Automates the formatting and uploading of agent transcripts and evaluation logs to the Docent analysis platform.
The Docent Ingestion skill provides a specialized, multi-stage workflow designed for AI researchers and developers to move agent run data into Docent for deep qualitative analysis. It intelligently maps complex directory structures, infers schemas from diverse log formats, and handles metadata mapping to ensure data is organized into logical collections. Whether you are working with Inspect AI files or custom JSON trajectories, this skill automates the generation of ingestion scripts and verifies the final upload using the Docent SDK, making it easier to debug agent behaviors and compare model checkpoints.
Key Features
01Strategic schema inference from sample transcript files
02Guided ingestion planning with field-to-class mapping
03Post-upload verification using the Docent SDK
041 GitHub stars
05Automated directory structure and naming pattern detection
06Native support for Inspect AI (.eval) file formats
Use Cases
01Comparing agent trajectories between different model versions to identify regressions
02Converting raw JSON/YAML transcripts into structured Docent collections for human annotation
03Uploading large batches of agent evaluation logs for semantic search and clustering
