fix: correct Anthropic provider inputOther token counting by hobostay · Pull Request #2 · MoonshotAI/kimi-code

hobostay · 2026-05-22T14:49:21Z

Summary

inputOther was set to raw input_tokens which already includes cache tokens, causing double-counting of non-cache input tokens
Subtract cache_read_input_tokens and cache_creation_input_tokens from the total, matching the OpenAI provider's calculation pattern (promptTokens - cached)

Root Cause

In packages/kosong/src/providers/anthropic.ts, _extractUsage assigned usage.input_tokens directly to inputOther. Per the Anthropic API, input_tokens is the total billed input tokens which already includes cache-read and cache-creation tokens. This inflated the reported non-cache input count.

Test plan

Verify existing Anthropic provider tests still pass
Confirm inputOther + inputCacheRead + inputCacheCreation now sums correctly to input_tokens

🤖 Generated with Claude Code

inputOther was set to raw input_tokens which already includes cache tokens, causing double-counting. Subtract cache_read_input_tokens and cache_creation_input_tokens to get the actual non-cache input count, matching the OpenAI provider's calculation pattern. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

liruifengv · 2026-05-23T04:58:55Z

Thank you for your interest in contributing to kimi-code.
However, we currently do not accept bulk AI-generated submissions that have not undergone thorough human review, so we will be closing these PR.

@ifoo

…imentalDecorators (P2.1) Smoke test for Phase 2 @ifoo migration: switch `services.set(IConnectionRegistry, new ConnectionRegistry())` to `services.set(IConnectionRegistry, new SyncDescriptor(ConnectionRegistry))` so the container drives construction through `_createAndCacheServiceInstance` and the @ifoo auto-injection path. ConnectionRegistry has 0 service deps so the auto-inject step is a no-op — this is the minimum viable proof that the new path replaces ad-hoc `new C()` registrations cleanly. Logger stays as an externally-built `PinoLogger` instance (its ctor takes the pino DaemonLogger, not a DI dependency). Also enable `experimentalDecorators: true` in `packages/{daemon,services}/tsconfig.json` — required prerequisite for `@IFoo`-style parameter decorators landing in P2.2 onwards. The root tsconfig stays untouched; only the two packages migrating to @ifoo opt in. esbuild (vitest transpiler) and tsdown (rolldown bundler) both honor this flag for legacy parameter-decorator emit. Per phase-1 reviewer handoff #1: this is the smoke test confirming a single descriptor-based registration migrates cleanly before fan-out. agent-core/src/di/ untouched (sealed for Phase 2). Daemon test green (241 pass after one flake-rerun on fs-watch.e2e.test.ts AC #2 burst window — known flake per phase-1 reviewer handoff #5). VERDICT: PASS — start.ts:296 descriptor swap + tsconfig experimentalDecorators=true; 241/241 daemon tests green.

…er-domain move Cleanup tail of the per-domain restructure. Step B.1's atomic move deleted 8 positive-service interfaces and impls but left the 3 broker interfaces (approval/question/event) and 2 broker adapters in the old layout while their new domain copies (approval/, question/, event/) were already wired into the barrel — making the old files dead orphans. Drop them and let the empty parent directories vanish. The barrel src/index.ts already points exclusively at the new per-domain folders; nothing in packages/services or packages/daemon imports the old paths. Quality gates match the Phase A baseline exactly: services tsc 13 pre-existing errors, daemon tsc 30, services tests 140/140, daemon tests 261/262 (the known fs-watch.e2e.test.ts AC MoonshotAI#2 flake).

…er-domain move Cleanup tail of the per-domain restructure. Step B.1's atomic move deleted 8 positive-service interfaces and impls but left the 3 broker interfaces (approval/question/event) and 2 broker adapters in the old layout while their new domain copies (approval/, question/, event/) were already wired into the barrel — making the old files dead orphans. Drop them and let the empty parent directories vanish. The barrel src/index.ts already points exclusively at the new per-domain folders; nothing in packages/services or packages/daemon imports the old paths. Quality gates match the Phase A baseline exactly: services tsc 13 pre-existing errors, daemon tsc 30, services tests 140/140, daemon tests 261/262 (the known fs-watch.e2e.test.ts AC #2 flake).

- kimy wrapper: move hash writes after smoke test (MoonshotAI#1) - kimy wrapper: add public/ to web hash inputs (MoonshotAI#3) - kimy wrapper: widen vis hash to include config files (MoonshotAI#4) - kimy wrapper: move lockfile from /tmp to ~/.kimy/bin (MoonshotAI#5) - kimy wrapper: use explicit package list for native hash (MoonshotAI#7) - 01-bundle.mjs: skip vis-asset build when already done (MoonshotAI#2) - justfile: sync deploy with new wrapper, add deploy-full (MoonshotAI#9,MoonshotAI#10,MoonshotAI#11) - flake.nix: add unpin guidance to nixpkgs comment (MoonshotAI#12)

…watch AC #2 - minidb: fuzz-model 10 seeds x 500 steps -> 6 x 250; compaction-race heavy-writes 20k/5k -> 10k/2k (still spans five writeSnapshot yields and a >64KiB WAL tail); crash-recovery runs 8/4 -> 5/3 (cost is subprocess spawn, not rounds); compaction concurrent 10k/2k -> 5k/1k - replace real-second waits with injected short timeouts or fake timers: hooks runner timeout 1s -> 0.05s (both packages), lifecycle-hooks printWaitCeilingS 1s -> 0.05s, task-tools block-timeout driven by fake timers, mcp connection-manager slow-startup child exits on stdin end, kap-server 429 test injects a token-only auth service (12x bcrypt ~250ms was the real cost, not the rate limiter) - fs-watch AC #2: the burst test required chokidar to deliver >500 events inside one 200ms window, which trickles under CI load. Add an fsWatcherOptions seam to ServerStartOptions (default behavior unchanged) and run the test with a 500ms window / 100-event threshold, which still proves truncation but tolerates 10x slower delivery

#2: Batch appendLoopEvent in executeStepTools — collect all tool.call and tool.result events during for-await, then dispatch once at end. Eliminates N synchronous context array copies per step. #3: Wire tryNativeReadBatch into native-tools.ts — existing Rust nativeBatchRead now callable from the tool layer. Callers fall back to sequential nativeRead when native module unavailable. #1: Async-ify native_edit with spawn_blocking — no longer blocks event loop during parallel tool execution.

…, diff drawer, more rounded corners Five user-reported issues fixed in this batch: MoonshotAI#3 Font-size adjustment now actually works Root cause: App.svelte's :global(body) used the `font:` shorthand with a hardcoded 13px, which won over the global.css `body { font-size: var(--ui-font-size) }` rule (same specificity, later in cascade). The shorthand resets font-size and the slider had no effect. Fix: replace shorthand with individual properties (font-family/size/line-height), keep font-size: var(--ui-font-size). Verified via headless Chromium: default 14px, slider to 20px → 20px, back to 14px → 14px. MoonshotAI#8 Plugin enable/disable ACL error Root cause: PluginPanel called Command.sidecar('kimi', ['plugin', 'disable', id]) which needs shell:allow-execute permission that isn't (and shouldn't be) granted in tauri.conf.json. Fix: added Rust `toggle_plugin(plugin_id, enabled)` command that directly edits ~/.kimi-code/plugins/installed.json (atomic tmp+rename). PluginPanel now invoke()s the new command instead of spawning a sidecar. No ACL needed. MoonshotAI#11 Softer UI — bumped radius tokens --r-sm 6→8, --r-md 8→10, --r-lg 10→14, --r-xl 12→18. Legacy scale bumped in parallel (--radius-xs 6→8, etc.). All components using these tokens get the bump automatically. MoonshotAI#2 Sidebar workspace/session hierarchy Workspace headers are now visually bigger (30px tall, 12.5px bold, with a top border + subtle bg gradient) and the session rows are indented (margin-left: 20px, smaller 11.5px font-weight 400). Workspace count badge becomes a pill instead of plain text. The hierarchy reads cleanly now. MoonshotAI#1 Working-directory diff → second-level right drawer New DiffDrawer.svelte component (540px, slides in from right with spring animation). Clicking a file in the '工作区改动' section of RightPanel opens the drawer; inline expansion is gone. Commit- history diffs stay inline (only the working-directory section was changed per UX request). Esc / ✕ / mask click closes. MoonshotAI#9 Composer compact moved into the main control row '压缩' is now a pill button next to mode/model/thinking instead of a separate FAB below. Per UX request: 'mode / model / thinking / compact all in one row'. Verified: svelte-check 0 errors, cargo check 0 warnings.

liruifengv closed this May 23, 2026

zbl1998-sdjn mentioned this pull request May 29, 2026

Test suite is not portable on Windows (path separators, path.resolve, platform-conditional snapshots, shell) #225

Open

tt-a1i mentioned this pull request Jun 20, 2026

🤖 Code Audit: 10 potential issue(s) found #837

Open

This was referenced Jun 20, 2026

test(server): replace flaky fs-watch e2e burst test with FsWatcherService unit test #943

Open

test(server): replace flaky fs-watch burst e2e with FsWatcherService unit test #944

Closed

kermanx mentioned this pull request Jul 13, 2026

test: cut slow suite runtimes and isolate experimental flag from host env #1595

Merged

5 tasks

ping1999 mentioned this pull request Jul 20, 2026

[Bug] 0.27.0 web server process repeatedly dies with SIGSEGV in V8 GC marking — 6 crashes in ~26 h #1943

Closed

seasonmsg mentioned this pull request Jul 23, 2026

Interactive sessions ignore [[hooks]] and [[permission.rules]] from config.toml; headless (-p) sessions enforce them (0.28.1, Windows) #2070

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: correct Anthropic provider inputOther token counting - #2

fix: correct Anthropic provider inputOther token counting#2
hobostay wants to merge 1 commit into
MoonshotAI:mainfrom
hobostay:fix/anthropic-token-counting

hobostay commented May 22, 2026

Uh oh!

liruifengv commented May 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

hobostay commented May 22, 2026

Summary

Root Cause

Test plan

Uh oh!

liruifengv commented May 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants