earakely-scale/GhostEnv-REA

ghost

ghost

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

Tables

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

Run it

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

Ghost on AgentEnv, from a REA capture

What this is. An environment of Ghost 5.130 (MIT), built from one recorded browser session against a self-hosted instance seeded with synthetic data. REA projected the capture into an environment manifest (38 tools) and an environment state, the entities the capture's reads showed, and rea environment-export wrote the bundle you can run from this dataset. Reads the state can render are answered from it, write tools whose effect the capture confirmed change it, and data/reset restores it. Nothing executes Ghost. Tasks are graded deterministically, with no judge.

v0.2.0 rebuilds this environment from a new capture of an instance seeded the same way, recorded so each write's effect is shown: two scratch objects per write target, every write read back at once. A REA build that serves a state, confirmed writes and list queries projects it; the tag v0.1.1 keeps the earlier replay-only version.

What's in it

• 38 tools (34 GET, 2 POST, 1 DELETE, 1 PUT) over 98 recorded exchanges on http://127.0.0.1:3105; no recorded group excluded. 36 are offered; the other 2 are write tools the environment cannot apply (post_ghost_admin_posts, put_ghost_admin_posts_by_uid), left out rather than faked.
• Path parameters: 5 opaque-token. REA's four format rules (numeric, uuid, hex, iso-date) applied unconditionally; the four opt-in rules (after-uid-literal, opaque-token, sibling-variance, prefix-variance) were requested and are labelled per parameter. With the four format rules alone the manifest has 46 tools.
• State: 143 entities of 19 types. 32 read tools answer from it and 3 replay their recorded responses. 7 list tools also answer filters, search, sorting or paging from it. Served write tools: delete_ghost_admin_posts_by_uid (delete).
• Unknowns the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools.
• 12 retrieval tasks (evals/ghost-retrieval.toml): each asks one factual question about a seeded object, with a distinct answer that the app does not ship by default; reward 1 if the response contains the recorded value, else 0. Plus smoke, which deploys the environment and checks it through the MCP endpoint.
• 2 write tasks (evals/ghost-writes.toml): each asks for a change in plain language and is graded from the environment's state through data/get, not from the agent's reply; reward 1 only if every check passes. Their grade is in scores.state; reward holds the retrieval verifier's score, so it is empty for them, as for smoke.
• 24 retrieval episodes and 4 write episodes, plus the smoke run, 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_cdaa83889577a573b873a359053c9f0047a94e90bbf0070a067bb4f5e771af98 (scoped HAR sha256 75b2f8126b1a08c7228ab27b3c9ffe36525af0b090edc2a301aea6e731d13e2a, 706,821 bytes), manifest view ev_b77610bf142edddad7d0775499a6fc0b641565b0f693a24e1b8c113d911553c8, confidence inferred, authority historical-reference. The REA build is rea-agents 6.7.0 from main 3e7e102a with open pull requests merged: morluto/rea#6385 at 342933ea and morluto/rea#6897 at 08c2532a. The bundle installs it from the tarball it ships, so the environment runs without waiting for an npm release.

What the state means, and does not

Reads the state can render, item, list and document views of the entities the capture's responses showed, answer from it. List reads filtered, searched, sorted or paged answer from it too wherever the capture's recorded calls explain what each query parameter does, including values no call recorded; a parameter the recordings do not explain, or a flag's unrecorded value, replays. Other reads replay the recorded response whose path and query values best match the arguments, with replay.exact_match saying whether it matched exactly; a replayed read does not change after a write. A served write was confirmed by the capture itself: replaying the session against the extracted state, a later recorded read reproduced its effect and none contradicted it. A create generates ids and clock values deterministically, so the same calls after a reset give the same state; an update changes the members the call sends; a delete removes the entity. Each takes only the body members the recorded call showed and refuses others with a message that names the accepted ones. Its effect shows in the reads the state renders, lists included wherever a list lists its whole collection or a learned query answers the call; a list call nothing explains keeps its recorded members, dropping a deleted entity but not adding a new one. The data plane at /agentenv takes data/reset (restore the extracted state), data/add and data/get (return it), which is how write tasks are graded.

Not modelled: side effects a write has beyond its own entity (counters, copies in other collections, related entities), preconditions the application enforces before a write, list query parameters the recorded calls do not explain, paging members such as page counts and next links when a call asks for another page size or page, and anything the capture never showed. The manifest's unknowns list what a capture cannot establish.

Tasks and grading

Retrieval

task	tool	arguments	field	answer
qa-01-get_ghost_admin_tags_by_uid	get_ghost_admin_tags_by_uid	uid=6adb94d1794a8e0001ce7596	tags[0].description	Stone counts and quarry audits for the rota.
qa-02-get_ghost_admin_members_by_uid	get_ghost_admin_members_by_uid	uid=6adb94d2794a8e0001ce75af	members[0].email	ada.finch@example.test
qa-03-get_ghost_admin_posts_by_uid	get_ghost_admin_posts_by_uid	uid=6adb94d2794a8e0001ce759a, formats=mobiledoc,lexical, include=tags,authors,authors.roles,email,tiers,newsletter,count.clicks,post_revisions,post_revisions.author	posts[0].post_revisions[0].title	Granite Ledger: Quarterly Stone Count
qa-04-get_ghost_admin_tags	get_ghost_admin_tags	limit=100, page=1, order=name asc, include=count.posts, filter=visibility:public	tags[0].description	Menus that follow the tides.
qa-05-get_ghost_admin_posts_slug_blue_heron_ki	get_ghost_admin_posts_slug_blue_heron_kitchen_tide_table_men	no arguments	posts[0].slug	blue-heron-kitchen-tide-table-menu
qa-06-get_ghost_admin_labels	get_ghost_admin_labels	limit=100, page=1	labels[0].name	Rota Crew
qa-07-get_ghost_admin_posts	get_ghost_admin_posts	formats=mobiledoc,lexical, limit=5, filter=status:[published,sent], order=published_at desc	posts[0].slug	granite-ledger-quarterly-stone-count
qa-08-get_ghost_admin_members	get_ghost_admin_members	order=id, limit=1, page=1, include=newsletters,labels	members[0].labels[0].slug	rota-crew
qa-09-get_ghost_admin_members_by_uid	get_ghost_admin_members_by_uid	uid=6adb94d2794a8e0001ce75af	members[0].name	Ada Finch
qa-10-get_ghost_admin_posts_by_uid	get_ghost_admin_posts_by_uid	uid=6adb94d2794a8e0001ce759a, formats=mobiledoc,lexical, include=tags,authors,authors.roles,email,tiers,newsletter,count.clicks,post_revisions,post_revisions.author	posts[0].custom_excerpt	412 blocks counted, audited by Rui Castellanos.
qa-11-get_ghost_admin_posts_slug_blue_heron_ki	get_ghost_admin_posts_slug_blue_heron_kitchen_tide_table_men	no arguments	posts[0].title	Blue Heron Kitchen opens its tide-table menu
qa-12-get_ghost_admin_members	get_ghost_admin_members	order=id, limit=1, page=1, include=newsletters,labels	members[0].note	Keeps the tide tables for the winter rota.

Writes

Graded from the environment's state through data/get (scores.state in the episodes table; reward 1 only if every check passes).

task	prompt	checks
write-delete	On this Ghost site, delete the post "Blue Heron Kitchen opens its tide-table menu". Leave every other post as it is.	the Blue Heron Kitchen post is gone; the other three posts are still there, and nothing else was added
write-delete-lookup	On this Ghost site, the unpublished draft about the lighthouse rota's winter shifts is no longer needed: delete that draft, and leave the published posts as they are.	the winter shift draft is gone; the three published posts are still there, and nothing else was added

Episodes

model	retrieval episodes	passed	pass rate	mean tool calls	mean input tokens
openai/gpt-oss-120b:groq	12	12	100%	1.0	21,135
zai-org/GLM-5.3-Flash:zai-org	12	12	100%	1.0	25,638

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

model	write episodes	passed
openai/gpt-oss-120b:groq	2	1
zai-org/GLM-5.3-Flash:zai-org	2	2

The post create and edit are left out rather than faked. Ghost turns the html a post is sent into its excerpt and its lexical document, and REA ties those derived values to the html of the call it learned from, so the second scratch post's create and edit, which sent other text, leave them unexplained. gpt-oss-120b missed write-delete-lookup: it called get_ghost_admin_posts_slug_lighthouse_rota_winter_shift_draft, a name the manifest cuts to 60 characters (get_ghost_admin_posts_slug_lighthouse_rota_winter_shift_draf), and the model provider rejected the call with HTTP 400, so the episode ended before the model answered.

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/GhostEnv-REA --bundle ghost --task smoke          # Docker, no model
agent-env hf run earakely-scale/GhostEnv-REA --bundle ghost --eval ghost-retrieval --model openai/gpt-oss-120b:groq
agent-env hf run earakely-scale/GhostEnv-REA --bundle ghost --eval ghost-writes --model openai/gpt-oss-120b:groq

The evals need 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. Ghost 5.130 ran in Docker on a throwaway VM, bound to loopback, seeded through its own Admin API with synthetic records (owner Rea Admin, 2 tags, 2 published posts and 1 draft, 1 page, 1 member with a label); people and places are invented.
2. Playwright drove a recorded admin login, the posts, editor, pages, tags, members and settings screens, and a scripted read sequence through the Admin API. It then created, edited and deleted two scratch posts, renamed the seeded draft and deleted one seeded post, reading each write back at once by item and by whole list; the HAR was scoped to the Admin API, secret-looking string 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 and the state it serves; gen_tasks.py derived the retrieval tasks from the manifest and the recorded bodies; the write tasks were written by hand against the served write tools and the seeded names; 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	Ghost (this dataset)
Environment	A simulation: a container quay, ships, cranes, weather, with A2A ship agents	An application's API rebuilt from one captured session
Observation	Tool results computed by the simulator as the week unfolds	Answers rendered from the extracted state, or 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	A state extracted from the capture (143 entities); served writes change it; data/reset restores it
Reward	Deterministic, against a CP-SAT optimum or a re-planner anchor; continuous 0 to 1	Deterministic: recorded-value match for retrieval, state checks for writes; 0 or 1 per task
What it measures	Planning under uncertainty and negotiation	Tool discovery, parameter inference and faithful retrieval in a real API, and changing its state
Data provenance	Real 2024 port calls and weather; simulated workloads	Synthetic seed data on a self-hosted Ghost; 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 and the state's limits above say what it still leaves out.

Licence and credits

Bundle code and tasks: Apache-2.0. Recorded content: responses generated by a self-hosted Ghost instance over synthetic data; Ghost is MIT software by the Ghost Foundation 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.