Text / AI & Machine Learning
Agentic Coding Trajectory Data | Verbatim Harness Sessions | Claude Code, OpenClaw, OpenCode, Kilo | JSONL
Advertised by providerLicensing rights not independently verifiedPrice on request
Overview
Agentic Coding Trajectory Data | Verbatim Harness Sessions | Claude Code, OpenClaw, OpenCode, Kilo | JSONL is a text dataset. Typical applications include fine-tuning & instruction data, conversational AI, agents & tool use and 1 more. Relevant sectors: AI & Machine Learning. The listing mentions timestamps. Pricing is not published and is provided on request.
Primary applications
What the listing offers
- Listing mentions timestamps.
Most agent datasets get cleaned up before they reach the buyer. System prompts are swapped for a generic stand-in, tool results are truncated or templated, reasoning traces are dropped, and the message envelope is flattened into whatever shape was convenient for the vendor. Data like that looks fine in a schema check and then produces models that behave oddly in deployment, because the context distribution no longer resembles anything a real agent encounters. We capture at the API boundary and ship what we recorded. What each record contains Every row is one model turn with its complete preceding context. That means the full system prompt the harness actually assembled, including its kernel instructions, injected workspace files, the skills registry, and runtime and version lines. User messages arrive with every harness-injected context block intact, including system reminders, environment blocks, IDE context, and heartbeat prompts, along with original markup and attachment references. Assistant messages retain response text and tool calls together, plus reasoning content where the model produced it. Tool results are the genuine receipt: real stdout and stderr, real exit codes, real file paths, real identifiers and timestamps. Tool call IDs pair one to one with their corresponding results. The tool list authorized for that specific request ships alongside each record, with names, descriptions, and parameter schemas matching what the system prompt described. How we collect it
Related datasets
More in Code & Software →Audio · TextAdvertised
Call Center Conversation Dataset [Hindi] – Real customer-agent dialogues for AI and speech model training
This dataset includes authentic Hindi-language call center conversations between customers and support agents.
- Language
- Hindi
- Volume
- —
- Domain
- Customer Service +2
- Use case
- Speech recognition
USD 10,000
More text & language datasets
All 776 text & language datasets →- AI-Assisted Synthetic SOPs | Source-Grounded Engineering Knowledge
- AI & ML Training Data | 171MM+ U.S Identities for Model Training | Identity Resolution | Identity Verification
- AI & ML Training Data | Real Estate & Property Records for LLM Training | 158M+ Properties | Structured NLP-Ready | Bulk
- AI Training Data | US Transcription Data | Unique Consumer Sentiment Data: Transcription of the calls to the companies
- AI Training Dataset [Audio Calls and Transcriptions] – Real support conversations for training conversational and sentiment-aware AI
- Airframe MRO Events - Worldwide commercial aviation airframe maintenance, repair, and overhaul events
- Alternative Data | 72M+ Company Data, 915M+ Employee Data, and 486M+ Jobs Data Records | Global | 5B+ Records