Workflow Retrospective analyzes raw conversation histories from Codex, Claude Code, and ChatGPT to conduct evidence-based audits of recurring developer patterns and friction points. By parsing JSONL session exports, thread transcripts, and user corrections, it identifies genuine workflow recurrences rather than one-off topics, recommending precise, minimal configurations (such as AGENTS.md rules or custom skills) with exact diffs before any changes are applied. It is ideal for engineering teams and individual developers looking to standardize best practices, reduce repeated prompt corrections, and systematically improve AI-assisted coding efficiency.
主要功能
01Distinguishes genuine reusable workflows and recurring friction from isolated topics
02Safeguards sensitive data by redacting credentials, secrets, and auth tokens during audits
03Parses raw session transcripts and JSONL logs across Claude Code, Codex, and ChatGPT exports
04Generates unified diffs and explicit dry-run proposals prior to applying any changes
05Proposes granular, scoped improvements ranging from repository rules to dedicated Claude skills
061 GitHub stars
使用场景
01Post-sprint audits to eliminate repetitive prompt corrections and manual setup across team members
02Standardizing successful ad hoc coding patterns into persistent repository documentation or skills
03Investigating workflow breakdowns and tool failures across past development sessions