--- tags: - rl-environment - agentenv license: apache-2.0 configs: - config_name: directus_tasks data_files: - split: train path: tasks/directus.parquet default: true - config_name: directus_episodes data_files: - split: train path: episodes/directus.parquet --- # directus directus An [AgentEnv](https://www.agentenvframework.com) bundle with its recorded runs, published with [agentenv-hf](https://github.com/earakely-scale/agentenv-hf-plugin). ## Tables - `directus_tasks`: one row per task of the bundle, with its prompt, envs, verifiers and steps. - `directus_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/directus.jsonl` holds each run's record and native trajectory, and `bundles/directus/` the bundle itself. ## Run it ```bash pip install agentenv-framework huggingface_hub hf download earakely-scale/DirectusEnv-REA --repo-type dataset --local-dir agentenv-dataset agent-env run agentenv-dataset/bundles/directus ``` # Directus on AgentEnv, from a REA capture **What this is.** An environment of **Directus 11.17.4** (BSL-1.1), built from one recorded browser session against a self-hosted instance seeded with synthetic data. [REA](https://github.com/morluto/rea) projected the capture into an **environment manifest** (40 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 Directus. 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 `v0.1.0` keeps the earlier replay-only version. ## What's in it - **40 tools** (38 GET, 1 POST, 1 PATCH) over 96 recorded exchanges on `http://127.0.0.1:3223`; 2 recorded groups excluded (2 as non-JSON replies such as bodiless 204 writes). 40 are offered. - **Path parameters**: 9 `uuid`, 4 `numeric`. 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 40 tools. - **State**: 17 entities of 12 types. 31 read tools answer from it and 7 replay their recorded responses. 5 list tools also answer filters, search, sorting or paging from it. Served write tools: `post_items_provisions` (create), `patch_items_provisions_by_id` (update). - **Unknowns** the manifest states explicitly: authorization model, response semantics beyond recorded exchanges, state transitions between tools. - **12 retrieval tasks** (`evals/directus-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/directus-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_c5b6fd36acdc9d4a03b8785ae617dff618fb66b03fbc51bc30de105efa0e6a03` (scoped HAR sha256 `b72f2c447695c9ce90cd01aa2a36cfaf81698f473e85cf81f6497bb8a4d66e9d`, 234,020 bytes), manifest view `ev_078f1b8a2f8a7246add6366fe956c83241b1d7fe1599b832d95968b14b777342`, confidence `inferred`, authority `historical-reference`. The REA build is `rea-agents` 6.7.0 from main `3220e17b` with open pull requests merged: morluto/rea#5600 at `45a5c8b9` (which carries #5351 and #5352), #5787 at `27dd6062`, #5800 at `1b699306`, #6136 at `755dfca4` and #6138 at `58185484`. 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_items_crew_rota_by_id` | `get_items_crew_rota_by_id` | id=1 | `data.name` | `Ada Finch` | | `qa-02-get_presets_by_id` | `get_presets_by_id` | id=1 | `data.bookmark` | `Night shift crew` | | `qa-03-get_dashboards_by_uuid` | `get_dashboards_by_uuid` | uuid=b9c4734b-06ec-44bc-a263-b2a1f1d93c33 | `data.name` | `Depot Overview` | | `qa-04-get_files_by_uuid` | `get_files_by_uuid` | uuid=00e6fb70-f6ae-4f7f-9143-7dc8e7131353 | `data.title` | `Granite Ledger tally GL-2291` | | `qa-05-get_users_by_uuid` | `get_users_by_uuid` | uuid=09d0b64a-a70f-4bfa-beae-5ddaafc8eebb | `data.email` | `ada.finch@example.com` | | `qa-06-get_flows_by_uuid` | `get_flows_by_uuid` | uuid=1657b0c4-a642-4b6f-9b99-e3e418a1c560 | `data.name` | `Announce new provisions` | | `qa-07-get_items_provisions_by_id` | `get_items_provisions_by_id` | id=2 | `data.title` | `Pale Meridian Tea` | | `qa-08-get_policies_by_uuid` | `get_policies_by_uuid` | uuid=a5234d2f-f895-456b-a6ed-97e760d5f5fa | `data.description` | `Create and update crew rota entries and read provisions.` | | `qa-09-get_roles_by_uuid` | `get_roles_by_uuid` | uuid=1d0e26ce-8f7a-4cb0-b243-6569c182c507 | `data.description` | `Keepers of the crew rota for the Lighthouse Rota.` | | `qa-10-get_folders_by_uuid` | `get_folders_by_uuid` | uuid=994a6f6d-a316-41ec-9535-e57b9a6e5218 | `data.name` | `Granite Ledger Scans` | | `qa-11-get_comments_by_uuid` | `get_comments_by_uuid` | uuid=ae445e61-b33a-4b11-8838-22a5b6cf4185 | `data.comment` | `Ada swaps the January night watch with Tomas Vey.` | | `qa-12-get_panels_by_uuid` | `get_panels_by_uuid` | uuid=6fa9d6b8-446f-4631-8cc5-8a76b2ec90ed | `data.name` | `Provisions on hand` | ### 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-create` | In this Directus project, add a provision titled "Copperfield Lamp Wicks" in the Lamp category: 24 packs, stored in Bay B2, with the supplier note "Spare wicks for the North Lamp Gallery." | a provision titled Copperfield Lamp Wicks exists; it is 24 packs in the Lamp category; it is stored in Bay B2; its supplier note is the one asked for; the provisions the state started with are still there, and nothing else was added | | `write-update` | In this Directus project, Pale Meridian Tea was restocked: set its quantity to 20 and its supplier note to "Restocked after the October storm." | its quantity is 20; its supplier note is the one asked for; it is the same provision, still in Bay A1; no provision was added or removed | ## Episodes | model | retrieval episodes | passed | pass rate | mean tool calls | mean input tokens | |---|---|---|---|---|---| | `openai/gpt-oss-120b:groq` | 12 | 12 | 100% | 1.0 | 16,018 | | `zai-org/GLM-5.3-Flash:zai-org` | 12 | 12 | 100% | 1.0 | 18,919 | 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 capture leaves out the Data Studio's provision row click: its item page reads the provision with `fields=title`, which returns the item without its id, and that read would leave REA unable to tell which field identifies a provision. In `write-create`, `openai/gpt-oss-120b:groq` stored the bay as `B2` rather than `Bay B2`, the spelling every provision uses, so its storage check failed. The state holds the two provisions the capture read by id (Pale Meridian Tea and Northern Quill Ink). The provisions list replays its recording, so it shows the five seeded provisions and never a created one, and a write to one of the other three seeded provisions is refused rather than faked. ## Run it yourself ```bash 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/DirectusEnv-REA --bundle directus --task smoke # Docker, no model agent-env hf run earakely-scale/DirectusEnv-REA --bundle directus --eval directus-retrieval --model openai/gpt-oss-120b:groq agent-env hf run earakely-scale/DirectusEnv-REA --bundle directus --eval directus-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. Directus ran in Docker on a throwaway VM, bound to loopback, seeded through its own API with synthetic records (two user collections with notes and field notes, crew_rota (Ada Finch, Rui Castellanos, Mina Okafor, Tomas Vey, Marisol Okonkwo-Reyes with role titles, shifts, watch stations) and provisions (Saffron Tide Biscuits, Pale Meridian Tea, Ember Compass Matches, Silver Anvil Rivets, Northern Quill Ink with category, quantity, bay, supplier note); folders Lighthouse Rota Archive and Granite Ledger Scans with one uploaded text file (Granite Ledger tally GL-2291); policy Rota Editing with 4 permissions; roles Rota Editors and Depot Viewers; users Ada Finch and Rui Castellanos; the admin's profile;...); people and projects are invented. 2. Playwright drove a login, the main pages and a scripted read sequence through the app's JSON API, then wrote two scratch objects per write target (created, then updated and deleted where the app allows), reading each write back at once; the HAR was scoped to the API, the login exchange dropped, 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 | Directus (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`) | 40 tools mirroring the application's endpoints | | State | Full: a virtual clock, triggers, frozen windows, agents that answer | A state extracted from the capture (17 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 Directus; 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 Directus instance over synthetic data; Directus is BSL-1.1 software by Monospace, Inc. and the Directus 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.