Agent toolbox

What it does

Four small tools. A weekly scheduled scout finds new open-source plugins, skills and MCP servers for AI coding agents, compares them with what is installed, and writes a report. It installs nothing. A menu-bar gauge shows live subscription usage limits. A native app watches which AI coding agents are running and what they are doing. A read-only visualiser shows the memory store as a live network.

How it works

• The scout runs every Sunday. It searches for new agent tools, compares against installed ones, and writes a report. It installs nothing.
• The usage gauge polls the CLI for server-side usage figures and shows a live ring and percentage in the menu bar.
• The agent watcher tails transcripts, scans running processes, and shows a live dashboard of sessions and events.
• The memory visualiser opens the memory database read-only and renders a live network of nodes and links.

Guardrails and design choices

• The scout installs nothing. It only writes a report.
• The usage gauge asks the server for authoritative figures rather than computing a local estimate.
• The agent watcher exposes a local-only HTTP API. It is not reachable from other machines.
• The memory visualiser opens the database read-only. It cannot corrupt, lock or migrate the store.

Status

Built and running locally. The scout runs weekly. The gauge, watcher and visualiser are menu-bar or window apps. None are distributed.