[juliusmarminge]

Claude now hands out two kinds of banked reset: a weekly reset and a Fable reset. The Fable reset refills only the Fable allowance and gives back what Fable used on the weekly limit; it does not reset the whole week. T3 Code added every grant into one count and always redeemed whichever grant Claude listed next, so "Use reset" could spend the Fable reset when you wanted the weekly one, and the dialog promised to clear your windows either way.

Fix

• Server. claudeResetCredits groups the cedar_ember grants by kind, using the same rule as the Claude CLI (2.1.322):

  • a grant that gives back weekly usage is a Fable reset;
  • otherwise its declared type decides;
  • otherwise what it clears decides (5-hour, weekly or full).

  Each kind becomes a reset-credit option with its own grant, count, expiry and description.

• Redeeming. The instance consume RPC takes an optional creditId. The Claude driver redeems that option's grant, after checking it is still offered. As before, one request ID per login is kept until Claude answers, so a retry after an unconfirmed claim resends the same request ID instead of starting a new claim.

• Clients. Web (composer, usage page and pool popover) and mobile show one row per kind, each with its own Use button. The confirm dialog says what that reset refills. The shared resetCreditChoices helper sits in @t3tools/shared/usageLimits.

• Compatibility. The new contract fields are optional. Older clients keep the single total and next-grant redemption, and Codex has only one kind of reset, so it keeps the single-button UI.

Evidence

Both builds ran with the same temporary fixture: one weekly grant and one Fable grant, parsed through claudeResetCreditsToContract. Claude on the capture machine routes through a proxy, so it returns no real grants.

	Before	After
Usage → Limits, weekly segment
Confirm dialog

The "before" row shows the Fable grant's expiry (118 days) under a generic "2 banked", because the next grant Claude listed was the Fable one.

Verification

• claudeResetCredits.test.ts has a new case: separate weekly, Fable and 5-hour options, each pinned to its own grant. It fails on main.
• vp run check:changed passes: format, lint, knip, typecheck and the affected tests.

Limits

• I have not checked these mappings against a real Fable response. The grant shape comes from the Claude CLI's own schema.
• Claude accounts pooled through a CLIProxyAPI hub still can't redeem resets, as before.
• The mobile change is the same per-kind row as web. It was typechecked but not run on a device.
• As before, a grant Claude marks unusable right now (for example, one that only works once you hit a limit) isn't listed.

🤖 Generated with Claude Code (Claude Opus 5.5)

[macroscopeapp]

Approvability

Verdict: Approved at 18e168f

Macroscope's review found this PR approvable — Claude reset credits are now offered by refill type, letting users redeem Fable resets separately from weekly resets. Redemption remains explicitly confirmed and uses the existing account-scoped redemption safeguards.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.5 KiB	21.5 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	1	1	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	1	1	0 (0.0%)	8	✅

Baseline: 43915a4 · PR result: 18e168f · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

Dismissing prior approval to re-evaluate 18e168f

[coderabbitai]

Finishing Touches

Generate docstrings

• Commit to this branch
• Create a new PR

Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts