lot-map

A live map beside the chat that shows which tools, models, hooks, skills and subagents Claude Code is using right now.

Source repo
alexeyw/lot-map
Stars
★ 1k
Last updated
1d ago
License
MIT
Primary language
TypeScript

At a glance

Works with
Platform-specificClaude CodeClaude.ai (Partial support)
You'll need
Claude Code 2.1.321 (built and tested version)terminal with at least 144 columns for the docked paneGraphical terminal such as kitty, Ghostty, or iTerm2 3.7.8+ for the picture viewShell / CLILocal filesystem
Typical use
Developers who want to see, during a long session, where Claude Code is actually spending effort across files, shell, network and subagents.
Main limitation
Entirely dependent on Claude Code's plugin and hook mechanism; it cannot run standalone or against other agent runtimes.
Website
alexeyw.github.io/lot-map ↗ set on GitHub
Maintenance
Commits in the last 30 dayslast commit 2026-11-19latest release v0.12.0 (2026-11-19)
Source review
72/100 · Some gaps

What does this agent do, and when should you use it?

lot-map is a Claude Code plugin that opens a live map beside the chat, drawing every tool call, model step, hook, skill and subagent as a pulse travelling from the centre to a station (You, Model, Memory, Files, Shell, Network, Browser, Agents, Skills, Hooks). The map is SVG-rendered and animated, and appears in the desktop app, the VS Code extension and the terminal — the terminal can show log rows alone, a raster picture in kitty/Ghostty/iTerm2, or a braille-dot redraw in any terminal whose font has braille. It is strictly local: it makes no network requests, reads no files, runs no processes, calls no model and adds nothing to the conversation, and it keeps only name-like labels (last two path segments, URL host, command descriptions), never raw command lines or file contents. It ships 7 UI languages (en, ru, es, fr, de, it, pt) and roughly two dozen options covering theme, colours, layout, motion, context-cost estimates and MCP server routing. It is MIT-licensed and installs through the Claude Code plugin marketplace. Because the map follows tool calls rather than an API, it is handed each call whole; the repository documents in PRIVACY.md and PRINCIPLES.md what it reads and the checks that hold it to that.

The plugin hooks into Claude Code session start, prompts, turns, model steps, tool calls, skills and compaction, plus three settings-hook events (PreToolUse, UserPromptSubmit, SessionStart); hooks pass events through unchanged and the classify function in hooks/classify.ts extracts only station labels from inputs. drawMap renders the board: pulses fly out on call start and return on completion, errors and refusals turn red, a ring around the centre fills with context-window usage (amber at 70%, red at 90%), and the newest route leg draws itself and fades after 40 seconds plus a 20-second fade. Beneath the map sit a line of what runs now and the last 16 log rows (logRows=fill expands this in the desktop pane), with ●/·/✕/◆ status markers, per-row timings and a red error count until the next user message. For terminals, hooks/scene.ts reads the drawing, hooks/raster.ts paints it and hooks/png.ts packs frames using X.Org misc-fixed bitmap fonts at up to five frames per second; hooks/braille.ts redraws the same board as braille cells. The optional contextCost flag estimates how many tokens each call added, shown one model step late.

  1. Developers who want to see, during a long session, where Claude Code is actually spending effort across files, shell, network and subagents.
  2. Users running Claude Code side by side with a desktop app or VS Code who want a visual sidebar that doesn't clutter the conversation.
  3. Terminal users on kitty, Ghostty or iTerm2 3.7.8+ who want a live picture instead of scrolling logs.
  4. Windows Terminal or Linux terminal users who can use braille mode to get the map without image support.
  5. People debugging hook refusals, near-full context windows or mid-session model switches.
  6. Teams that need to verify a plugin's privacy boundary before installing it.

How do you install or deploy this agent?

At the prompt of a Claude Code terminal session:

/plugin install lot-map --marketplace alexeyw/lot-map

Answer y to add the marketplace, then choose a scope. From a shell:

claude plugin install lot-map --marketplace alexeyw/lot-map

If it says the plugin is not found in any marketplace, add the marketplace first:

claude plugin marketplace add alexeyw/lot-map
claude plugin install lot-map@lot-map

In the desktop app, click + next to the prompt box, choose Plugins, Add marketplace with alexeyw/lot-map, then Add plugin. To update, refresh the marketplace before updating the plugin:

claude plugin marketplace update lot-map
claude plugin update lot-map@lot-map

Then start a new session.

How do you use this agent?

The pane opens by itself when a session starts; with openAtStart off it waits for /lot-map. Each station represents a kind of work, pulses mark call start and end, and red marks errors or refusals. In a terminal the rows show alone by default; to get the map, use full-screen rendering and set terminalView, for example braille dots:

claude plugin configure lot-map@lot-map --values-stdin <<'EOF'
{"terminalView":"braille"}
EOF

Or the picture in a terminal that supports it:

claude plugin configure lot-map@lot-map --values-stdin <<'EOF'
{"terminalView":"picture"}
EOF

The terminal map needs the pane beside the conversation: type /tui fullscreen or start with CLAUDE_CODE_NO_FLICKER=1 claude; inside tmux set allow-passthrough on and focus-events on. On Team or Enterprise logins where notices like lot-map: its classic.PreToolUse hook was skipped appear, turn guardedHooks off:

claude plugin configure lot-map@lot-map --values-stdin <<'EOF'
{"guardedHooks":"false"}
EOF

Every option value is written as a string, including "false" and "2.4"; lists use commas.

What are this agent's strengths and limitations?

Pros
  • One drawing codebase feeds four renderings — desktop pane, VS Code, terminal picture and braille — so most Claude Code environments get the same board.
  • Explicit local-only contract: no network, no file reads, no processes, no model calls, and labels limited to names rather than command lines or content.
  • Installs in one command from the Claude Code plugin marketplace, and every option has a working default.
  • Seven UI languages plus deep appearance options (theme, 13 named colours, layout, motion, text size) for team preferences.
  • contextCost attributes context growth to specific calls and stations, which helps spot what is inflating the window.
Limitations
  • Entirely dependent on Claude Code's plugin and hook mechanism; it cannot run standalone or against other agent runtimes.
  • Terminal map requires full-screen rendering and at least 144 columns for the docked pane; otherwise only log rows appear.
  • Picture mode has hard terminal requirements: iTerm2 3.7.8 or later, with documented memory blow-up (about 17 GB in five minutes) on 3.7.7 (#81).
  • On Team/Enterprise logins, organisation guards skip five hook events, so the Hooks station, hook records, slash-command rows and compaction marks are lost unless guardedHooks is turned off.
  • Built and tested only on Claude Code 2.1.321 on macOS; Windows and Linux support rests on user reports, not the project's own testing.
  • The marketplace tracks the repository's main branch, so updates bring whatever is there and require reviewing the diff.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
lot-map This agent 72 · Some gaps — ★ 1k 1d ago TypeScript Claude Code
I Have ADHD Skill 54 · Major gaps Agent plugin / skillFree ★ 62k 1mo ago Python Claude Code
OSS Autopilot 80 · Good Agent plugin / skillFree ★ 30 today TypeScript Claude Code · Claude.ai
Jev Review 62 · Some gaps Agent plugin / skillFree + model costs ★ 253 2mo ago TypeScript Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Some gaps
72/ 100 5-point scale 3.6 / 5
Trust 18/29
Reliability 8/14
Adaptability 16/18
Convention 15/18
Effectiveness 10/13
Verifiability 5/8
Why each dimension lost points
Trust18 / 29 · 3.1/5

README and SECURITY.md state the mod only observes, makes no network requests, reads no files, starts no processes, and define what counts as a vulnerability; PRIVACY.md and PRINCIPLES.md are referenced but not in the packet, so data-flow transparency scores 3 while sensitive-data handling scores 2. Least privilege has a guardedHooks toggle and a CI surface-check script, but the actual hook list and permission scope are not shown, deducted. User confirmation: no explicit confirmation flow at install/enable beyond /plugin install and a config screen, scored 1. External effects and rollback: no destructive defaults, but no uninstall/rollback guidance, scored 1. Dependency security: CI pins action commit SHAs and Node 22, but runtime dependency manifest is absent, scored 2. Source attribution: plugin.json has author, homepage, repository, supportUrl, but publisher is unverified by FollowAgents, scored 2.

Reliability8 / 14 · 2.9/5

README is detailed and internally consistent (contextCost one-step lag, braille limitations), but dependency availability is not evidenced—no package.json or runtime deps shown, and CI depends on a pinned @anthropic-ai/claude-code 2.1.321, scored 1. Failure messages: red dots, ✕ marks, and hook-refusal first line are described, but concrete error text is not shown, scored 2.

Adaptability16 / 18 · 4.4/5

Audience and scenarios cover desktop, terminal, VS Code, Windows/Linux/macOS, 7 languages, scored 3. Capability boundaries are very clear (what is not drawn, not read, here-documents not parsed), scored 3. Trigger precision: shell commands classified by name with known misclassification (python -c shows as Shell), scored 2. Environment fit: iTerm2 version detection, tmux config, terminal width requirements all documented, scored 3.

Convention15 / 18 · 4.2/5

Information architecture is excellent with clear README sections, scored 3. Install notes cover terminal, shell, desktop app, and update flow, scored 3. Naming stability: plugin and marketplace names align, but version 0.12.0 is pre-1.0, scored 2. Examples and FAQ: config examples and screenshots exist but no dedicated FAQ, scored 2. Known limitations are thorough (braille omissions, iTerm2 3.7.7 issue #81), scored 3. License MIT complete, scored 3. Versioning/changelog: CHANGELOG.md referenced but not in packet, scored 2. Maintenance responsibility: SECURITY.md states solo maintenance and latest-version-only fixes, but publisher identity is unverified, scored 2.

Effectiveness10 / 13 · 3.8/5

Output usability is high: map, row list, context ring, token estimates, multi-dimensional visualization, scored 3. Marginal value: richer spatial view than Claude Code's built-in status line, but incremental, scored 2. Cost-benefit: high thresholds (144 columns, fullscreen, iTerm2 3.7.8+), scored 2.

Verifiability5 / 8 · 3.1/5

claim_traceability: README references PRIVACY.md, PRINCIPLES.md, FONTS.md, CHANGELOG.md, none in the packet, so claims cannot be traced, scored 2. cross_source_corroboration: README, plugin.json, SECURITY.md, and CI are mutually consistent, scored 2. fact_inference_separation: README explicitly marks contextCost as an estimate versus fact, scored 2.

Risks and how to mitigate them
  • PRIVACY.md, PRINCIPLES.md, FONTS.md, and CHANGELOG.md are referenced but absent from the packet, so privacy and safety claims cannot be verified.
  • Publisher is unverified by FollowAgents; identity is unknown and must not be used to infer reliability.
  • Runtime dependency manifest and actual hook list are not shown, so least privilege and dependency security cannot be fully confirmed.
  • No uninstall/rollback guidance, and only the latest version receives fixes.
  • High terminal thresholds (144 columns, fullscreen, iTerm2 3.7.8+) limit functionality in some environments.
Evidence confidence: Low Reviewed Nov 19, 2026 Reviewed revision e3a93ec01f72
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/lot-map.svg)](https://followagents.com/en/agents/lot-map)
<a href="https://followagents.com/en/agents/lot-map"><img src="https://followagents.com/badge/lot-map.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v0.12.0 lot-map 0.12.0
    Release notes excerpt
    - New option openAtStart: with false the pane stays closed when a session starts and opens on /lot-map alone. The map is kept from the session's first call all the same, so it is whole when it is opened. true, as before, stays the default. Suggested by Thombutter. (#116, #117) - The terminal picture is back in iTerm2, from iTerm2 3.7.8 on. iTerm2 3.7.7 kept every picture it was sent, so since 0.9.0 the mod drew none there; iTerm2 fixed that on its side. Measured in 3.7.8: 1466 frames in four and a half minutes left its memory between 327 and 351 MB. An earlier iTerm2, or one that does …
    View on GitHub ↗
  2. v0.11.0 lot-map 0.11.0
    Release notes excerpt
    - New option contextWindow: with compaction the ring and its percentage measure the context against the auto-compact window, such as one set with autoCompactWindow, where it is smaller than the model's; at 450k tokens of a 1M model with compaction at 600k it reads 75%, not 45%. model, as before, stays the default. For this the mod asks Claude Code how the context divides up, at a session's first model step and again after a compaction or a change of model, and takes only the window's size from the answer; PRIVACY.md says so. Contributed by philarete173. (#97, #101) …
    View on GitHub ↗
  3. v0.10.0 lot-map 0.10.0
    Release notes excerpt
    - New option serverStations: put an MCP server's tools on a station of your choice, each as server=station, with files, shell, web, browser, agents, skills or you. The row still reads server: tool. Memory keeps memoryServers, which wins if a server is in both. Contributed by arandevcode. (#55, #86) …
    View on GitHub ↗

FAQ

Does it send my code or commands anywhere?
No. The plugin states it makes no network requests, reads no files, runs no processes, calls no model and adds nothing to the conversation. It extracts only name-like labels from tool calls: the last two path segments, a URL host, a command description and similar. The only cross-session state is the last model name per folder, stored in Claude Code's own plugin store.
Can I see the map without the desktop app?
Yes, with conditions. The terminal shows log rows by default; terminalView: picture needs a terminal that renders images (kitty, Ghostty, iTerm2 3.7.8+) plus full-screen rendering, while terminalView: braille works in any terminal whose font has braille, such as Windows Terminal or most Linux terminals. Either way the pane must sit beside the conversation, with a body of at least 50 columns by 25 rows.
We use a Team or Enterprise login — will it work?
It installs, but organisation guards skip five hook events and the transcript shows notices like lot-map: its classic.PreToolUse hook was skipped. Setting guardedHooks to false silences them at the cost of the Hooks station, hook-added records, the slash-command row and the compaction mark.
How accurate is the token figure from contextCost?
It is an estimate shown one model step late: the plugin only learns how much the context grew between model steps and assigns that to the calls whose answers went with the step. Two or three calls in one step share the growth evenly and are marked ~; subagents keep their own context and get no figure.
Does it work with agents other than Claude Code?
No. It is a Claude Code plugin that depends on that product's marketplace and hook events (session, prompt, tool call, compaction and others); the repository documents no adapter for other agents or model providers.
Is lot-map open source?
Yes. The source code is on GitHub (alexeyw/lot-map) under the MIT licence.
Is lot-map safe to use?
FollowAgents' FARS review (Nov 19, 2026) scored lot-map 18/29 on trust and safety. None of the eight safety controls the review checks for was missing outright.
Is lot-map still maintained?
Yes. The latest commit on GitHub is from 2026-11-19, within the last 30 days. The latest release is v0.12.0 (2026-11-19).
Who makes lot-map?
lot-map is published on GitHub by the account alexeyw. FollowAgents has not verified which person or company is behind that account.
What is lot-map's official website?
The project's GitHub repository lists https://alexeyw.github.io/lot-map/ as its website. That link is set by the repository owner; the source code itself is on GitHub (alexeyw/lot-map).
Was this page helpful?
View on GitHub ↗ Install ↓

Host ecosystems

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts