productivity-kit

productivity-kit wires Claude Code and Cursor into your day: you get alerted when something needs you or finishes, and you keep structured session logs for standups, time review, and debugging.

• Sound (bundled MP3s) and desktop notifications when runs stop, fail, or need permission
• Session logs captured from assistant replies into daily JSON under work-logs/
• Small Node scripts only (no extra packages); copy into a project or use from ~/.claude/

---

Table of Contents

• How It Works
• What This Solves
• Hooks
• Session Logging
• Commands
• Quick Start
• Log Analysis
• Detailed Docs
• Vision
• Contributing
• License

---

How It Works

Quick Start

1. Copy this repo into your project, or copy .claude/ into your user-level config (~/.claude/).
2. Install Node.js and ensure node is on your PATH (hooks use Node built-ins only).
3. In Cursor, run the project custom command setup (.cursor/commands/setup.md) from the commands menu or chat.
4. Run a short test task in Cursor and Claude Code; optionally trigger a permission flow to hear/see the approval hook.

Setup details

• setup creates work-logs/, copies example configs when missing, optionally updates .git/info/exclude, runs hook tests, and asks whether local configs and logs should be tracked in git.
• Session logging is toggled with session_tracking.enabled in .claude/config.json and .cursor/config.json (defaults write under work-logs/).

Keep local files out of Git (manual alternative)

If you skip setup or need extra paths, append to .git/info/exclude:

work-logs/
.claude/config.json
.cursor/config.json

These are machine-local ignores and don't touch the shared .gitignore.

Smoke-test the hooks

Ask the agent to do something that requires approval (e.g. create a file in a path the tool must confirm) to exercise the permission hook.

Filesystem scope: hooks write only under the project root (default work-logs/, with path safeguards) and read only transcript paths supplied by Claude Code or Cursor.

---

How It Works

1. Hooks listen to Claude Code / Cursor lifecycle events (session start/end, stop, failure, permission, after-agent reply).
2. Scripts under .claude/hooks/ and .cursor/hooks/ play sounds / show notifications or parse replies and update the store.
3. Logs are written as one file per day — work-logs/session-work-log-YYYY-MM-DD.json — when session tracking is enabled.
4. You skim or query those files (and transcripts) to improve prompts, estimate time, and see what actually shipped.
5. It runs out of the box with the bundled hook wiring (.claude/settings.json, .cursor/hooks.json).

---

What This Solves

Problem	How the kit helps
Missing important events	Finish, failure, and approval prompts surface without watching the terminal
Babysitting long runs	Step away; get alerted when something matters
Slow failure diagnosis	Logs and transcripts help pinpoint errors and delays
Prompt quality drift	Review what the model did vs. what you asked for
Lost work context	Daily files preserve context after sessions end

---

Hooks

---

Hooks

Tool	Events	Role
Claude Code	SessionStart, SessionEnd	Daily JSON session logs
Claude Code	Stop, StopFailure, PermissionRequest	Desktop notifications on completion, failure, or approval needed
Cursor	sessionStart, sessionEnd	Same session log model as Claude Code (finalize + transcript fallback)
Cursor	afterAgentResponse	Appends a timestamped entry to session_logs while the session is open

Wiring is in .claude/settings.json and .cursor/hooks.json.

---

Session Logging

When session tracking is enabled, each session row stores session_logs: an array of objects, one per captured reply, each containing:

captured_at | user_intent | prompt_summary | provided_context | what_i_did | open_issues | next_best_step

See project rules in CLAUDE.md / .cursor/rules/task-summary.md for full field definitions.

Example session log entry

{
  "captured_at": "2026-04-19T12:34:56.789Z",
  "user_intent": "Ship the README refresh.",
  "prompt_summary": "Restructure intro, add How it works, tighten Quick start.",
  "provided_context": "README.md and prior feedback bullets.",
  "what_i_did": "Rewrote sections and added an example JSON block.",
  "open_issues": "None.",
  "next_best_step": "Optional: add a real screenshot path under docs/."
}

---

Commands

---

Commands

Command	File	What it does
setup	setup.md	One-time setup: creates work-logs/, copies example configs, updates .git/info/exclude, runs hook tests
/logger	logger.md	Reads your session log for a given date, groups entries into major tasks with time estimates, outputs a standup-ready summary

/logger — daily work summary

Run /logger in Cursor chat at any point during or at the end of your day. It will:

• Parse the daily session-work-log-YYYY-MM-DD.json for a date you choose (defaults to today)
• Cluster related log entries into major tasks, ignoring noise and trivial retries
• Estimate time per task using captured_at timestamps, deduplicating overlapping Claude + Cursor sessions
• Output a clean summary:

## Daily Log — 2026-04-19

**Total active time:** 8h 9m

### Major Tasks
1. **Session hook refactoring** — ~50 min
   Summary: Simplified session lookup and centralized shared utilities...

### Other minor work
- Re-read rules file to confirm session tracking was enabled

---

Quick Start

1. Copy this repo into your project, or copy .claude/ into your user-level config (~/.claude/).
2. Install Node.js and ensure node is on your PATH (hooks use Node built-ins only).
3. In Cursor, run the project custom command setup (.cursor/commands/setup.md) from the commands menu or chat.
4. Run a short test task in Cursor and Claude Code; optionally trigger a permission flow to hear/see the approval hook.

Setup details

• setup creates work-logs/, copies example configs when missing, optionally updates .git/info/exclude, runs hook tests, and asks whether local configs and logs should be tracked in git.
• Session logging is toggled with session_tracking.enabled in .claude/config.json and .cursor/config.json (defaults write under work-logs/).

Keep local files out of Git (manual)

If you skip setup or need extra paths, append to .git/info/exclude:

work-logs/
.claude/config.json
.cursor/config.json

These are machine-local ignores — no shared .gitignore needed.

Smoke-test the hooks

---

---

Log Analysis

The log-analysis/ directory contains a Python pipeline that runs your session logs through four LLM analysis passes (prompt audit, cost optimization, skills clustering, rules extraction) using the Gemini API, then synthesizes the results into a single prioritized report.

See the log-analysis README for setup and usage.

---

Detailed Docs

Doc	Contents
Claude hooks	Config, each hook, edge cases, filesystem scope
Cursor hooks	Config, each hook, edge cases, filesystem scope
log-analysis/README.md	Full analysis setup and command reference

---

Vision

Longer term, this kit is meant to help you spot inefficiencies, tighten prompts, and move toward more agentic workflows — while building intuition for which models and patterns save time and cost.

---

Contributing

Contributions are welcome! To get started:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-feature)
3. Commit your changes (git commit -m 'Add my feature')
4. Push to the branch (git push origin feature/my-feature)
5. Open a Pull Request

---

License

See LICENSE.