jsonl-corelisted
Install: claude install-skill taxueseek/session-digger
# JSONL Core — Conversation Parsing Infrastructure
## Architecture
All parsing logic lives in `${CLAUDE_PLUGIN_ROOT}/scripts/echolib/` — a Python package (stdlib only, Python 3.6+). User-facing tools are `sd-recall.py` (search/list/stats), `index-builder.py` (FTS index), `topic-segmenter.py` (segmentation), `trend-engine.py` (longitudinal aggregation), `skill-gap-finder.py` (pain-point mining), `format-detector.py` (unknown agent format detection), `topic_classify.py` (topic routing), `chat-profiles.py` (group chat participant extraction), and `remember.py` (auto memory generation).
## Data Locations
- **Session index (fast path)**: `~/.claude/projects/<encoded-path>/sessions-index.json`
- **Fallback index (built by session-digger)**: `~/.claude/projects/<encoded-path>/.session-digger-index.json`
- **SQLite unified index**: `~/.session-digger/index.db` — auto-maintained by `index-builder.py`, stores session stats, tool usage JSON, timestamps for all environments
- **Full conversations**: `~/.claude/projects/<encoded-path>/<uuid>.jsonl`
- **Subagent conversations**: `~/.claude/projects/<encoded-path>/<uuid>/subagents/agent-<id>.jsonl`
- **Global prompt history**: `~/.claude/history.jsonl`
**Important:** Only ~10% of projects have `sessions-index.json`. The scripts automatically build a fallback index from raw `.jsonl` files for the remaining 90%, cached in `.session-digger-index.json`.
## Strategy: Fast Path First
Always start with the index before opening any `.jsonl`