[Bug] Subagents billed after monthly spend limit exceeded; false clean review result on agent failures

Bug Description
Subject: Fable credit adjustment request — subagents billed while failing on an already-hit spend limit Summary (with the honest cause up front): On 2026-07-08, three xhigh multi-agent code-reviews were run on Claude Fable 5 in a Claude Code ship workflow. Running these on Fable was an operator mistake — established practice here is to run heavy multi-agent reviews on Opus, and that was not followed. Each review fans out ~14 subagents; three launched in parallel (~42 agents) exhausted the monthly Fable spend limit mid-run. I'm still filing this because two platform behaviors turned a bad launch decision into wasted spend and a misleading result: 1. Billing on doomed launches. Once the spend limit was reached, additional subagents still launched and immediately failed with "You've hit your monthly spend limit" — and that failed spend still appears to count against the budget. Subagents that cannot do any work because the budget is already exhausted should not bill. 2. False "clean" review result. The review workflow then returned "No findings survived verification" — a clean pass — synthesized from a run in which most agents had died on the spend limit (agents_error >= 6 per run; finders and/or verifiers never completed). A code-review tool reporting a clean result when the run actually crashed is a real hazard; it should surface an inconclusive/failed state whenever agents_error is nonzero, not an empty findings list. Evidence: three review runs on 2026-07-08 — task IDs wpag33m0p, wzztqwyzb, wj1fwiglv — each with agents_error >= 6 and "monthly spend limit" failure messages in their failures block. Ask: - Credit adjustment for the Fable spend consumed by subagents that failed on the already-hit spend limit (issue 1). - Fix consideration for the false-clean-result reporting behavior (issue 2). That's the full report in a code block — copy it straight into the bug report. If you'd rather have it without the em-dashes or with your account/email added, say the word and I'll adjust.

Environment Info

• Platform: darwin
• Terminal: xterm-256color
• Version: 2.1.202
• Feedback ID: 333f6feb-bf86-470a-aa9f-5b91be106df5

Errors

[]