fix: auto-fix CI formatting issues for PR #873

[github-actions]

CI Auto-Fix

Original PR: #873
Failed CI Run: PR Build Check

Fixes Applied

File	Fix	Type
MetadataTab.tsx	Import ordering (alphabetical)	format
SummaryTab.tsx	Import ordering (alphabetical)	format
price-list.tsx	Import ordering (alphabetical)	format
response-handler.ts	Import ordering + let to const	format
session.ts	Import ordering (alphabetical)	format

Not Auto-Fixable

No errors that couldn't be safely auto-fixed.

Verification

• `bun run typecheck` passes
• `bun run format:check` passes
• No logic changes made

---

Auto-generated by Claude AI

Greptile Summary

This PR is an automated CI formatting fix for PR #873, reordering imports alphabetically across 5 files and converting one let declaration to const in response-handler.ts. All import changes are safe formatting-only fixes; however, the let → const change on inTailMode in response-handler.ts surfaces a pre-existing logic issue worth addressing.

Key changes:

• MetadataTab.tsx, SummaryTab.tsx, price-list.tsx, session.ts: Pure import reordering, no functional impact.
• response-handler.ts: Import reordering + let inTailMode = false → const inTailMode = false. The const change is lint-correct since the variable is never reassigned — but this also reveals that inTailMode is permanently false, meaning joinChunks() always early-returns with only the head buffer's content. Tail data accumulated for large streaming responses (> 1 MB) is buffered but never included in the assembled SSE text used for cost/usage extraction and fake-200 detection. This is a pre-existing bug from PR feat(pricing): resolve provider-aware billing for multi-provider models #873 that this PR makes structurally explicit.

Confidence Score: 2/5

• Formatting changes are safe to merge, but a pre-existing logic bug in response-handler.ts is exposed and warrants resolution.
• Four of five files have purely mechanical, risk-free import reorderings. The fifth file (response-handler.ts) has an additional let → const change that is lint-correct (variable is never reassigned) but highlights a pre-existing logic flaw: inTailMode is always false, causing joinChunks() to silently discard all tail buffer data for large streaming responses (> 1 MB). This can lead to missed cost/usage events. The bug is pre-existing (not introduced by this PR), but since this PR makes it structurally permanent with const, it should be addressed before or alongside merging this change into the feature branch.
• src/app/v1/_lib/proxy/response-handler.ts — address the inTailMode logic issue.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Incoming stream chunk] --> B{headBufferedBytes < MAX_STATS_HEAD_BYTES?}
    B -- Yes --> C[Push to headChunks]
    B -- No --> D[Push to tailChunks via pushToTail]
    D --> E{tailBufferedBytes > MAX_STATS_TAIL_BYTES?}
    E -- Yes --> F[Evict oldest tail chunks\nwasTruncated = true]
    E -- No --> G[Tail buffered]

    H[joinChunks called] --> I{inTailMode?\nalways false}
    I -- false\nalways --> J[Return headText only\ntail data IGNORED]
    I -- true\nnever reached --> K[Return headText + tailText\nwith truncation marker]

    style I fill:#f96,stroke:#c00,color:#000
    style J fill:#f96,stroke:#c00,color:#000
    style K fill:#9f9,stroke:#090,color:#000

Loading

_{Last reviewed commit: cba447b}

Greptile also left 1 inline comment on this PR.

[greptile-apps]

inTailMode is always false — tail data is never joined

The CI lint fixer correctly changed let to const because this variable is never reassigned. However, this surfaces a pre-existing logic issue: inTailMode is permanently false, which means joinChunks() (line 1164) always takes the early-return branch and returns only headText — the tail buffer is populated when the head fills beyond 1 MB, but its contents are never included in the assembled output used for stats/cost parsing.

For large streaming responses (> 1 MB), this means the tail portion of the SSE stream — which typically contains the final usage/cost events and fake-200 markers — is buffered but silently dropped when joinChunks() is called. Cost calculations, usage extraction, and error detection will all miss any data that was pushed into the tail buffer.

It looks like tail-mode was intended to be activated by a inTailMode = true assignment when the head buffer fills up, but that assignment was never added. The const change in this PR makes the situation explicit — consider either:

1. Removing inTailMode entirely and relying solely on headBufferedBytes < MAX_STATS_HEAD_BYTES to gate head vs. tail routing, or
2. Restoring let and adding the missing assignment that switches to tail mode once the head is full.

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/app/v1/_lib/proxy/response-handler.ts
Line: 1155

Comment:
`inTailMode` is always `false` — tail data is never joined

The CI lint fixer correctly changed `let` to `const` because this variable is never reassigned. However, this surfaces a pre-existing logic issue: `inTailMode` is permanently `false`, which means `joinChunks()` (line 1164) always takes the early-return branch and returns only `headText` — the tail buffer is populated when the head fills beyond 1 MB, but its contents are **never included** in the assembled output used for stats/cost parsing.

For large streaming responses (> 1 MB), this means the tail portion of the SSE stream — which typically contains the final usage/cost events and fake-200 markers — is buffered but silently dropped when `joinChunks()` is called. Cost calculations, usage extraction, and error detection will all miss any data that was pushed into the tail buffer.

It looks like tail-mode was intended to be activated by a `inTailMode = true` assignment when the head buffer fills up, but that assignment was never added. The `const` change in this PR makes the situation explicit — consider either:
1. Removing `inTailMode` entirely and relying solely on `headBufferedBytes < MAX_STATS_HEAD_BYTES` to gate head vs. tail routing, or
2. Restoring `let` and adding the missing assignment that switches to tail mode once the head is full.

How can I resolve this? If you propose a fix, please make it concise.