What problem does it solve? Claude Code records every session as a JSONL transcript, but that raw data is hard to interpret. This Skill turns those transcripts into an evidence-backed report showing what the agent is actually used for, where it fails, and what it costs. ## Core Features & Use Cases - Deterministic extraction pipeline: Runs gaia.factory.harvest modules (scan, report, context, savings) to normalize sessions into traces, intents, and aggregate statistics without sending data anywhere. - Use-case classification: Guides batch labeling of session intent with a consistent taxonomy so reports can break down behavior by workflow type. - Metric rigor: Documents five common analysis pitfalls (subagent scoping, argument hashing, error taxonomy ordering, token carry, harness turns) plus honesty requirements for reporting. - Use Case: A developer wants to know which tools fail most often in their Claude Code sessions and what oversized tool results cost in tokens; the pipeline produces per-tool failure rates and savings estimates in ~/.gaia/cache/factory/. ## Quick Start Analyze my Claude Code session history and produce a report on usage patterns, failure rates, and token costs.
