Last week I ran out of my Claude Max window halfway through a feature. I had Codex, and a second Claude account, and neither could see anything the first session knew. The plan, the decisions and my half-written code were stuck in a transcript no other tool can read.
That is not a bug in any one tool. Every coding agent keeps its session in its own store. So the only place that every agent can read is the repository.
SpecWeave 3.0 keeps three things in git for each piece of work:
And one AGENTS.md of about 760 tokens that Claude Code, Codex, Grok Build, Gemini CLI and Cursor all read.
In the tool you're leaving:
specweave handoff --reason "out of tokens"
It releases your task claims, writes the handoff note, scrubs secrets, and pushes your branch plus a snapshot of the uncommitted diff. Your working tree is left as it was.
In the next tool:
specweave pickup
It fetches the handoff, applies the edits when your checkout is clean, and prints the next task with its acceptance criteria, files and test. The agent starts with context instead of "let me explain this project again".
specweave auto-handoff on
On your own machine this hands off at 90% of the 5-hour or weekly window in Claude Code and Codex, and right after a rate limit in Grok Build. Cloud sessions have no usage meter to read, so there you just say "hand off".
It can't read another tool's transcript. Nobody can. It only works because the important parts were never in the transcript to begin with.
npm i -g specweave@3
specweave init
It's open source: github.com/anton-abyzov/specweave. The full guide is at spec-weave.com, and the usage-limit walkthrough is at spec-weave.com/docs/guides/claude-code-usage-limit.
How do you handle a limit mid-task today? A HANDOFF.md by hand, or just wait for the reset?