What problem does it solve? Reviewing weeks of Claude Code session transcripts by hand is impractical, so teams lack evidence-based feedback on how their human-plus-AI development workflow actually performed and what guidance in CLAUDE.md should change. ## Core Features & Use Cases - Transcript Digestion: Condenses raw JSONL session transcripts into per-day Markdown digests via scripts/digest.py, stripping low-signal tool noise while keeping user and assistant text. - Parallel Batch Analysis: Fans out digested batches to parallel background agents, then merges and deduplicates findings into one report. - Two-Sided Retrospective: Evaluates both Claude's behavior and the human supervisor's behavior across continue/stop/improve categories, ending with prioritized CLAUDE.md edit candidates. - Use Case: After a three-week development stretch, ask for a process review since a given date and receive a structured retrospective covering both team members plus concrete guidance changes. ## Quick Start Ask the assistant to run a development process review since a specific date, for example: review our development process since 6/23/2026.