Sign in to view source links and access this dataset
Description
4,333 local coding-agent session transcripts exported with the share-codex tool. The dataset contains 16,482 user turns, with each row including an ordered sequence of messages, user prompts, assistant responses, and tool calls. It was created by author nmuendler and last updated on June 18, 2026.
Use Cases
Training conversational AI models for code generation based on the sequence of user prompts and assistant responses.
Analyzing patterns in human-AI collaboration for software development based on the multi-turn message transcripts.
Benchmarking the performance of coding agents based on the recorded tool calls and their outputs.
Studying prompt engineering strategies for coding assistants based on the user-provided prompts.
Strengths
Contains 4,333 distinct coding-agent sessions, providing a substantial corpus.
Includes 16,482 user turns, indicating multi-turn, interactive conversations.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
Source
huggingface
Collection Method
Exported from local coding-agent sessions using the share-codex tool.
Freshness
Last updated 2026-06-18 13:54:51; freshness should be verified.
License is unknown; users should verify permissions before use.