What problem does it solve? When debugging or planning new work, developers often forget what was already tried in previous AI assistant sessions. This Skill searches historical session transcripts across Claude Code, Codex, and Cursor to surface what was attempted, what failed, and what decisions were made, so you never repeat dead ends. ## Core Features & Use Cases - Cross-platform session discovery: Finds JSONL session files from Claude Code, Codex, and Cursor filtered by repo, time window, and git branch. - Keyword-based relevance filtering: Ranks sessions by keyword matches in actual user/assistant text, excluding tool outputs and metadata noise. - Skeleton and error extraction: Condenses multi-megabyte transcripts into compact conversation skeletons and error logs without loading full files into context. - Subagent synthesis: Dispatches a historian subagent to summarize findings into sections like what was tried, what didn't work, and key decisions. - Use Case: Before fixing a recurring auth bug, ask what was tried last week and get a summary of prior debugging attempts across all your AI tools. ## Quick Start Ask the assistant to search your session history for what was tried when debugging the auth middleware last week.
