earakely-scale/ListmonkEnv-REA

listmonk

listmonk

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

Tables

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

Run it

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

listmonk on AgentEnv, from a REA capture

What this is. A replay environment of listmonk 6.2 (AGPL-3.0), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (29 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 listmonk. Tasks are graded deterministically, with no judge.

What's in it

• 29 tools (26 GET, 1 POST, 1 DELETE, 1 PUT) over 68 recorded exchanges on http://127.0.0.1:3214; no recorded group excluded.
• Path parameters: 11 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 29 tools.
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/listmonk-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_40b38ad41b4906c91a10812bd33a562c757b5bbf249f84f2f5b173e3fe59f0ba (scoped HAR sha256 db06948f2b2e2525759589dd9788f84edfa163926ae67c9c4e49c2f3b64962d0, 134,271 bytes), manifest view ev_2219c491a30c79107897a70884bbe76d8376ccb7f2c67943c7583baeea100cac, confidence inferred, authority historical-reference. The REA build is a prerelease of the environment commands proposed upstream in morluto/rea#4954, #4955, #4975 and #4999; 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_subscribers_by_id	get_subscribers_by_id	id=3	data.lists[0].description	Weekly shift notes for the lamp keepers of the Lighthouse Rota.
qa-02-get_users_by_id	get_users_by_id	id=2	data.email	ada.finch@example.test
qa-03-get_templates_by_id	get_templates_by_id	id=5	data.name	Tide Table Letter
qa-04-get_subscribers_export_by_id	get_subscribers_export_by_id	id=3	profile[0].name	Ada Finch
qa-05-get_campaigns_by_id	get_campaigns_by_id	id=2	data.subject	Shift three now starts at 22:00
qa-06-get_media_by_id	get_media_by_id	id=1	data.filename	granite-ledger-banner.png
qa-07-get_campaigns	get_campaigns	page=1, order_by=created_at, order=desc, no_body=true	data.results[0].name	Blue Heron Kitchen Spring Menu
qa-08-get_roles_lists	get_roles_lists	no arguments	data[0].name	Kitchen List Editors
qa-09-get_subscribers_by_id	get_subscribers_by_id	id=3	data.attribs.city	Granite Point
qa-10-get_users_by_id	get_users_by_id	id=2	data.user_role.name	Ledger Auditor
qa-11-get_subscribers_export_by_id	get_subscribers_export_by_id	id=3	profile[0].attribs.duty	tide-table keeper
qa-12-get_campaigns_by_id	get_campaigns_by_id	id=2	data.name	Winter Rota Briefing

Episodes

model	tasks	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	12	100%	1.0	6,997
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	8,840

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/ListmonkEnv-REA --bundle listmonk --task smoke          # Docker, no model
agent-env hf run earakely-scale/ListmonkEnv-REA --bundle listmonk --eval listmonk-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. listmonk ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (4 lists (Lighthouse Rota Dispatch, Granite Ledger Members, Blue Heron Kitchen Regulars, Saltmarsh Depot Notices), 6 subscribers with attributes (one blocklisted), 3 templates (2 campaign, 1 transactional), 3 draft campaigns, 1 user role, 1 list role, 1 extra user (ada.finch), 1 uploaded PNG, site name Lighthouse Rota Dispatch; install-time samples removed; nothing sent); 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	listmonk (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)	29 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 listmonk; 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 listmonk instance over synthetic data; listmonk is AGPL-3.0 software by Kailash Nadh and the listmonk 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.