Get a heads-up before your coding agent hits its limit.

It warns you at 80% and 95% of your usage window, then writes a "Where I left off" note so another agent can carry on.

Join the listWorks with Claude Code and Codex.

Runs on your machine. No account, no telemetry, makes no network calls.

~/app · claudeexample session
  • refactor the auth middleware so sessions are checked in one place
  • Reading src/middleware/auth.ts
  • Reading src/routes/account.ts
  • Plan: 5 steps. Starting with auth.ts.
5-hour window18%

Limit Rescue is watching 18%

You're forty minutes into a refactor.

The agent knows which files you've touched, why that one test is skipped, and the approach you already tried and dropped.

It's going well.

Meanwhile the number you weren't watching is getting close to the line.

So Limit Rescue taps you on the shoulder.

While there's still time to finish the thought you're on.

~/app · claudeexample session
  • Edited src/middleware/auth.ts +42 -18
  • Edited src/routes/account.ts +6 -31
  • Tried per-route guards. Dropped: same check in 9 routes.
  • Skipped billing.spec.ts › renews session (flaky on CI)
  • Tests: 212 passed, 1 skipped
  • Edited src/routes/admin.ts +4 -22
Limit RescueHeads up: you're getting low. Good moment to wrap up this step.
5-hour window91%

Usage limit reached.

Resets in 3h 41m. Or open another agent and spend twenty minutes explaining everything again.

Unless something wrote it down.

A handoff note, so the next agent isn't starting from zero.

Saved in your repo at 95%, with the git facts. Pick up in whichever agent you've still got.

Claude Codelimit reached
  • Edited src/routes/admin.ts +4 -22
  • Tests: 212 passed, 1 skipped
Codex
  • continue from the handoff note
  • Read handoff note (5 lines)
  • Picking up at: move the check out of settings.ts
  • Editing src/routes/settings.ts
5-hour window4%

How it works

One command sets it up. After that it runs in the background while you work.

  • Reads real usage from files Claude Code and Codex already write on your machine.
  • Warns once at 80% and once at 95%, in the statusline, as a desktop notification and in the terminal.
  • At 95% it gets the running agent to write the note, then adds the git facts itself.
limitrescue
limitrescue initSet up Claude Code, plus Codex if it is installed. Backs up first, only appends.
limitrescue statusUsage for Claude and Codex, from local files.
limitrescue handoffWrite the note now, from the transcript.
limitrescue resume <codex|claude>Open the other agent in the same repo, pointed at the note.
limitrescue doctorCheck the setup.

privacy checkIt makes no network calls, and the tests check that two ways. One scan fails if any source file imports a networking module or calls fetch. The other runs every command with all of Node's network functions replaced by traps that fail and log.

Where it works

SetupWhat you get
Claude Code, in the terminalWarnings and the handoff note
Codex, in the terminalWarnings and the handoff note
Codex, in the ChatGPT desktop appWarnings and the handoff note
Claude desktop appA fallback note when the limit hits, and a rough estimate in status. No live usage numbers: the app doesn't run the statusline, and Limit Rescue doesn't call any usage endpoint to fill the gap.

Tested on macOS. Windows and Linux: not tested yet. Needs Node 18.17 or newer. Claude's usage numbers need Claude Code in a terminal on a Pro or Max plan, after the first reply.

Questions you might have

Can I use it today?

It works, but it isn't public yet. We're letting a few developers in first. Join the list and we'll send you the install.

Does it send anything anywhere?

No. It runs on your machine, with no account and no telemetry, and it makes no network calls. It only reads files Claude Code and Codex already write locally. The resume command starts claude or codex, which talk to their own servers as they normally do.

Which agents does it work with?

Claude Code and Codex, in the terminal. Codex also works in the ChatGPT desktop app. The Claude desktop app gets a fallback note when the limit hits and a rough estimate, but no live usage numbers. Other agents aren't supported yet.

What happens when I get close to the limit?

At 80% and 95% of the window you get one warning each, in the statusline, as a desktop notification and in the terminal. At 95% it asks the running agent for a "Where I left off" note and saves it in your repo at .limitrescue/handoff.md. If the agent doesn't answer, or the limit hits first, it writes a plainer note from the transcript and marks it as a guess.

Will it mess with my Claude Code or Codex setup?

It's built not to. Setup backs up your files first and only appends. It never overwrites a statusline you already have, and uninstall removes only what it added. It never edits CLAUDE.md or AGENTS.md. After setup, restart Claude Code. Codex asks you to trust the new hook once: run codex in a terminal and type /hooks.