Astra usage limit reached again in under an hour after one reset during a travel-site workflow (GJC)

What issue are you seeing?

I am reporting unexpectedly rapid Astra allowance consumption and asking for an investigation into the usage accounting.

I used one usage-limit reset, but the renewed allowance did not last even one hour before I hit the limit again. The work I was doing was planning a trip, checking the information, turning it into an HTML website, and publishing/updating that website. Having to spend a reset and still being unable to sustain even an hour of this workflow feels unreasonable and makes the service difficult to rely on.

This is my observed usage-limit experience, not a verified measurement of incorrectly billed tokens. I do not have an exact token total or a before/after usage-meter export to attach, and I cannot yet identify which quota window or accounting component caused the depletion.

Client and environment

• Client used for the affected work: GJC, a third-party coding agent. The currently installed version is gjc/0.16.4.
• Platform: Windows 11.
• Model: Astra (identified as gpt-6-astra in the workflow).
• Date reported: September 6, 2026, KST.
• This is not a verified reproduction in the official Codex Windows app. I am explicitly identifying the third-party client so that the attribution is not misleading.
• The subscription tier, effective reasoning/service tier for the entire affected period, and server-side token breakdown have not been independently captured for this report.

What steps can reproduce the bug?

These describe the affected workflow; repeatability and the underlying cause are not yet confirmed:

1. Use Astra through GJC for a travel-planning project.
2. Research/check the plan, organize it into static HTML pages, validate the pages, and publish/update the site.
3. After reaching the usage limit, apply one available usage reset.
4. Continue the same project work and observe that the allowance becomes unavailable again in less than one hour.

The agent performed tool calls and delegated some work. I am not claiming this was a single short prompt, and I have not isolated context re-sends, parallel agent activity, retries, compaction, or other account activity from the account-level deduction.

What is the expected behavior?

Usage should be predictable enough to complete a practical planning-and-static-site workflow. A reset followed by another limit in under an hour needs an understandable explanation, rather than leaving the user guessing whether Astra's consumption is intended or an accounting defect.

Please investigate:

1. Whether the allowance deductions and the effect of the usage reset were calculated correctly.
2. How Astra model weighting, input/cache accounting, reasoning output, context re-sends, retries, and agent activity contributed to the deduction. These are questions to investigate, not confirmed causes.
3. Whether a third-party client such as GJC could be causing excessive requests or losing expected caching behavior, and what diagnostics can distinguish that from a backend accounting issue.
4. If this is intended consumption, please expose a clear per-task breakdown and review whether the effective allowance is practical for this workload.
5. If incorrect metering is confirmed, please review restoration of the affected allowance/reset.

Additional information

Related reports describe rapid Astra allowance depletion, but I am not assuming they share the same cause: #42987 and #43029. This report adds the specific combination of a third-party GJC workflow and less than one hour of use after an explicit reset.

No credentials, private travel details, reservation identifiers, or private repository contents are included here. Please advise on a supported private diagnostic channel for any account-specific investigation, or route this to the appropriate quota/support team if it cannot be investigated in this repository.

The report names no repository files, tests, or code entry points, and it does not provide token totals or usage-meter exports. Start by determining whether this belongs with the quota or support team and gather account-level diagnostics; done would require identifying the deduction cause and providing an explanation or confirmed correction.