Greptile Summary

This PR fixes a context-utilization display bug in WebChat where session.inputTokens (cumulative across all API sub-calls in a run) was used instead of session.totalTokens (the prompt-size snapshot from the final API call). The result was premature 100% context warnings and a blocked chat input with multi-call models like MiniMax-M2.5.

Changes:

• ui/src/ui/views/chat.ts — renderContextNotice now reads session.totalTokens instead of session.inputTokens.
• ui/src/ui/chat/slash-command-executor.ts — /status context % calculation switched from inputTokens to totalTokens, matching the TUI behaviour.
• ui/src/ui/views/chat.browser.test.ts — fixture updated from inputTokens to totalTokens so the existing context-notice test continues to pass.

Issue: slash-command-executor.node.test.ts was not updated. Its fixture already has inputTokens: 1200, totalTokens: 1500, and contextTokens: 4000. The old code yielded 1200 / 4000 = 30% (what the test asserts); the new code yields Math.round(1500 / 4000 * 100) = 38%, so that test will fail.

Confidence Score: 3/5

• The semantic fix is correct, but the paired node test was not updated and will fail CI.
• The logic change across chat.ts and slash-command-executor.ts is correct and well-reasoned. The browser test fixture was updated. However, slash-command-executor.node.test.ts still asserts 30% context while the new code produces 38% — a test that was passing before will now fail in CI, blocking the merge until fixed.
• ui/src/ui/chat/slash-command-executor.node.test.ts — expected context percentage string must be updated from 30% to 38%.

This is a comment left during a code review.
Path: ui/src/ui/chat/slash-command-executor.ts
Line: 310-311

Comment:
**Existing node test will break with new percentage**

`slash-command-executor.node.test.ts` (line ~325) was not updated and now produces the wrong expected value. The fixture already has `inputTokens: 1200`, `totalTokens: 1500`, and `contextTokens: 4000`. The old code computed context % from `inputTokens`: `1200 / 4000 = 30%`, which matches what the test asserts. With this change, `contextUsed = session.totalTokens = 1500`, so the new result is `Math.round(1500 / 4000 * 100) = Math.round(37.5) = 38%` — but the test still expects `Context: **30%** of 4k`.

The fix is to update the expected string in `slash-command-executor.node.test.ts`:

```
"**Session Usage**\nInput: **1.2k** tokens\nOutput: **300** tokens\nTotal: **1.5k** tokens\nContext: **38%** of 4k\nModel: `gpt-4.1-mini`"
```

(or align the fixture's `totalTokens` to a value that yields the same 30% to minimize test delta).

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

_{Last reviewed commit: "fix(webchat): use to..."}

Closing this as implemented after Codex review.

Current main already implements this fix on the affected Control UI surfaces and includes matching tests, so PR #51536 is superseded by code now on main. The later timeline cross-reference to PR #71462 matches the behavior now present in the repo.

What I checked:

• /usage now computes context utilization from the current context snapshot: executeUsage derives contextSnapshotTotal from session.totalTokens, gates on totalTokensFresh, and uses that value for the displayed context percentage instead of cumulative inputTokens. (ui/src/ui/chat/slash-command-executor.ts:385, 42514156e03c)
• Context notice uses totalTokens on main: getContextNoticeViewModel reads session.totalTokens for used-context calculations and suppresses the notice when the snapshot is stale. (ui/src/ui/chat/context-notice.ts:64, 42514156e03c)
• Regression test matches the PR's expected /usage behavior: The node test fixture uses totalTokens: 1500 and expects Context: **38%** of 4k, which is the same correction called out in the PR discussion and bot review. (ui/src/ui/chat/slash-command-executor.node.test.ts:490, 42514156e03c)
• Context-notice test explicitly guards against cumulative inputTokens over-reporting: The current test asserts that a session with inputTokens: 757_300 and totalTokens: 46_000 does not show the warning, and that the rendered notice uses 190k / 200k rather than the cumulative input figure. (ui/src/ui/chat/context-notice.test.ts:32, 42514156e03c)
• GitHub discussion already pointed at the same landing direction: The provided timeline for PR fix(webchat): use totalTokens for context utilization display #51536 includes a cross-reference to closed PR fix(ui): use current context usage in Control UI #71462 (fix(ui): use current context usage in Control UI), which aligns with the behavior now present on main.

So I’m closing this as already implemented rather than keeping a duplicate issue open.

Review notes: reviewed against 42514156e03c; fix evidence: commit 42514156e03c.