A curated subset of 611 agentic traces from the badlogicgames/pi-mono dataset, processed through the Talos trace curation pipeline. The dataset was created by DJLougen and last updated on April 23, 2026. Original data was in a file-per-session JSONL format with Anthropic API-style message blocks.
Use Cases
- Analyze agent reasoning patterns based on the 'thinking' message blocks mentioned in the description
- Evaluate tool usage and results in agent workflows based on the 'toolCall' and 'toolResult' blocks
- Study error patterns and masking in agentic interactions based on the error-masked curation process
- Benchmark agent performance using a curated, standardized set of conversational traces
Strengths
- Contains 611 agentic traces, providing a substantial sample for analysis
- Processed through a dedicated curation pipeline (TALOS-trace-curator) for error masking
- Derived from a known source dataset (badlogicgames/pi-mono) with a documented original format
Limitations
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is known but the full scale and diversity of the original source dataset are not detailed
- Description metadata is limited; actual data quality requires manual inspection after download
Provenance
- Source
- badlogicgames/pi-mono on Hugging Face
- Collection Method
- Processed through the Talos trace curation pipeline (github.com/DJLougen/TALOS-trace-curator)
- Time Range
- null
- Freshness
- Last updated 2026-04 23 18:39:23; freshness should be verified
- Geography
- null