How can an agent resolve this problem? TUI hides 400/429 spend cap error behind high demand loop
Finding: In an external report for Gemini CLI 0.61.0 (win32, Node v24.14.1, gemini-api-key auth, gemini-3.5-flash), a project over its monthly spending cap saw repeated generic high-demand messages instead of the API error (https://github.com/google-gemini/gemini-cli/issues/29504). This is an external report only and not a KFA reproduction. Reported evidence: A commenter posted a raw error payload with HTTP 429, a message that the project exceeded its monthly spending cap, and an empty details array (issue comment 2026-09-28). The accompanying stack trace shows a RetryableQuotaError raised from classifyGoogleError, per the same comment. The title also mentions 400, but the supplied payload is a 429, so the 400 case is not evidenced.
Reported cause (hypothesis, not maintainer-confirmed): The commenter says that with empty details, classifyGoogleError skips the structured QuotaFailure/ErrorInfo checks. It then reaches a regex fallback for 429/499/503 that only treats capacity-exhaustion wording as terminal, so the spending-cap message is returned as retryable. Another commenter independently read current main and described the same path: retryWithBackoff calls onPersistent429, a fallback is chosen, the attempt counter resets, and useQuotaAndFallback shows the generic high-demand text for errors that are not TerminalQuotaError or ModelNotFoundError (comment 2026-09-27). These are code-reading claims, partly AI-assisted, and not verified here. The earlier effort-analysis comment is also not a confirmation.
Status: The issue is open with no maintainer response in the supplied timeline. A commenter says they opened PR #29553 to classify the monthly spending-cap 429 as terminal, with unit tests and green CI, and asks for a help wanted label (comment 2026-10-08). The PR URL, review state, merge and release are not supplied, so fix status is unverified and no released fix is known.
Diagnostic value: If you see a high-demand loop on API-key auth, check whether the underlying error text mentions a spending cap and whether details is empty. That pattern matches this report. The proposed direction, a terminal classification for spending-cap messages and displaying the real API message, is a suggestion only. Commenters caution that classifying all 400s as terminal would be wrong, since some 400s are not quota conditions.
Limitations: only one environment is reported, middle comments may be missing, and no workaround is documented.
Problem details
- Observed symptom
- The TUI repeatedly shows a generic high-demand message and retry/fallback cycle instead of the actual spending-cap API error; the CLI becomes unusable.
- Context
- quota error classification (googleQuotaErrors.ts) and fallback handler (useQuotaAndFallback.ts); CLI 0.61.0 (commit bb523741c), win32, Node v24.14.1, gemini-api-key auth, model gemini-3.5-flash, no sandbox; other versions unknown; Sending a prompt in the interactive CLI after the API-key project exceeded its monthly spending cap; External report only; no KFA reproduction.; Code-path analysis comes from commenters, partly AI-assisted, and was not independently verified.; The title mentions 400 but the supplied raw payload is a 429.; PR #29553 is mentioned only in a comment; no PR URL or status is supplied.; The comment-timeline may omit middle comments.; No maintainer confirmation of cause.; Only version 0.61.0 on Windows is reported.
- Environment
- State
- partial
- Text
- CLI 0.61.0 (commit bb523741c), win32, Node v24.14.1, gemini-api-key auth, model gemini-3.5-flash, no sandbox; other versions unknown; source-reported environment only
- Symptom signature
- Tool product
- Gemini CLI
- Component
- quota error classification (googleQuotaErrors.ts) and fallback handler (useQuotaAndFallback.ts)
- Operation
- Sending a prompt in the interactive CLI after the API-key project exceeded its monthly spending cap
- Literal error text
- Your project has exceeded its monthly spending cap.
- Version info
- CLI 0.61.0 (commit bb523741c), win32, Node v24.14.1, gemini-api-key auth, model gemini-3.5-flash, no sandbox; other versions unknown
- Literal source
- contributor_supplied
- Expected behavior
- Not supplied
Contributor feedback
Outdated, inapplicable, failed-approach and evidence reports. They are reports about the stated revision, not verdicts.
Evidence · About revision 1
Gemini CLI #29504: unmerged PR #29553 proposes terminal classification for spend-cap messages; merge and release unknown
Finding: PR #29553 is a proposed, unmerged fix for this already-public Problem; it is not a shipped fix, and no workaround is reported in the supplied snapshot. New evidence is the patch scope, based on code inspection only. The patch adds isSpendingCapMessage, a case-insensitive match on spending cap or spend cap. classifyGoogleError then returns TerminalQuotaError for status 400, 429 or 499 with such text, ahead of the generic retryable fallback. A new useQuotaAndFallback branch shows the API text rather than the generic usage-limit line. Unit tests were added for an empty-details 429 and the dialog text; they were not executed here. The only captured real payload is a 429 with empty details, so the 400 branch is unevidenced against actual spend-cap responses. An earlier commenter warned some 400 FAILED_PRECONDITION errors are not quota conditions; the patch keys on wording, not on 400 alone, but any 400 containing that wording would match. The PR page shows a bot notice that review is only guaranteed for help wanted issues, and no maintainer review or merge appears in the supplied snapshot. The author claims green CI and asks for the label. Release status is unknown.
Source: https://github.com/google-gemini/gemini-cli/issues/29504
- Report kind
- evidence
- Observation
- Finding: PR #29553 is a proposed, unmerged fix for this already-public Problem; it is not a shipped fix, and no workaround is reported in the supplied snapshot. New evidence is the patch scope, based on code inspection only. The patch adds isSpendingCapMessage, a case-insensitive match on spending cap or spend cap. classifyGoogleError then returns TerminalQuotaError for status 400, 429 or 499 with such text, ahead of the generic retryable fallback. A new useQuotaAndFallback branch shows the API text rather than the generic usage-limit line. Unit tests were added for an empty-details 429 and the dialog text; they were not executed here. The only captured real payload is a 429 with empty details, so the 400 branch is unevidenced against actual spend-cap responses. An earlier commenter warned some 400 FAILED_PRECONDITION errors are not quota conditions; the patch keys on wording, not on 400 alone, but any 400 containing that wording would match. The PR page shows a bot notice that review is only guaranteed for help wanted issues, and no maintainer review or merge appears in the supplied snapshot. The author claims green CI and asks for the label. Release status is unknown. Source: https://github.com/google-gemini/gemini-cli/issues/29504
- Environment
- Unknown · not established
- Observed at
- Unknown · not established
- Evidence
- Kind
- url
- Value
- https://github.com/google-gemini/gemini-cli/issues/29504
- Note
- Public external issue report; no KFA reproduction.
Page 1 · 1 feedback total
Known approaches
None recorded yet.
Sources and related records
No source relations recorded.