Claude Qmd Sessions

qmd-sessions

Claude Code skill that converts JSONL session transcripts into searchable markdown and indexes them in qmd.

Installation

1. Copy the skill directory to your Claude Code skills folder:

2. Start a new Claude Code session and run:

The skill walks through all setup interactively — output directory, conversion, qmd collection, embeddings, MCP server, hooks, and CLAUDE.md guidance. Each step prompts for confirmation.

What it does

1. Reads session JSONL files from ~/.claude/projects/
2. Extracts user messages + assistant text responses (skips tool_use, tool_result, thinking blocks)
3. Outputs organized markdown: {project}/{date}-{slug}-{id}.md
4. Indexes in qmd for semantic + keyword search via MCP

Files

Usage

refresh outputs both CLAUDE.md files and the last ~50 exchanges (100 turns, capped at 14,000 characters) collected from recent sessions into context. Useful for manually restoring context in a fresh session.

The setup wizard walks through interactively: output directory, conversion, verification, then checks for Bun, qmd, collection, embeddings, MCP server, hooks, and CLAUDE.md guidance.

Hooks

Four Claude Code hooks keep the index current and restore context (configured in ~/.claude/settings.json):

- PreCompact — converts session before context compaction, runs qmd update && qmd embed
- SessionEnd — converts session on exit, runs qmd update && qmd embed
- SessionStart (compact/resume/clear) — converts the session, then outputs both CLAUDE.md files and the last ~50 exchanges (from multiple recent sessions, sorted by cwd match, capped at 14,000 characters) to stdout so Claude receives instructions and conversation context after compaction.
- SessionStart (startup) — outputs CLAUDE.md files to stdout. If loadContextOnStartup is enabled in…