenv-hf reads.

raw/pinry.jsonl holds each run's record and native trajectory, and bundles/pinry/ the bundle itself.

Run it

pip install agentenv-framework huggingface_hub
hf download earakely-scale/PinryEnv-REA --repo-type dataset --local-dir agentenv-dataset
agent-env run agentenv-dataset/bundles/pinry

Pinry on AgentEnv, from a REA capture

What this is. A replay environment of Pinry 2.1.13 (BSD-2-Clause), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (12 tools), and rea environment-export wrote the bundle you can run from this dataset. Every tool call replays a retained record from the capture, cites the record and the Evidence it came from, and nothing executes Pinry. Tasks are graded deterministically, with no judge.

What's in it

• 12 tools (9 GET, 1 POST, 1 DELETE, 1 PATCH) over 568 recorded exchanges on http://127.0.0.1:3403; 38 recorded groups excluded (4 as other origins, 34 as UI pages or assets).
• Path parameters: 5 numeric. REA's four format rules (numeric, uuid, hex, iso-date) applied unconditionally; the three opt-in rules (after-uid-literal, opaque-token, sibling-variance) were requested and are labelled per parameter. With the four format rules alone the manifest has 12 tools.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 6 retrieval tasks (evals/pinry-retrieval.toml): each asks one factual question about a seeded object, with a distinct answer that the app does not ship by default, answerable through a recorded exchange; reward 1 if the response contains the recorded value, else 0. Plus smoke, which deploys the environment and checks replay through the MCP endpoint.
• 12 recorded episodes by openai/gpt-oss-120b:groq, zai-org/GLM-5.3-Flash:zai-org through the Hugging Face router, with their chat transcripts.
• Provenance: capture ev_0878c47a60e7fec405c266932194cb3b41b645d635c765b4593c2c6e5d382ac5 (a capture_browser_scenario session REA recorded itself; capture file sha256 921645f30e35acb43127cb5cc01dd241da871bfdefefb333be9d58cc75f42951, 2,303,521 bytes), manifest view ev_3953f182d0e7f30e5531f5d130571e19971d958e79ae1655db5a82879e766c44, confidence inferred, authority controlled-replay. The REA build is a prerelease of the environment commands proposed upstream in morluto/rea#4954, #4955, #4975, #4999 and #5182; the bundle installs it from the tarball it ships, so the environment runs without waiting for an npm release.

What replay means, and does not

A tool call returns the recorded response whose path and query values best match the arguments, with replay.exact_match telling you whether it matched exactly or fell back to the closest record. Reads of seeded objects behave like the application. Writes return the recorded answer of the one write the session made, and the environment has no state: a created object does not appear in a later list. This is what makes retrieval tasks hard-gradable and write tasks weak; the manifest's unknowns say so.

Tasks and grading

task	tool	arguments	field	answer
qa-01-get_pins_by_id_2	get_pins_by_id_2	id=1	description	Signal card LR-0417: amber over teal, eleven horizontal stripes
qa-02-get_boards_by_id	get_boards_by_id	id=1	name	Lighthouse Rota Signal Colours
qa-03-get_boards_auto_complete	get_boards_auto_complete	submitter__username=ada-finch	[0].name	Copperfield Annex Window Studies
qa-04-get_tags_auto_complete	get_tags_auto_complete	no arguments	[0].name	lamp-room
qa-05-get_pins	get_pins	format=json, ordering=-id, limit=30, offset=0	results[0].description	Window study CA-09: arched pane, peach sky fading to lemon
qa-06-get_boards	get_boards	submitter__username=ada-finch, offset=0, limit=50	results[0].cover.description	Fog warning swatch LR-0431: grey fading to white, top to bottom

Episodes

model	tasks	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	6	6	100%	1.2	5,541
zai-org/GLM-5.3-Flash:zai-org	6	6	100%	1.2	6,787

Every episode is in the episodes config with its transcript; rewards are the deterministic verifier scores.

Every recorded episode passed.

Run it yourself

uv tool install agentenv-framework --with 'agentenv-hf @ git+https://github.com/earakely-scale/agentenv-hf-plugin@v0.3.0'
agent-env hf run earakely-scale/PinryEnv-REA --bundle pinry --task smoke          # Docker, no model
agent-env hf run earakely-scale/PinryEnv-REA --bundle pinry --eval pinry-retrieval --model openai/gpt-oss-120b:groq

The retrieval eval needs a model endpoint: set [model] base_url and api_key in .agentenv/config.toml (the Hugging Face router works with any hf_ token). The bundle's agents/replay-qa is a small OpenAI-compatible tool-calling agent; any A2A agent can replace it.

How it was made

1. Pinry ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (accounts: ada-finch ada.finch@example.test, rui-castellanos rui.castellanos@example.test, mina-okafor mina.okafor@example.test, tomas-vey tomas.vey@example.test, rea-admin rea-admin@example.test (capture account; owns nothing seeded); boards: Lighthouse Rota Signal Colours (ada-finch; 3 pins), Granite Ledger Core Samples (rui-castellanos; 3 pins), Blue Heron Kitchen Tile Glazes (mina-okafor; 3 pins), Saltmarsh Depot Crate Stencils (tomas-vey; 3 pins), Copperfield Annex Window Studies (ada-finch; 3 pins, one shared with the signal board); pins: 14; pin codes: LR-0417, LR-0422,...); people and projects are invented.
2. REA's capture-browser-scenario launched Chromium and drove the app's own UI, with any credential given as a declared secret; it recorded every network transaction with credential headers and declared secrets redacted, and the environment-manifest view read that capture directly, with no HAR or scrubbing step.
3. rea inspect-analysis-view projected the environment-manifest view; rea environment-export wrote this bundle; gen_tasks.py derived the retrieval tasks from the manifest and the recorded bodies; agent-env run recorded the episodes; agent-env hf publish wrote this dataset. Scripts: https://github.com/earakely-scale/rea-environments.

Compared with PortSimEnv on AgentEnv

	PortSimEnv-AgentEnv	Pinry (this dataset)
Environment	A simulation: a container quay, ships, cranes, weather, with A2A ship agents	A replay of a real application's API, captured from one session
Observation	Tool results computed by the simulator as the week unfolds	Recorded responses, cited to Evidence
Action space	Planning tools (check_plan, confirm_berths, advance, message_agent)	12 tools mirroring the application's endpoints
State	Full: a virtual clock, triggers, frozen windows, agents that answer	None: every call is independent; writes replay one recorded answer
Reward	Deterministic, against a CP-SAT optimum or a re-planner anchor; continuous 0 to 1	Deterministic, recorded-value match; 0 or 1 per task
What it measures	Planning under uncertainty and negotiation	Tool discovery, parameter inference and faithful retrieval in a real API
Data provenance	Real 2024 port calls and weather; simulated workloads	Synthetic seed data on a self-hosted Pinry; nothing real
How it was built	Hand-written plugin, months of work	Generated from a capture in one scripted pass
Where it is strong	Long-horizon decisions with hard rewards	Breadth: any app you can capture becomes an environment

The two are complementary: PortSimEnv shows what a hand-built simulation can grade; this dataset shows how far a capture alone gets you, and the manifest's unknowns list exactly what a stateful version would still have to specify.

Licence and credits

Bundle code and tasks: Apache-2.0. Recorded content: responses generated by a self-hosted Pinry instance over synthetic data; Pinry is BSD-2-Clause software by the Pinry contributors and is not redistributed here beyond the recorded JSON responses. REA is MIT software by morluto; AgentEnv Framework is Apache-2.0 by Scale AI.