task_id stringlengths 10 33 | split stringclasses 1
value | component stringclasses 5
values | domain stringclasses 7
values | behavior stringlengths 51 100 | skill_family stringclasses 8
values | source_dataset stringclasses 5
values | source_revision stringclasses 5
values | source_url stringclasses 5
values | source_record stringclasses 1
value | reward stringclasses 3
values | instance_policy stringclasses 3
values | original_instruction stringlengths 47 71 | original_instruction_json_pointer stringclasses 1
value | original_instruction_sha256 stringlengths 64 64 | computational_tools bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
awm-accounting-1-invoice | train | awm | office-white-collar | Business entity lookup, invoice creation, discount and tax calculations, and parent-child line items | business_api | Snowflake/AgentWorldModel-1K | dde80a0283fe781bdc51656bce57063dc5650213 | https://huggingface.co/datasets/Snowflake/AgentWorldModel-1K/tree/dde80a0283fe781bdc51656bce57063dc5650213 | gen_tasks.jsonl#scenario=accounting_1;tasks[1] | Unchanged original pure-code state verifier, binary 0/1. | Fixed original request and pristine database copied at reset. | components/awm/private/tasks/awm-accounting-1-invoice/SOURCE.json | /task | a08b48d978d8a70cfea5c6e871d079622fc99ac8438fb38d6d6816bdfa399c99 | null |
incident-3 | train | loui | software-engineering | Repair event ordering, sliding windows and reset handling | code_repair | Louistiti/openenv-python-repair | f82d0dabb0da9b1b913ef95f89f54c7a8f39d0ad | https://huggingface.co/datasets/Louistiti/openenv-python-repair/tree/f82d0dabb0da9b1b913ef95f89f54c7a8f39d0ad/v3 | null | Original mean of per-behavior-group pass fractions. | Original task generator; default seed 0, explicit reset seed supported. | components/loui/tasks/incident-3/instruction.md | null | 682c7db0f0911e85174652574349fbe957c1cc235a68d224c569e405d9674452 | null |
task_008667_2151f271 | train | tmax | finance-economics | Cross-account ownership mapping, transaction cycle detection, and deduplication | database_workflows | TMaxxx/TMax-15K-Harbor | 48a77eb0b017606c643ed905f96db41672914798 | https://huggingface.co/datasets/TMaxxx/TMax-15K-Harbor/tree/48a77eb0b017606c643ed905f96db41672914798 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | components/tmax/tasks/task_008667_2151f271/source/instruction.md | null | 50d1d10121dfa3b565842fd10f2589ceb91b4459126787af7793cec9cc0d06b1 | null |
construct-mc-costas-array-l3-s0 | train | math | mathematics-or-formal-reasoning | Implementing constructive constraints and complete validation | constraint_construction | amphora/MathConstructOptimize-Envs-harbor | 66cd5c87d51113c76faaef1b05397f1a6282a0f7 | https://huggingface.co/datasets/amphora/MathConstructOptimize-Envs-harbor/tree/66cd5c87d51113c76faaef1b05397f1a6282a0f7 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | components/math/upstream/construct-mc-costas-array-l3-s0/instruction.md | null | 88e4a26e71d8483fbc89627210e4ed07f6dfef0fb16ce3e6d31e84a04bd050f9 | null |
task_000260_7b7cc286 | train | tmax | media-content-production | Frame-by-frame video analysis and structured output | media_analysis | TMaxxx/TMax-15K-Harbor | 48a77eb0b017606c643ed905f96db41672914798 | https://huggingface.co/datasets/TMaxxx/TMax-15K-Harbor/tree/48a77eb0b017606c643ed905f96db41672914798 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | components/tmax/tasks/task_000260_7b7cc286/source/instruction.md | null | 79c9fd4e4d41524db109d90b0415a4f0a2b9d24bf690fa8d7340ce3cb5ce0975 | null |
task_000663_dff2c340 | train | tmax | office-white-collar | Backup restoration dependencies, time boundaries, and resource aggregation | database_workflows | TMaxxx/TMax-15K-Harbor | 48a77eb0b017606c643ed905f96db41672914798 | https://huggingface.co/datasets/TMaxxx/TMax-15K-Harbor/tree/48a77eb0b017606c643ed905f96db41672914798 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | components/tmax/tasks/task_000663_dff2c340/source/instruction.md | null | feae25374139b4b1d8ec9759d553c591841d9330a071217f46a2e464bd32bbd7 | null |
pdbthink-s09-7ba4aa076b6bb5633953 | train | pdbthink | natural-science | Coordinate extraction and oriented geometry classification without tools | tool_free_geometry | open-athena/pdbthink-coordinate-tasks | 3734406cb97b1702844319f9a5d860cbbf8fe660 | https://huggingface.co/datasets/open-athena/pdbthink-coordinate-tasks/tree/3734406cb97b1702844319f9a5d860cbbf8fe660 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | upstream/pdbthink/pdbthink-s09-7ba4aa076b6bb5633953/instruction.md | null | 21dc586748f23a8630118831399e1c40d62443444d454bf6f5e7f60ba334b06f | false |
expression-3 | train | loui | software-engineering | Repair tokenization, operator precedence and unary expressions | code_repair | Louistiti/openenv-python-repair | f82d0dabb0da9b1b913ef95f89f54c7a8f39d0ad | https://huggingface.co/datasets/Louistiti/openenv-python-repair/tree/f82d0dabb0da9b1b913ef95f89f54c7a8f39d0ad/v3 | null | Original mean of per-behavior-group pass fractions. | Original task generator; default seed 0, explicit reset seed supported. | components/loui/tasks/expression-3/instruction.md | null | 09ef2d8fc856ca3dc5662c279df705e9accc827f4472dc34758ffe769d116128 | null |
task_000056_f726766b | train | tmax | industrial-physical-systems | Waveform system identification, numerical fitting, and result verification | system_identification | TMaxxx/TMax-15K-Harbor | 48a77eb0b017606c643ed905f96db41672914798 | https://huggingface.co/datasets/TMaxxx/TMax-15K-Harbor/tree/48a77eb0b017606c643ed905f96db41672914798 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | components/tmax/tasks/task_000056_f726766b/source/instruction.md | null | 714f780883970365cfffe408e2c76e75f499005f87551fc84b55d79368e77197 | null |
task_005684_3c8594a3 | train | tmax | office-white-collar | Communication graph computation, database joins, and sensitive access checks | database_workflows | TMaxxx/TMax-15K-Harbor | 48a77eb0b017606c643ed905f96db41672914798 | https://huggingface.co/datasets/TMaxxx/TMax-15K-Harbor/tree/48a77eb0b017606c643ed905f96db41672914798 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | components/tmax/tasks/task_005684_3c8594a3/source/instruction.md | null | f4eee4e0643fcfb7399df10e1947ec9dbda28c74bacf307167f443506268f583 | null |
task_008998_ae4cbe72 | train | tmax | industrial-physical-systems | State parsing from compressed manufacturing logs and consistent multi-file output | stream_parsing | TMaxxx/TMax-15K-Harbor | 48a77eb0b017606c643ed905f96db41672914798 | https://huggingface.co/datasets/TMaxxx/TMax-15K-Harbor/tree/48a77eb0b017606c643ed905f96db41672914798 | null | Unchanged original binary success reward; 0 or 1. | One fixed original train instance per task_id; no random question substitution. | components/tmax/tasks/task_008998_ae4cbe72/source/instruction.md | null | f81f86b07681931baabdba313cc4e988c01bff7a1f1d83349dfa939cdaa20669 | null |
Eleven original training tasks served by five OpenEnv images.
| Task type | Count | Source |
|---|---|---|
| Python repair | 2 | Louistiti/openenv-python-repair |
| Structured data workflows | 3 | TMax |
| Business API workflow | 1 | AgentWorldModel |
| Manufacturing log processing | 1 | TMax |
| Video analysis | 1 | TMax |
| Physical parameter identification | 1 | TMax |
| Mathematical constraint construction | 1 | MathConstruct |
| Protein geometry | 1 | PDBThink |
tasks.jsonl records task IDs, original instructions and pinned sources. Each ID selects one original task. Reset restores its workspace or database; the Python repair tasks use the original seeded generator, with seed 0 by default. Original graders determine rewards. Python repair uses the original mean of behavior-group pass fractions; the other tasks use binary rewards.
components/ contains runtime adapters, task inputs, original graders and dependency records. upstream/ preserves the selected Harbor materials. All images serve schema.json through one persistent WebSocket connection at /ws. Shell tasks use exec and finish; Python repair and business workflows use call and finish; protein geometry uses submit_answer through call.
REGISTRATION.json records the 11 available task IDs. SOURCE_LOCK.json records provenance; MANIFEST.json and SHA256SUMS cover package files. Licenses and attribution are in LICENSES/README.txt and the component notices.
Validate a downloaded dataset and restore required empty source directories:
python3 -B prepare_source_tree.py
- Downloads last month
- 171