File size: 4,658 Bytes
ab9d003 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 | ---
pretty_name: "codex Agent Traces"
task_categories:
- text-generation
tags:
- "agent-traces"
- "format:agent-traces"
- "codex"
- "distillation"
- "teich"
configs:
- config_name: default
data_files:
- split: train
path: "**/*.jsonl"
---
This dataset was generated using [teich](https://github.com/TeichAI/teich) by [TeichAI](https://huggingface.co/TeichAI) <img src="https://cdn-avatars.huggingface.co/v1/production/uploads/6837935ac3b7ffe0d2559ce9/-AxyvV4wfUY8uo87kNKkK.png" width="20" height="20" style="display: inline-block; vertical-align: middle; margin: 0 3px;">
# codex Agent Traces
This directory contains raw agent trace files generated by teich.
JSONL files: 359
## Training-ready tools
Generated agent traces carry configured or recovered tool schemas so tools remain available for training even when a session did not call them.
Native Claude Code imports recover schemas for Claude Code and Claude Desktop built-ins, plus conservative name-derived MCP schemas, when the raw transcript only records tool names or calls.
A complete dataset-level `tools` schema snapshot is stored in `tools.json` to keep this dataset card upload-safe.
`load_traces` applies the dataset snapshot to each loaded example as a fallback `tools` field.
## Create A Similar Dataset
This dataset was staged from existing local agent sessions with `teich extract`.
To build your own local dataset, install Teich and point it at one of the supported providers:
```bash
teich extract codex --out data
```
Use `--sessions-dir /path/to/store` when your sessions are not in the default location,
and `--model <substring>` when you only want sessions whose model metadata matches a value.
## Format
Each file is newline-delimited JSON representing a single captured agent session.
The trace schema is designed for upload-first preservation so you can keep the original session history and convert it later for training.
Teich normalizes split assistant fragments during trace copy and conversion so the semantic order is reasoning first, optional assistant text second, and tool calls last.
Native Claude Code conversion also preserves runtime context such as skills, MCP instructions, hook context, permission state, date changes, and session recaps as masked `system` messages when the raw transcript provides them.
Common top-level event groups:
- `session_meta`
- `turn_context`
- `event_msg`
- `response_item`
- `session`
- `message`
- `session_info`
- `model_change`
- `thinking_level_change`
- `external_session_meta`
- `external_message`
- `external_stderr`
## Example
```json
{"timestamp": "2025-12-01T18:44:41.452Z", "type": "session_meta", "payload": {"id": "019adb3b-1fb2-7a80-a940-625566dbc902", "timestamp": "2025-12-01T18:44:36.402Z", "cwd": "c:\\Users\\user1\\AppData\\Local\\RLBotGUIX\\MyBots\\DropBot\\src", "originator": "codex_vscode", "cli_version": "0.61.1-alpha.1", "instructions": null, "__truncated__": "2 keys omitted"}}
```
## Training
Use this dataset as `username/repo` with Teich's data preparation and training utilities.
If you do not want Teich to handle chat-template formatting or masking, run `teich convert` to write standalone OpenAI-style JSONL rows with `prompt`, `messages`, `tools`, and `metadata`.
Training setup details evolve over time, so the maintained guide lives in the [Teich training docs](https://github.com/TeichAI/teich/blob/main/docs/training.md).
For loading, mixing, converting, and validating Teich datasets, see [Preparing Data](https://github.com/TeichAI/teich/blob/main/docs/prepare-data.md).
## Tool schema snapshot
The complete dataset-level tool schema snapshot was written to `tools.json` because it is too large to embed safely in the Hugging Face dataset card.
<details>
<summary>Tool names in snapshot</summary>
```json
{
"tool_count": 27,
"tools": [
"_fetch_file",
"_hf_doc_fetch",
"_hf_doc_search",
"_hub_repo_details",
"_search",
"apply_patch",
"bash",
"exec",
"exec_command",
"js",
"js_reset",
"list_mcp_resources",
"load_workspace_dependencies",
"mcp__codex_apps__base44_create_base44_app",
"mcp__codex_apps__github_fetch_file",
"mcp__codex_apps__hugging_face_dataset_search",
"mcp__codex_apps__hugging_face_hf_doc_search",
"mcp__codex_apps__hugging_face_hub_repo_details",
"read_mcp_resource",
"read_thread_terminal",
"request_user_input",
"shell",
"shell_command",
"update_plan",
"view_image",
"wait",
"web_search"
]
}
```
</details>
|