fix(vscode): make cache hit rate write-aware

[marius-kilocode]

The VS Code session usage bar currently calculates cache hit rate as cache reads divided by input plus cache reads. Because cache writes are excluded, sessions with significant write volume can display nearly 100% despite paying for millions of written tokens.

This changes the rate to include cache writes in the denominator, matching the write-aware calculation used by the TUI:

cache read / (input + cache read + cache write)

The screenshots below use the exact same persisted session and token counts. The display changes from 100.0% to 49.7%, while In 6, Cache R 19,242, and W 19,442 remain unchanged.

Before

After

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

The extracted cacheRate helper matches the CLI's formatRate in packages/opencode/src/kilocode/plugins/model-usage.ts:65 (read / (input + read + write)), the zero-total guard is preserved, and the test expectation (20 / 35 = 57.1%) is arithmetically correct. Change is confined to Kilo-owned paths and includes a user-facing changeset.

Files Reviewed (4 files)

• .changeset/write-aware-cache-rate.md
• packages/kilo-vscode/tests/unit/model-usage.test.ts
• packages/kilo-vscode/webview-ui/src/components/chat/TaskUsage.tsx
• packages/kilo-vscode/webview-ui/src/context/model-usage.ts

---

_{Reviewed by claude-opus-5 · Input: 20 · Output: 3.4K · Cached: 449.8K}

_{Review guidance: REVIEW.md from base branch main}