This skill should be used when the user asks to "analyze sessions", "review my Claude usage", "session insights", "what have I been working on", "improve my workflow", "analyze conversation history", or mentions reviewing past Claude Code sessions. Routes to /session-review or /session-stats commands.
From session-insightsnpx claudepluginhub nthplusio/functional-claude --plugin session-insightsThis skill uses the workspace's default tool permissions.
Executes pre-written implementation plans: critically reviews, follows bite-sized steps exactly, runs verifications, tracks progress with checkpoints, uses git worktrees, stops on blockers.
Guides idea refinement into designs: explores context, asks questions one-by-one, proposes approaches, presents sections for approval, writes/review specs before coding.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
Analyze your Claude Code conversation history with interactive selection, deep drill-down, and workflow improvement generation.
Session Insights provides programmatic analysis of your local Claude Code session data:
/insightsThe built-in /insights command generates a broad HTML report across up to 50 sessions with Haiku-level analysis. Session Insights complements it with:
/session-review — Deep Analysis WorkflowA 6-phase interactive workflow:
/session-stats — Quick DashboardFast overview of usage statistics:
All analysis uses streaming Node.js scripts at ${CLAUDE_PLUGIN_ROOT}/scripts/:
| Script | Purpose |
|---|---|
list-projects.js | Enumerate projects with session counts and sizes |
list-sessions.js | List sessions for a project with metadata |
extract-session.js | Structured turn-by-turn summary with detail levels (low/medium/high) |
extract-history.js | Parse history.jsonl with filters |
aggregate-stats.js | Cross-session aggregate statistics |
Scripts read JSONL files via readline streaming — they never load entire files into memory. Output is JSON to stdout.
IMPORTANT: Always use scripts to parse session data. Never read .jsonl files directly into context — they can be megabytes in size and would consume the entire context window.
~/.claude/history.jsonl — One line per user message with {display, sessionId, project, timestamp}~/.claude/projects/<dash-path>/<session-id>.jsonl — Full transcripts with user, assistant, progress, file-history-snapshot types-home-user-Source-project)All analysis is performed locally. No session data leaves your machine. Scripts read from ~/.claude/ and output structured JSON — the raw conversation content stays in the JSONL files.