gss1147 commited on
Commit
d4ec4a1
·
verified ·
1 Parent(s): e5dbb36

Upload 2 files

Browse files
Files changed (2) hide show
  1. LICENSE +5 -0
  2. README.md +84 -0
LICENSE ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ Apache License
2
+ Version 2.0, January 2004
3
+
4
+ This repository ships a template dataset package.
5
+ For the full Apache-2.0 text, see: https://www.apache.org/licenses/LICENSE-2.0
README.md ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: OpenToolTrace-X (Platinum)
3
+ language:
4
+ - en
5
+ license: apache-2.0
6
+ task_categories:
7
+ - text-generation
8
+ - question-answering
9
+ - reinforcement-learning
10
+ tags:
11
+ - agents
12
+ - tool-use
13
+ - trajectories
14
+ - verification
15
+ - code
16
+ - bash
17
+ - git
18
+ size_categories:
19
+ - n<1K # sample pack; replace after scaling
20
+ dataset_info:
21
+ creator: "Within US AI"
22
+ contact: "Within US AI"
23
+ created: "2025-12-30T16:53:41Z"
24
+ schema: "See Features section below"
25
+ ---
26
+
27
+ # OpenToolTrace-X (Platinum)
28
+
29
+ **Developer/Publisher:** Within US AI
30
+ **Version:** 0.1.0 (sample pack)
31
+ **Created:** 2025-12-30T16:53:41Z
32
+
33
+ ## What this dataset is
34
+ `OpenToolTrace-X` is a **replayable, verifiable** corpus of tool-using agent trajectories.
35
+
36
+ Each record contains:
37
+ - A user goal (`prompt`) and constraints
38
+ - An `initial_state` describing the starting environment/repo snapshot
39
+ - A `trajectory` (tool calls + observations)
40
+ - A `final_state` (artifacts/diff/output)
41
+ - `verification` (tests, checksums, exit codes) to make outcomes **machine-checkable**
42
+
43
+ ## Features / schema (JSONL)
44
+ - `task_id` (string)
45
+ - `domain` (string; e.g., `python`, `bash`, `git`, `data`)
46
+ - `difficulty` (int; 1–5)
47
+ - `prompt` (string)
48
+ - `constraints` (string)
49
+ - `initial_state` (object)
50
+ - `trajectory` (list of objects)
51
+ - `final_state` (object)
52
+ - `verification` (object)
53
+ - `tags` (list of strings)
54
+ - `created_utc` (string; ISO 8601)
55
+ - `license_note` (string)
56
+
57
+ ### Trajectory step format
58
+ Each step is a dict:
59
+ - `tool` (e.g., `bash`, `python`, `git`)
60
+ - `action` (command / code / args)
61
+ - `observation` (stdout / structured output)
62
+ - `exit_code` (int)
63
+ - `stderr` (string, optional)
64
+ - `artifacts_written` (list of strings, optional)
65
+
66
+ ## Data splits
67
+ - `data/train.jsonl`
68
+ - `data/validation.jsonl`
69
+ - `data/test.jsonl`
70
+
71
+ ## Replay harness (scaffold)
72
+ See `replay_harness/` for a safe, non-executing replay viewer.
73
+ Integrate your own sandbox executor for real replays.
74
+
75
+ ## How to load
76
+ ```python
77
+ from datasets import load_dataset
78
+ ds = load_dataset("json", data_files={
79
+ "train": "data/train.jsonl",
80
+ "validation": "data/validation.jsonl",
81
+ "test": "data/test.jsonl",
82
+ })
83
+ print(ds["train"][0])
84
+ ```