3 Commits

Author	SHA1	Message	Date
你的姓名	062978ec97	release: v0.5.8	2026-06-11 02:03:00 +08:00
小橙子	fe4bf78448	fix(provider): auto-detect & fall back from thinking-incompatible providers (#6) ... When a third-party Anthropic-compatible gateway (typical: Bedrock-backed proxies like the user's mimo setup) can't relay Anthropic's `thinking` field, it returns 400s like "该模型不支持 'additionalModelRequestFields' 字段" — the gateway has wrapped the unknown Anthropic param into AWS Bedrock's `additionalModelRequestFields`, and Bedrock rejects on the target model. The user reported a 4.5 model on mimo failing this way while the same model on kiro-account-manager 1.7.5 (which doesn't go through Bedrock) was fine. End-to-end fix: detect the failure pattern, sticky-mark the provider, auto-restart the sidecar with `CLAUDE_CODE_DISABLE_THINKING=1`, and surface a Settings badge so the user knows what happened. Mirrors the desktop providerCompatStore pattern from PR #4 (fake tool_use detection) — two compat dimensions, same re-arm-on-edit semantics, same UX shape. Server side ----------- - `SavedProvider` schema gains `thinkingIncompatible?: boolean` and `thinkingIncompatibleReason?: string` (max 500 chars). Fully back-compat: missing field reads as undefined. - `ProviderService.markThinkingIncompatible(id, reason)` writes the flag, idempotent for identical reasons (no thrash on a burst of the same error). Returns null for unknown ids and openai-official. - `ProviderService.updateProvider` auto-clears the flag on any edit — user changing config = "I'm fixing it" → fresh chance. - `buildProviderManagedEnv` injects `CLAUDE_CODE_DISABLE_THINKING=1` when the flag is true, so the next sidecar launch goes out without thinking entirely. - WS handler runs `detectThinkingIncompatMessage` on every error ServerMessage forwarded to the desktop. On match, fires: 1. `markThinkingIncompatible` (persists) 2. `provider_compat_event` WS message to the desktop 3. `enqueueRuntimeTransition` → `scheduleRestartSessionWithRuntimeConfig` (same graceful path used by set_runtime_config — never tears down a streaming response mid-flight) Process-local dedup so a burst of identical errors doesn't restart the sidecar repeatedly. - Detection regex deliberately narrow: /additionalModelRequestFields/i /\bthinking\b[^.]*\b(not supported|unsupported|invalid| disabled|rejected)\b/i /unknown.{0,40}\bthinking\b/i False positives would permanently disable thinking on the wrong provider until the user edited its config. Desktop side ------------ - New ServerMessage variant `provider_compat_event` plumbed through both server and desktop chat type unions. - `providerCompatStore` gains `thinkingIncompatibleProviderIds: Set<string>`, `recordThinkingIncompatible(id, reason)` action, `hasProviderThinkingIncompatible(id)` helper. Persisted to the same localStorage key as the fake-tool_use counter, alongside it. Dedup: re-firing for an already-flagged provider doesn't re-toast. - `clearProvider(id)` now clears BOTH dimensions, matching the server-side updateProvider behavior. - `chatStore` switch dispatches the new event to the store. - `Settings → Provider` row renders a separate "思考不兼容" badge (psychology_alt icon) next to the existing "工具调用异常" badge. Both badges can show simultaneously when a provider has both problems; both are cleared by the same edit-and-save action. - 5 locales (en/zh/zh-TW/jp/kr) get 3 new keys each: badge label, badge tooltip, and toast message. Tests ----- - `src/server/ws/thinkingIncompat.test.ts` (NEW, 6 cases): regex pinning — Bedrock additionalModelRequestFields rejections / thinking-rejection phrases / unrelated 4xx / passing mentions of thinking / null-empty defensive / case-insensitive. - `src/server/__tests__/providers.test.ts` (5 added): markThinkingIncompatible persists / truncates 500-char reason / idempotent / null-for-unknown-id / null-for-openai-official / updateProvider auto-clears. - `src/server/__tests__/provider-runtime-env.test.ts` (2 added): `CLAUDE_CODE_DISABLE_THINKING=1` injected when flag true / absent when flag false (back-compat with v0.5.7 providers.json). - `desktop/src/stores/providerCompatStore.test.ts` (6 added): records flag + one-time toast / persists set / hydrates from localStorage / clearProvider clears both dimensions / null-empty defensive / hasProviderThinkingIncompatible reflects current state. Folded into the unreleased v0.5.8 release notes — fold both this fix and the PR #5 (handoff fixes) into the same 0.5.8 release. Tested: - bun run lint (desktop) clean - bun test src/server/{ws/thinkingIncompat,__tests__/providers, __tests__/provider-runtime-env}.test.ts → 87 + 6 = 93 pass - bunx vitest run src/stores/providerCompatStore.test.ts → 14 pass - bunx vitest run src/components/chat/AssistantMessage.faketooluse.test.tsx → 6 pass (regression check, no breakage from the new chatStore dispatcher case) Confidence: high Scope-risk: narrow — additive types throughout (back-compat with v0.5.7 providers.json), narrow detection regex, dedup guard prevents restart loops, edit-to-clear gives users an easy escape hatch. Co-authored-by: 你的姓名 <you@example.com>	2026-06-11 01:43:21 +08:00
小橙子	2b6202d381	fix(handoff): reuse empty session + ship raw tail for v0.5.8 (#5) ... Two bugs in the v0.5.7 "Continue from here" hand-off path landed in the wild and the user caught them while exercising 0.5.8 prep: 1. **Stray empty session.** When the user previously opened a "New session in X" tab from the sidebar, then closed the tab to declutter, the freshly-created empty session was already on disk and visible in the sidebar. Coming back to the welcome screen (activeTabId = null → EmptySession route) and clicking "Continue from here" called `createSession` unconditionally, minting yet another empty session next to the now-stale one. The user ended up with `Untitled Session 27 minutes ago` lingering in the sidebar while the hand-off ran in a completely separate fresh session. Extract the picker into `desktop/src/lib/sessionReuse.ts` — `pickReusableEmptySession(sessions, workDir, excludeSessionId?)` filters by exact-workDir match, `messageCount === 0`, excludes the previous session being handed off FROM, and sorts by `modifiedAt` desc. Caller in `EmptySession.onAutoHandoff` runs this before `createSession`; on hit, openTab on the existing sessionId and let ContentRouter switch to ActiveSession naturally; on miss, create fresh as before. 7 unit tests cover the boundaries: workDir mismatch, non-zero messageCount, excludeSessionId honored, null workDir treated as "no candidates" (so we never silently merge home-dir sessions into a project hand-off), empty workDir argument, empty session list, freshness sort order. 2. **Summary too abstract; AI doesn't know specific recent state.** The previous implementation only injected the LLM-summarized `main` + `recent` paragraphs into the next session's system prompt. That summarization tends to wash out exact wording, file paths, error messages, and the user's literal last question — the user reported "AI doesn't know what we just hit a wall on." Add `recentRaw?: string` to `SessionSummary`. New helper `buildRecentRawSlice` keeps the LAST ~12 turns (env-tunable via `CLAUDE_CODE_HANDOFF_RAW_TURNS`, range 0–50, default 12), each truncated to 400 chars preserving the `USER:` / `ASSISTANT:` role prefix, total capped at ~8000 chars (env-tunable via `CLAUDE_CODE_HANDOFF_RAW_CHARS`, range 500–30000, default 8000). Older turns drop first when the total cap is busted by dense recent turns. `formatHandoffSystemPrompt` renders the raw slice inside a fenced code block AFTER the abstracted recent summary, so the next session sees both the digest and the literal text. Section is omitted entirely when `recentRaw` is absent — v0.5.7 caches still load and produce the exact same prompt bytes (back-compat). 8 unit tests cover the formatter's back-compat path, the raw block placement, fence rendering, tail-N selection, all-fits case, per-turn truncation with role prefix preservation, total cap busting (older drops first), and the `RAW_TURNS=0` opt-out. Both fixes fold into the unreleased v0.5.8 — release-notes/v0.5.8.md gets two new fix bullets and the 范围 / 验证 sections call out the new files and 15 added tests. desktop/package.json bump to 0.5.8 stays as-is (preset earlier in the same worktree before this commit). Tested: - bun run lint (desktop) clean - bunx vitest run sessionReuse.test.ts → 7/7 pass - bun test sessionSummaryService.test.ts → 8/8 pass - bunx vitest run EmptySession.test.tsx → 22/22 pass (regression check after the helper extraction — the existing tests don't go through the hand-off branch but exercise the rest of the file) Confidence: high Scope-risk: narrow (no shape changes to any persistence — recentRaw is an additive field, old caches deserialize cleanly because `tryParseSummaryResponse` already only required main+recent) Co-authored-by: 你的姓名 <you@example.com>	2026-06-10 20:46:46 +08:00