alexeyw/lot-map

What it does

lot-map is a Claude Code plugin that adds a pane beside the chat and visualizes the session as a small map. Instead of scrolling a command list, you see a pulse travel from the centre to a station for each activity: model requests, memory access, file operations, shell commands, network calls, browser tools, subagents, skills and hooks. Each station gets brighter as it is used, so a session leaves a visible trace of where the work went. The demo animation is replayed through the plugin's own drawing code (demo).

It is meant for people who already use Claude Code and want to observe agent behaviour in real time: which tools fire, when a hook blocks a call, when subagents are running and how full the context window is.

Install and start

Install the plugin from the terminal inside a Claude Code session:

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

Answer y to add the marketplace and pick a scope. From a shell you can run the equivalent:

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

In the desktop app, use + next to the prompt box, then Plugins, Add marketplace with alexeyw/lot-map, then Add plugin.

The pane opens automatically when a session starts. In a terminal it needs at least 144 columns and renders text-only; otherwise type /lot-map. Configuration is available during install or later via /config or claude plugin configure, followed by a session restart.

Reading the map

A thin route line links the last stations the main loop passed through, and spokes thicken with usage. Two clock hands show call duration (one minute and ten minutes per revolution), parallel calls appear side by side with counts such as ×3, and errors or refusals leave a red dot until your next message. The ring around the centre tracks context-window fill, amber at 70% and red at 90%. Subagents circle the Agents station, and the feed below the map keeps the last 16 rows. The README documents each station and visual cue (reading the map).

Options and localisation

Labels come in English, Russian, Spanish, French and German via the language option. Other settings cover the centre label, which MCP servers or shell commands light the Memory station, memory path parts and labels, hook record/reason display, theme and layout. Lists are comma-separated, and options can be set non-interactively, for example:

echo '{"language":"ru","memoryServers":"my-memory"}' | claude plugin configure lot-map@lot-map --values-stdin

Privacy and constraints

The plugin makes no network requests, reads no files, runs no processes and calls no model, and it does not add to the conversation. It does receive every tool call's full input, including command lines and file text, but only extracts names; classification lives in classify in hooks/map.ts. Labels hold file path tails, host names, command descriptions or names, memory identifiers and subagent descriptions, never contents, and control characters are stripped. Across sessions it stores only the last model name per folder. A map fault never reaches tool calls or model turns.

A marketplace added from this repository follows its main branch, so updates bring whatever is there. It was built and tested on Claude Code 2.1.321 on macOS; Windows support is reported by users. Development checks are claude plugin test . and claude plugin validate ..

HotRepo's take

This is a young project: it entered the new-project board for TypeScript on 2026-11-11 with 110 stars, 8 of them gained in a roughly one-hour window, with no GitHub all-language daily rank recorded. The useful signal is not the absolute count but the design: it treats agent observability as an animated map with its own station vocabulary, and the README goes out of its way to constrain what the plugin extracts from tool calls. That specificity makes it a practical instrument for inspecting Claude Code sessions rather than a generic dashboard.