Keep working when a provider runs out

You end with a thread that continued on another provider under a rule you set for the project.

SpecWithCode
1:39
Claude Code and Codex
Narrated, with captions

Recorded on SpecWithCode 1.0.43 with SpecWeave 3.1.0, 18 October 2026, added 18 October 2026. The current release is SpecWithCode 1.0.44.

Before you start: SpecWithCode for Mac; two signed-in coding agents (Claude Code and Codex in this video); a git repository with a SpecWeave increment planned.

Length 1:39Captions

Steps

Open the thread in the project's right panel. Here it runs on Claude Code and has finished T-01 of increment 0001.
Send the next request. When the provider stops on its usage limit, the thread shows a Usage limit reached banner with the reset time and what the project rule will do, for example Continuing with Codex by project rule.
Under the default rule, Switch automatically, SpecWithCode hands the stopped turn to Codex with its context, in the same worktree. The thread shows Context handoff and Claude Code limit reached, continued on Codex. The warning row then reads Usage limit resolved — chat continued: that means this chat continued, not that Claude Code's limit has reset.
Before you send the next message, check the composer's model picker. In SpecWithCode 1.0.43 it returns to Claude Code a few seconds after the switched turn ends, so the next message would go to Claude Code, stop on the limit and hand off again.
Run specweave task list in the repo. The By column shows which agent finished each task, with the test run that proved it, and git log --oneline shows both agents' commits.
Choose the rule in Project settings, Usage limits. When a provider runs out offers Switch automatically, Offer the fallback and Wait for the reset. With Offer the fallback, the thread waits until you click Continue with Codex.
Use Switch before the limit to move an idle thread early. The default is At 5% left.

Recorded headlessly in one take on a private SpecWithCode 1.0.43 server with a disposable home and a sample repo, 18 October 2026. A real usage limit cannot be forced, so the limit is simulated with the app's documented test switch, STUDIO_SIMULATE_USAGE_LIMIT=claudeAgent, which makes every Claude Code turn stop with a limit that resets in one hour. T-01 was a real Claude Code turn, started off camera through the coordinator before the switch was turned on. T-02 was a real Codex turn, and the handoff, banner and ledger are the app's own. Both agents were signed in with the author's accounts through short-lived tokens copied into that home. Full access was chosen for the demo only. Nothing in Codex's turn is sped up; two still screens (the limit banner and the open plan list) play slower to fit the narration, and nothing moves in them. After the switch the composer's last frame is held for about 2 s, marked Still frame, and the walk to Project settings, Usage limits plays at 3×, marked Sped up 3×. The ledger beat is a captured terminal run of `specweave task list` and `git log --oneline -3`. Its first line, '0001-personal-greeting — agent cli@demo', is left out: cli@demo is the capture shell's own identity, not a third agent. Known issues in 1.0.43, shown in the video: the model picker returns to Claude Code after the automatic switch (the docs say it follows the switch), so the next message bounces back to Claude Code and hands off again; and the row 'Usage limit resolved — chat continued' does not mean Claude Code's limit has reset. Narration is an AI voice (ElevenLabs, the author's cloned voice); captions are timed from the voice service's character timestamps.

Transcript

Built from the caption file of the narration.

0:00 What you will see

Keep working when a provider runs out.

0:03 A thread on Claude Code

This greet-app thread runs on Claude Code. It finished T-01 of the increment, and one task is left.

0:11 Claude Code hits its usage limit

Ask for T-02, and Claude Code stops on its usage limit. Here the limit is simulated with SpecWithCode's test switch. The banner shows the reset time, and that the project rule continues with Codex.

0:25 The thread moves to Codex

Under the default rule, Switch automatically, SpecWithCode hands the stopped turn to Codex, with its context, in the same worktree. The row saying Usage limit resolved only means this chat continued. Claude Code stays limited until the reset.

0:41 Codex finishes T-02

Codex runs npm test and commits the change. Then, unasked, it runs specweave handoff. Its summary: T-02 is done on main, and all four tests pass. The plan shows both tasks done. One catch in this build: the picker then flips back to Claude Code, so check it before your next message.

1:02 Who did what: the ledger

In a terminal, specweave task list reads the ledger: Claude finished T-01, and Codex finished T-02, each with the test run that proved it. Both agents' commits are on the branch.

1:16 Choose the rule

Set the rule per project, under Usage limits: switch automatically, offer the fallback, or wait for the reset. With Offer the fallback, nothing moves until you choose Continue with Codex. An idle thread can also move early, at 5% left.

1:33 Next: sw-done

Next, sw-done verifies and closes the increment.

Download SpecWithCode for Mac Read the reference

Learning path

Run SpecWithCode

Understand the setup, run a project through its coordinator, then add plans, memory and a second provider.

See how SpecWithCode runs your agents
2:54, SpecWithCode
Start SpecWithCode with a coordinator and two workers
1:48, SpecWithCode
Run a project with a coordinator and workers
1:29, SpecWithCode
Plan, verify and schedule project work
1:35, SpecWithCode
Keep project memory and read honest usage
1:06, SpecWithCode
Continue a thread on Codex at a usage limit
1:01, SpecWithCode
Keep working when a provider runs out (this guide)
1:39, SpecWithCode

3 more guides are planned for this path.

Hand off from Claude Code to Codex at a usage limit1:18, SpecWeave CLI
Keep project memory and read honest usage1:06, SpecWithCode

All tutorials