earakely-scale/TandoorRecipesEnv-REA

tandoor

tandoor

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

Tables

• tandoor_tasks: one row per task of the bundle, with its prompt, envs, verifiers and steps.
• tandoor_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/tandoor.jsonl holds each run's record and native trajectory, and bundles/tandoor/ the bundle itself.

Run it

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

Tandoor Recipes on AgentEnv, from a REA capture

What this is. A replay environment of Tandoor Recipes 2.6.15 (AGPL-3.0 with Commons Clause), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (36 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 Tandoor Recipes. Tasks are graded deterministically, with no judge.

What's in it

• 36 tools (32 GET, 2 POST, 1 DELETE, 1 PATCH) over 130 recorded exchanges on http://127.0.0.1:3505; no recorded group excluded.
• Path parameters: 15 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 36 tools.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/tandoor-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.
• 24 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_be15a45685ef07f05c6b78885936293e3d2d94956f1aca0b3ac8d7a1323b69bb (scoped HAR sha256 23755079b8b61c6fee1f7430fb103e5f93e10f97a43e25badcfbe618a0f7f91b, 287,781 bytes), manifest view ev_6ff8d7d19cdde1e4bcb4a2732f7526d35c925dc217536b14209af7e38ec8606a, confidence inferred, authority historical-reference. The REA build is rea-agents 6.7.0 from main at commit 3fca2e0c, which merged the environment commands (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_recipe_by_id	get_recipe_by_id	id=2	keywords[0].description	Suppers timed to the spring tide at Blue Heron Kitchen.
qa-02-get_shopping_list_by_id	get_shopping_list_by_id	id=2	description	Dry stores for the quarry crews, signed off by Rui Castellanos.
qa-03-get_shopping_list_entry_by_id	get_shopping_list_entry_by_id	id=2	shopping_lists[0].name	Saltmarsh Depot Run
qa-04-get_keyword_by_id	get_keyword_by_id	id=4	label	Saltmarsh Forage
qa-05-get_meal_plan_by_id	get_meal_plan_by_id	id=2	title	Quarry count lunch
qa-06-get_recipe_book_by_id	get_recipe_book_by_id	id=1	description	Everything the night shift may cook between 22:00 and 06:00.
qa-07-get_cook_log_by_id	get_cook_log_by_id	id=2	comment	Too much ginger for the 02:00 watch; cut to two thumbs next time.
qa-08-get_supermarket_by_id	get_supermarket_by_id	id=1	name	Saltmarsh Depot Market
qa-09-get_food_by_id	get_food_by_id	id=7	name	Copperfield Ginger
qa-10-get_keyword	get_keyword	page=1, page_size=10	results[0].description	Small bites for the Lighthouse Rota between 22:00 and 06:00.
qa-11-get_supermarket	get_supermarket	no arguments	results[0].description	Stalls open at the 05:30 low tide and run north to south along the sea wall.
qa-12-get_recipe_book	get_recipe_book	no arguments	results[0].name	Lighthouse Rota Night Book

Episodes

model	tasks	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	12	100%	1.0	10,785
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	13,048

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/TandoorRecipesEnv-REA --bundle tandoor --task smoke          # Docker, no model
agent-env hf run earakely-scale/TandoorRecipesEnv-REA --bundle tandoor --eval tandoor-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. Tandoor Recipes ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (space: Blue Heron Kitchen (renamed from the auto-created space; message about the 05:30 tide-day opening); keywords: Tide Table Suppers, Quarry Lunches, Night Watch Snacks, Saltmarsh Forage; units: ladle, crate, quarry tin; foods: Marsh Samphire, Smoked Kelp, Sea Beet, Granite Rolled Oats, Heron Pearl Barley, Brown Lentils, Copperfield Ginger, Lantern Honey; supermarket: Saltmarsh Depot Market with categories Tidal Greens and Dry Stores; recipes: Saltmarsh Samphire Fritters, Kelp-Smoked Heron Barley Risotto, Granite Ledger Oatcakes, Lamp-Room Ginger Cordial, Copperfield Annex Lentil Pie,...); people and projects are invented.
2. Playwright drove a login, the main pages and a scripted read, write, read sequence through the app's JSON API; the HAR was scoped to the API, the login exchange dropped, secret-looking body fields scrubbed, headers and cookies removed.
3. rea inspect-web-network-capture retained the exchanges as Evidence; 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	Tandoor Recipes (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)	36 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 Tandoor Recipes; 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 Tandoor Recipes instance over synthetic data; Tandoor Recipes is AGPL-3.0 with Commons Clause software by the Tandoor Recipes authors 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.