idle-intel

The status line is the most-watched line on a developer's screen while an AI agent thinks. Today it says "Thinking...". It could say something worth knowing.

idle-intel replaces Claude Code's empty wait state with rotating micro-briefings about the project you're already working on:

◈ npm audit: 0 critical, 1 high severity vulns -> run `npm audit fix`
◈ git: 2 unpushed commits on this branch
◈ git: 1 stash sitting around -- still needed?
◈ 2 FIXME markers in the codebase -- oldest debts bite hardest

One line at a time, rotating every 12 seconds. Zero interruption, zero network beyond your own package manager, zero telemetry.

Why

Tools have appeared that fill the agent wait state with ads. This project is the opposite bet: the wait state is valuable because it's empty — the only moment a developer looks at the screen without acting. Filling it with generic content destroys it; filling it with one glanceable fact about your own project respects it. Vulnerabilities you haven't patched, commits you haven't pushed, debt you've stopped seeing.

What it detects

Each collector activates only if its manifest and tool are present, and fails silently otherwise. All facts are local. Nothing is uploaded anywhere.

Architecture

Two-speed design:

• Fast path (statusline.sh, pure bash, ~20ms): reads one line from a cache, rotates by time. Runs on every status line refresh.
• Slow path (collect.py, stdlib-only Python): scans the project and rewrites the cache atomically. Triggered in the background when the cache is older than 30 minutes.

The status line never waits for a scan. Cache lives in <project>/.claude/idle-intel/cache.txt.

See everything at once

The status line shows one insight at a time. To re-scan and see all of them in the chat, run:

/intel

This also works where the status line is not shown, such as the VS Code extension.

Install

As a Claude Code plugin

/plugin marketplace add demartinogiuseppe/idle-intel
/plugin install idle-intel@the-deeper-layer

Then ask Claude Code to "set up idle-intel" — the skill walks it through wiring your settings.json.

Manual (Linux / macOS)

mkdir -p ~/.claude/idle-intel
cp plugins/idle-intel/skills/idle-intel/scripts/* ~/.claude/idle-intel/
chmod +x ~/.claude/idle-intel/statusline.sh

Add to ~/.claude/settings.json:

"statusLine": {
  "type": "command",
  "command": "~/.claude/idle-intel/statusline.sh"
}

Manual (Windows)

Windows needs a .cmd wrapper because Claude Code runs status line commands outside Git Bash, where quoting nested paths with spaces ("Program Files") breaks, and where C:\Windows\System32\bash.exe (WSL) shadows Git's bash. Create statusline.cmd next to the scripts:

@echo off
"C:\Program Files\Git\usr\bin\bash.exe" "C:\path\to\idle-intel\scripts\statusline.sh"

And point settings.json at the wrapper with forward slashes:

"statusLine": {
  "type": "command",
  "command": "C:/path/to/idle-intel/scripts/statusline.cmd"
}

Requirements: git (ships with Git Bash), python 3.8+. npm and pip optional.

Test before you trust

The test-kit/ folder contains make-demo.sh: it builds a disposable demo project containing seven planted defects (a real lodash CVE, unpushed commits against a local fake remote, a stash, a backdated stale branch, known TODO/FIXME counts) so you can verify every collector against a known-answer checklist:

bash test-kit/make-demo.sh
bash plugins/idle-intel/skills/idle-intel/scripts/test.sh ./idle-intel-demo

Expected: 7/7 insights, statusline execution well under 300ms. Tested on Linux and Windows 10/11 (Git Bash).

Extending

Collectors are plain functions in collect.py: (root: Path) -> list[str], registered in one tuple, never allowed to raise. PRs welcome for Cargo.toml / go.mod support, license-conflict detection, or changelog highlights for recently bumped dependencies.

Uninstall

Remove the statusLine block from ~/.claude/settings.json. Delete <project>/.claude/idle-intel/ caches freely.

License

MIT