Skip to main content

Track active coding time with Claude Code and nudge yourself to take breaks

Project description

Garlic 🧄 — the AI Vampire 🧛 Warding Tool

Garlic is used to ward off vampires. According to Steve Yegge, AI tools have a vampiric effect on us, draining us of energy and making us tired and exhausted. Not because they are not good at coding, or do not make us much more productive, but simply because we get dopamine for getting stuff done quicker, leading us to work longer and think harder. In short, we need to touch grass. Instead of going hard for 12 hours straight with our coding agent of choice and burning ourselves out to only create value for our employer, we should be mindful of the $/hr formula and consider a new balance. He estimates there are no more than 3-4 hours of good work that we can do in a day with all this uplift without burning our own candles a little too brightly. As someone quite sensitive to the effects of extended dopamine release on the mind and body, I tend to agree with him. So I created garlic, a CLI tool that helps you keep the draining to a minimum and maintain your own energy levels so we can continue to be healthy little worker bees for years to come.

The idea came from this article by Steve Yegge.

How does it work?

garlic hooks into Claude Code using its hooks system. It tracks three events:

  • Session start — when you open a new Claude Code session
  • Prompt submit — when you send a message to Claude
  • Stop — when Claude finishes responding

From these events, garlic estimates how much time you have spent actively coding each day. It works across multiple concurrent Claude Code sessions by sharing a single state file with file locking.

The time model counts your full engagement cycle: the time Claude spends generating a response (up to a configurable cap — 2 hours by default — to guard against hung processes or forgotten sessions inflating your daily total), plus the time you spend reading it and thinking before your next prompt. If your thinking time exceeds 40 minutes (configurable), garlic assumes you stepped away and counts nothing for that gap. Gaps within the limit are counted in full. The limit is intentionally generous: it covers the time you spend reading docs, answering a Slack message, checking email, or getting back into context — adjacent work that's still part of your coding session.

As you approach configurable thresholds (every 30 minutes up to 4 hours by default), garlic asks Claude to gently nudge you to consider taking a break. You choose how it nudges — gentle, firm, or spicy. Each threshold only fires once, so you won't be nagged on every prompt. The final threshold delivers a more definitive "session over" message.

If you're still coding in the hour before the daily reset (1 AM by default, when reset_hour is 2), garlic sends a bedtime nudge — a distinct "wrap up and get some sleep" message that fires once per night.

Compatibility

  • Python 3.11+
  • macOS, Linux, and WSL
  • Not supported on native Windows (fcntl is unavailable)

Setup

Install garlic with uv:

uv tool install garlic-cli

Run setup to install the Claude Code hooks:

garlic setup

Setup interactively prompts for key preferences (nudge interval, max prompt gap, reset hour, nudge style) with sensible defaults — just press Enter to accept them all. To skip prompts entirely and use defaults, pass -y:

garlic setup -y

This does three things:

  1. Creates ~/.garlic/config.toml (with your chosen settings or sensible defaults)
  2. Adds garlic's hooks to ~/.claude/settings.json so they run across all your projects
  3. Appends a nudge-relay instruction to ~/.claude/CLAUDE.md so Claude always relays garlic's nudge messages verbatim as the last line of its response

To reset your config to the latest built-in defaults (useful after upgrading):

garlic setup --defaults

Setup is idempotent — safe to run again if you need to repair or update hooks.

Upgrading

uv tool install garlic-cli --upgrade

Then re-run garlic setup to update your hooks if the release notes mention hook changes.

Usage

# Check your installed version
garlic version

# See how long you have been Clauding today
garlic status

# Output status as JSON (for scripting and statusline integrations)
garlic status --json

# Output a compact single-line string for the Claude Code status bar
garlic statusline

# See your rolling 7-day usage summary
garlic week

# See monthly totals, streaks, and averages
garlic stats

# Disable nudging for the rest of the day (tracking continues)
garlic ignore

# Update a config value without editing the file
garlic set nudge_style=spicy
garlic set max_prompt_gap_minutes=60

# Reset the daily timer to zero
garlic reset

Slash command

After running garlic setup, you can use /garlic directly in Claude Code to check your status without leaving the conversation.

Status line

To wire garlic into Claude Code's built-in status bar, run this in Claude Code:

/statusline add the output of the `garlic statusline` command to our status line

This shows a single-line readout like 🧄 2h 15m / 4h — your accumulated time and daily target — refreshed on each Claude Code event.

Configuration

Edit ~/.garlic/config.toml to customize:

# Max thinking time (minutes) between Claude stopping and your next
# prompt that still counts as active coding. If you take longer than
# this, garlic assumes you stepped away and counts nothing for that gap.
max_prompt_gap_minutes = 40

# Max generation time (minutes) to count per response. If Claude runs
# longer than this (hung process, forgotten session), the time is
# clamped to this cap instead of inflating the daily total.
max_generation_minutes = 120

# Hour of day (0-23) when the daily timer resets.
reset_hour = 2

# Accumulated minutes at which garlic will nudge you.
# Each threshold fires only once per day. The final threshold uses
# a more definitive "session over" message.
nudge_thresholds_minutes = [30, 60, 90, 120, 150, 180, 210, 240]

# Nudge personality: "gentle", "firm", or "spicy".
nudge_style = "gentle"

Project layout

Source modules in src/garlic/:

  • cli.py — argparse entry point and subcommand dispatch
  • config.py — loads/creates ~/.garlic/config.toml with defaults
  • state.py — reads/writes ~/.garlic/state.toml under fcntl.flock, handles daily reset
  • engine.py — gap calculation, accumulation, threshold checking
  • nudges.py — hardcoded gentle/firm/spicy message pools
  • hooks.py — handlers for the session-start, prompt, and stop hook subcommands
  • setup.py — installs/updates hooks in ~/.claude/settings.json

Runtime state lives in ~/.garlic/: config.toml (settings) and state.toml (daily tracking, fcntl-locked for concurrent sessions).

Claude Code hooks written to ~/.claude/settings.json by garlic setup:

  • SessionStart (matcher "startup") → garlic hook session-start
  • UserPromptSubmitgarlic hook prompt
  • Stopgarlic hook stop
  • SessionEndgarlic hook session-end — finalizes in-flight generation time and clears last_event_time so a crashed or killed session can't leak time into the next one

Each hook reads JSON from stdin and either writes plain text to stdout (a nudge) or exits silently.

Things I should know?

No prompt injection risk. The nudge messages output by garlic's hooks are hardcoded in the project. There is no mechanism for external input to influence what gets sent to your agent. You can audit every possible message in src/garlic/nudges.py.

No runtime dependencies. Garlic uses only the Python standard library at runtime. Development dependencies (pytest, build, twine) exist for testing and releasing, but are not shipped with the package. This is an intentional choice — garlic runs on every prompt you send, so the supply chain should be as small and auditable as possible.

No data leaves your machine. All state lives in ~/.garlic/ and is never transmitted anywhere.

Built with Claude. This project was built with Claude Code, which is fitting given what it does.

Release process

  • Run make release BUMP=patch|minor|major
  • Merge release PR
  • Draft release updates
  • Publish release
  • Release workflow runs

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

garlic_cli-0.2.1.tar.gz (38.2 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

garlic_cli-0.2.1-py3-none-any.whl (22.4 kB view details)

Uploaded Python 3

File details

Details for the file garlic_cli-0.2.1.tar.gz.

File metadata

  • Download URL: garlic_cli-0.2.1.tar.gz
  • Upload date:
  • Size: 38.2 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.13

File hashes

Hashes for garlic_cli-0.2.1.tar.gz
Algorithm Hash digest
SHA256 9a41227e31c0d77c47c49a56c30a9ded36cba6b14b1c808a309e7efcb2e72d0e
MD5 ec008f9b62b2c61f63df4b8447342f70
BLAKE2b-256 6eb132ca547a3c4cd6fe4d60b053c5a8086eff0c2ea0abc04e4435854a58b7fa

See more details on using hashes here.

Provenance

The following attestation bundles were made for garlic_cli-0.2.1.tar.gz:

Publisher: release.yml on justanotherspy/garlic

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file garlic_cli-0.2.1-py3-none-any.whl.

File metadata

  • Download URL: garlic_cli-0.2.1-py3-none-any.whl
  • Upload date:
  • Size: 22.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.13

File hashes

Hashes for garlic_cli-0.2.1-py3-none-any.whl
Algorithm Hash digest
SHA256 8b0a141f987565b78b95e63e96aa274853c4e9a57616addf2c4cb2ce138ce98a
MD5 51e32f2b3b4676435130ef2de04dc8bb
BLAKE2b-256 89d3dccdb7e126e60e2aabf7248e65a9590648d03cef5541b055bcfd3dca401a

See more details on using hashes here.

Provenance

The following attestation bundles were made for garlic_cli-0.2.1-py3-none-any.whl:

Publisher: release.yml on justanotherspy/garlic

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page