@kodax-ai/kodax 0.7.39 → 0.7.40

package/CHANGELOG.md

@@ -4,6 +4,49 @@ All notable changes to this project will be documented in this file.
 
 > Full history for versions prior to v0.7.0: [CHANGELOG_ARCHIVE.md](docs/CHANGELOG_ARCHIVE.md)
 
+## [0.7.40] - 2026-05-13
+
+### Theme
+
+**Envelope Spillover + Vision Bridge** — Two parallel-developed features close gaps in the child-agent communication path (FEATURE_121) and the REPL input path (FEATURE_134). FEATURE_121 routes child task summaries through the existing `tool-result-policy.ts` spillover system (50KB per-banner + 200KB envelope aggregate cap, mirror of claudecode's `MAX_TOOL_RESULTS_PER_MESSAGE_CHARS`), removing two prior hard truncations (`orchestration.ts:1033` 1600-char slice + `dispatch-child-tasks.ts:256` 200-char slice) that were silently losing 95%+ of child output bytes. A follow-up LLM blob summarizer fallback handles the residual `spillFailed + content > 100KB` data-loss edge that the main slice's inline fallback would otherwise resolve by blowing the Worker context window. FEATURE_134 adds 5 paste sources (bracketed paste / `@<path>` file refs / macOS Cmd+V auto-link / Windows Alt+V explicit / macOS-Linux Ctrl+V backup) on top of the existing `KodaXImageBlock` AI-layer vision serialization, completing the round-trip from screenshot in clipboard to multimodal `user` message at any of the 12 KodaX providers. A late P0 regression in REPL transcript rendering (transcript items invisible during agent execution under AMA mode on Windows ConPTY) was diagnosed as `useDeferredValue` starvation in Ink under Node.js (no DOM idle-scheduling bridge), fixed by removing the deferral indirection — the 200-item UUID-anchored cap from FEATURE_060 Tier 2 retains the perf protection that was the actual fix for SSH-resume O(N) blow-up.
+
+### Added
+
+- **FEATURE_121 — Envelope Spillover Gap-Fix + LLM Blob Summarizer Fallback**. Two-commit landing: main slice (`0a0f844e`) + follow-up LLM blob summarizer (`ba0c82f9` + review fixes `05259ab2`). Removes the `orchestration.ts:1033` 1600-char `truncateText` + `dispatch-child-tasks.ts:256` 200-char `slice` two-layer hard truncation that silently dropped 95%+ of child task output (a 25KB audit report reached the Worker as ~50 tokens). Routes every `<task-completed>` banner through `applyToolResultGuardrail('child_task_summary', ...)` — 50KB head + spill-to-file under `getAgentConfigPath('tool-results')/<id>.txt`, banner now carries a preview + spill path that the Worker reads via standard Read tool. `composeIdleYieldUserMessage` in `@kodax/agent/orchestration/idle-yield.ts` gains a 200_000-chars aggregate envelope cap (mirror of claudecode's `MAX_TOOL_RESULTS_PER_MESSAGE_CHARS=200_000`) — when N banners individually fit but together exceed envelope budget, the enforcer calls `applyToolResultGuardrail(..., { forceSpill: true })` to reclaim space. Capability sections include a `LARGE CHILD OUTPUT (FEATURE_121 v0.7.40)` block teaching the Worker the spillover-path Read pattern. 4-alias × 3-case × 5-runs Layer 2 eval clears PARTIAL SHIP (3/4 aliases ≥80% on each case; mmx/m27 weak floor 60-80% documented in test guide). **Follow-up — LLM Blob Summarizer (last-resort fallback)**: when `persistToolOutput` fails (ENOSPC / EACCES / EROFS / SELinux denial) AND raw content > `LARGE_CONTENT_THRESHOLD_BYTES` (100KB), `dispatch-child-tasks` now calls `ctx.summarizeBlob(content, {maxChars: 8000})` to compress to a ~2-8KB lossy summary preserving file paths / line numbers / error codes / identifiers / findings verbatim, banner-wrapped with `[SPILL FAILED — original ${size} compressed via LLM summarizer; raw content unavailable. Worker: treat this summary as LOSSY...]`. The summarizer callback is injected into `KodaXToolExecutionContext` via lazy-once memoization in `runner-driven.ts` bound to the Worker's own provider/model — layer-independent (`@kodax/agent` stays unaware of LLM client). If the summarizer itself fails (provider error / abort), falls back to inline full content with an emergency banner `[SPILL FAILED AND LLM SUMMARIZER FAILED — original ${size} inlined as last-resort emergency dump...]` so the Worker never silently receives oversized opaque content. Honors the FEATURE_121 contract: **silent data loss is the worst outcome; over-budget but observable is acceptable**. 4-alias × 2-case × 3-runs Layer 2 eval clears SHIP gate (4/4 aliases × 100% retention on audit_report + grep_findings cases). Test guide: `docs/test-guides/FEATURE_121_v0.7.40_TEST_GUIDE.md`. Design doc: `docs/features/v0.7.40.md#feature_121-envelope-spillover-gap-fix--child-task-summary-接入-tool-result-policy`.
+- **FEATURE_159 — MessageQueue as Single Source of Truth + Idle-Yield Mode-Split**. Main commit `948b8879` (Phase 3 mode-split synthetic + unified queued-followup predicate) + design `daa8e846` + follow-up `9d4c6ae4` (queue filter scope + verdict.summary echo) + test-isolation `29369a2a` (MessageQueue test isolation + compaction flake fix). User-reported during v0.7.40 RC: after Worker dispatched 3 child tasks and the main agent was still tool-calling, user typed a follow-up "你是派出了子Agent再做嘛？" — status bar showed `Queue 1` / `Queued follow-ups: 1`, but the prompt looked **swallowed** when the agent finished its investigation (not folded into the answer). Forensic showed two stacked failures: (1) `waitForWakeEvent` consumed messages via `MessageQueue.dequeue()` with no reverse notification to REPL, so the React `state.pendingInputs` retained stale entries and the `Queue N` indicator never cleared; (2) drained prompts were marked `_synthetic: true` and hidden from transcript — the user prompt appeared dropped even though it had landed in the model context. Fix flips the substrate: `MessageQueue` becomes the **single source of truth** with typed event subscribe + frozen snapshot read + `mode`/`id`/`predicate` filtering APIs; REPL reverses its sync direction (queue→React mirror via `subscribe`); `composeIdleYieldUserMessage` branches on `msg.mode` — real user prompts emit as **non-synthetic** user-bubble messages (visible in transcript), task-notifications stay synthetic (silent background framing). Net result: claudecode-parity messageQueueManager semantics + SDK-grade observable substrate, without importing claudecode's `commandLifecycle` module, `recordQueueOperation` file sink, or 3-tier priority (kept KodaX's binary user/background). Net −70 LoC, zero new modules. Test guide: `docs/test-guides/FEATURE_159_v0.7.40_TEST_GUIDE.md`. Design doc: `docs/features/v0.7.40.md#feature_159-messagequeue-as-single-source-of-truth--idle-yield-mode-split`.
+- **FEATURE_134 — Image / Screenshot Paste Input (REPL Vision Bridge)**. Commit `2e9674bb`. Adds the REPL paste entry point on top of KodaX's existing `KodaXImageBlock` AI-layer vision serialization (already implemented in `packages/ai/src/providers/anthropic.ts:770` + `openai.ts:904` + `image-serialization.ts`). 5 paste sources per claudecode `usePasteHandler.ts` parity: (1) bracketed paste base infra (DEC 2004 `ESC[200~/201~` wrapping — note FEATURE_134's redundant `enableBracketedPasteMode()` pre-render write was removed in `ca009b3a` since `KeypressContext.tsx:134` already owns DEC 2004 lifecycle via Ink's managed stdout); (2) file path paste — `extractImagePaths` splits on `/` or `[A-Za-z]:\\` + extension; (3) macOS Cmd+V auto-link via `osascript NSPasteboard` read; (4) Windows Alt+V explicit keybind (Ctrl+V is system-paste-reserved on Windows, same as claudecode); (5) macOS/Linux Ctrl+V backup via `wl-paste` / `xclip` / `osascript`. Building blocks in `packages/repl/src/paste/`: `bracketed-paste-mode.ts` (DEC 2004 lifecycle), `image-normalize.ts` (jimp decode + clamp 2000px + PNG→JPEG quality ladder 80/60/40 to fit 3.75MB), `clipboard-image.ts` (cross-platform reader, never throws), `persist-image.ts` (writes to `$TMPDIR/kodax-paste/` + returns path-based `KodaXImageBlock`), `paste-handler.ts` (5-source orchestrator). Library choice: jimp (~10MB pure JS) over sharp (~30MB native binary) — jimp install never fails on cross-platform CI; claudecode picked sharp for broader use cases but KodaX only paste-uses, so jimp is the simpler fit. Integration: paste event → `@<resolved-path>` text-token translation via existing `common/input-artifacts.ts:preparePromptInputArtifacts` pipeline which already converts `@<path>` refs to `KodaXInputArtifact[]` on submit — no changes needed downstream. Test guide: `docs/test-guides/FEATURE_134_v0.7.40_TEST_GUIDE.md`. Design doc: `docs/features/v0.7.40.md#feature_134-image--screenshot-paste-input--repl-vision-bridge`.
+
+### Fixed
+
+- **REPL transcript rendering starvation — `useDeferredValue` removed from `displayHistory` chain**. Commit `ca009b3a`. User-reported P0 (AMA mode, Windows ConPTY): pressing Enter on a query showed nothing in the transcript for the entire agent run — header banner + TodoListSurface + spinner + status bar rendered normally, but assistant thinking blocks / tool calls / tool results / assistant text remained invisible until task completion forced a re-render. Root cause: `useDeferredValue(displayHistory)` (introduced in v0.7.30 FEATURE_060 Tier 2 as a polish on top of the 200-item cap that was the real perf fix) marks transcript rebuild as low-priority React work. Under React DOM, this work flushes during browser idle time via the scheduler's idle-callback bridge. Ink uses react-reconciler without that bridge — under Node.js, high-priority work (spinner ticks @30fps, streaming setState bursts, tool-state updates) perpetually pre-empts the deferred work, so the low-priority track never flushes until a "big" state change like `setIsLoading(false)` forces a sync re-render. v0.7.40 surfaced the latent starvation: FEATURE_121's envelope spillover replaced `slice(0, 200)` child task truncation with up-to-50KB head + spill content, raising per-item `buildPromptSurfaceItems` cost from ~0.1ms (200 chars) to ~10-50ms (50KB) — single-item cost crossed the React scheduler's low-priority deferral threshold, starving the deferred update perpetually under high-frequency setState bursts. Fix: replace `useDeferredValue(displayHistory)` with direct passthrough. The 200-item UUID-anchored cap (the real perf protection from FEATURE_060 Tier 2) and the transcript-mode 30-message cap remain — together with React's built-in `useMemo` memoization (which prevents spinner ticks from triggering `buildPromptSurfaceItems` when history is unchanged), per-render cost stays bounded at O(min(N, 200)). Trade-off: long-session `kodax -c` resume on Windows-SSH may add ~10-40ms one-time first-paint cost — well below human perception threshold (~100ms). Length-thresholded fallback pattern documented inline (`displayHistory.length > 100 ? lazyDeferred : displayHistory`) for future repro.
+- **FEATURE_121 v0.7.40 follow-up build break + memoize + emergency banner**. Commit `05259ab2`. Fixes three review-uncovered issues from `ba0c82f9`: (1) **CRITICAL build break** — `blob-summarizer.ts` imported `KodaXProvider` (not exported from `@kodax-ai/llm`); replaced with `KodaXBaseProvider` everywhere. `tsc --noEmit` missed this; `tsc -b` with declaration emission caught it during `npm run build:packages`. (2) **HIGH memoize** — `runner-driven.ts` `summarizeBlob` was rebuilding the factory closure (including `resolveProvider`) on every call; changed to lazy-once memoize (cached on first invocation, reused for the rest of the Worker run). (3) **MEDIUM emergency banner** — when LLM summarizer ITSELF fails AND we fall back to inline 100KB+ content, the Worker received the raw blob with no banner; now prepends `[SPILL FAILED AND LLM SUMMARIZER FAILED — original ${size} inlined as last-resort emergency dump]` so the Worker sees a clear signal to expect possible downstream truncation and re-run upstream with narrower scope.
+- **`fix(build,v0.7.39): make 'npm run build' produce shippable dist/`** (commit `b77fa0a3`). Footgun caught during clean-room publish dry-run: `npm run build` previously ran `npm run build:packages && tsc` — the trailing bare `tsc` read root `tsconfig.json` (`outDir: ./dist`, `declaration: true`) and overwrote the esbuild bundle with unbundled `tsc` output containing bare `import '@kodax-ai/coding'` specifiers, which would have shipped a broken tarball — consumers running `npm install @kodax-ai/kodax` would hit `ERR_MODULE_NOT_FOUND`. Fix: `"build": "npm run build:packages && npm run build:bundle && tsc --emitDeclarationOnly"`. The `--emitDeclarationOnly` flag is the critical guard — tsc skips `.js` entirely and only writes `.d.ts` on top of the esbuild bundle. Now any caller (developer, CI, `release.mjs`) can run `npm run build` at any time and get a complete, shippable dist/. Side benefit: SDK consumers get real TypeScript types for all 5 subpath entries (was untyped before).
+- **`fix(release,v0.7.40): bake KODAX_VERSION into bundle via esbuild --define`** (commit `b70048b7`). The previous v0.7.39 release bundled with `process.env.KODAX_VERSION` left unresolved, so the runtime version was `undefined` in the published `dist/kodax_cli.js` — visible to users as `KodaX undefined` in the banner. Fix: `scripts/build-bundle.mjs` reads root `package.json` at build time and injects via esbuild `--define`, so the bundled CLI carries the correct version literal. `release.mjs` flow unchanged: bumps `package.json` first, then `npm run build` (which reads the bumped version).
+- **FEATURE_134 follow-up: Alt+V duplicate-image-file accumulation in temp dir**. User-reported regression during v0.7.40 RC validation: pressing Alt+V on the same screenshot created many identical-content files with different UUID names under `$TMPDIR/kodax-paste/`. Two root causes: (1) `prompt-input-controller.ts:triggerExplicitImagePaste` had no single-flight guard — OS-level key autorepeat (Windows ConPTY in particular) fires multiple Alt+V keypresses on a brief hold, each spawning a concurrent `readClipboardImage` + `persistImageAsBlock` pair; (2) `persist-image.ts` used `randomUUID()` filenames so even identical buffer content wrote N distinct files. Fix in [packages/repl/src/paste/persist-image.ts](packages/repl/src/paste/persist-image.ts): filename derived from `sha256(buffer).slice(0,16)` — identical content reuses one path, `writeFile` is idempotent on rewrite. Fix in [packages/repl/src/ui/utils/prompt-input-controller.ts](packages/repl/src/ui/utils/prompt-input-controller.ts): `explicitImagePasteInflightRef` boolean ref guard drops re-entrant invocations until the in-flight clipboard read settles. New tests: `persist-image.test.ts` adds "reuses the same path for identical content" + "produces distinct paths for different content"; `prompt-input-controller.test.ts` adds "Alt+V autorepeat fires the clipboard read only once (single-flight guard)". Together the two fixes cap temp-dir growth per unique screenshot at one file per session.
+- **FEATURE_134 follow-up: Gemini-CLI vision via `@<path>` token injection** (commit `71d45783`). The ACP base class (`packages/ai/src/providers/acp-base.ts`) was silently dropping image blocks at the prompt flatten step (`.filter(b => b.type === 'text')`) before forwarding to the CLI bridge. New extension point `serializeImageBlockToPromptToken(block)` on `KodaXAcpProvider` defaults to `null` (preserves silent-drop for Codex-CLI which has no `codex exec --json` image surface), and `KodaXGeminiCliProvider` overrides to return `@<absolutePath>` — Gemini CLI 2.x's file-include syntax inlines any readable file (including images) into the model context. The CLI-bridge capability profile gets a new sibling `IMAGE_INPUT_CLI_BRIDGE_PROVIDER_CAPABILITY_PROFILE` so the policy gate sees the right metadata. 2 new tests in `cli-bridge-providers.test.ts` pin the exact wire format for both Gemini (`@<path>`) and Codex (null).
+- **FEATURE_134 follow-up: paste tmp dir age-based GC at REPL bootstrap** (commit `0eb6cbb4`). Per-session content-hash dedup (commit `12589a46`) prevented Alt+V autorepeat from stacking duplicates within one session, but cross-session accumulation was still unbounded — files only cleared on OS tmpdir reboot. New `prunePasteTmpDir(now?)` exported from `packages/repl/src/paste/persist-image.ts` deletes `paste-*` files older than `PASTE_TMP_TTL_MS` (24h). Non-paste files in the same dir survive (e.g. user-dropped `notes.txt`); per-file errors swallowed so concurrent KodaX instance races don't break REPL startup. Wired in `InkREPL.tsx` bootstrap as a fire-and-forget dynamic import — never blocks first paint. Active-session paste files (always within TTL) are preserved across overnight idle. 4 new tests cover dir-not-exist / age-based deletion / non-paste preservation / TTL boundary.
+- **FEATURE_134 follow-up: custom provider vision opt-in documented** (commit `db36dc69`). README.md + README_CN.md gain a complete `customProviders` example showing the `capabilityProfile.multimodalSupport: 'image-input'` nested shape. Built-in 12 providers ship with the flag already enabled; only custom providers needed opt-in instructions because the field was implicitly available via `KodaXCustomProviderConfig.capabilityProfile` but undocumented.
+- **AMA in-turn compaction parity gap — three-phase lifecycle restored (microcompact + snapshot-aware trigger + graceful fallback)**. User-reported during v0.7.40 RC validation: status bar climbed from 124k → 138k → 150k across three AMA rounds with zhipu-coding/glm-5.1, no compaction / microcompaction / graceful prune ever firing despite crossing the 60% × 200k = 120k threshold. Latent since FEATURE_114 v0.7.36 (4→2 role consolidation) but masked by FEATURE_076's over-aggressive round-exit reshape collapsing context to ~1k after every round; once the v0.7.40 reshape fix above let context grow naturally, the gap surfaced. Three structural deltas vs SA path's `runCompactionLifecycle` ([compaction-orchestration.ts:358](packages/coding/src/agent-runtime/middleware/compaction-orchestration.ts#L358)): (1) **trigger metric mismatch** — the AMA `compactionHook` ([_internal/managed-task/compaction.ts](packages/coding/src/task-engine/_internal/managed-task/compaction.ts)) compared `estimateTokens(transcript)` to the threshold, but transcript-only estimate excludes the system prompt + tools schema (Worker role prompt + AGENTS.md + REPO INTELLIGENCE TOOLS teaching + repo-intel capsule + 12 tool definitions ≈ 20–35k tokens after FEATURE_114 worker consolidation + FEATURE_161 prompt growth). 200k-window 60% trigger needs API total > 120k, but the hook saw ~95–115k transcript and never fired — the status bar's 138k was the real total (system + tools + transcript) sourced from `streamResult.usage.totalTokens`. (2) **no microcompact phase** — SA path runs `microcompact(messages, DEFAULT_MICROCOMPACTION_CONFIG)` every turn at zero LLM cost ([run-substrate.ts:603](packages/coding/src/agent-runtime/run-substrate.ts#L603)) to prune old tool_results / image blocks past `maxAge=20` turns; AMA hook bypassed it entirely. (3) **no graceful degradation fallback** — SA path's third phase `applyGracefulDegradationGate` ([compaction-orchestration.ts:250](packages/coding/src/agent-runtime/middleware/compaction-orchestration.ts#L250)) deterministically prunes tool_results when LLM compact threw / returned `compacted: false` / left context still above `triggerTokens × pruningGapRatio`; AMA hook bailed silently on LLM failure, letting context grow unbounded. Fix in [packages/coding/src/task-engine/_internal/managed-task/compaction.ts](packages/coding/src/task-engine/_internal/managed-task/compaction.ts): hook now mirrors SA path's three-phase lifecycle — (Phase 1) `microcompact` every call (free, prunes old tool blocks); (Phase 2) `intelligentCompact` LLM summary with snapshot-aware trigger check `resolveContextTokenCount(transcript, snapshot)` instead of raw `estimateTokens` — `snapshot.currentTokens` carries the LAST LLM call's API-reported `usage.totalTokens` (system + tools + transcript) so the threshold check uses the same metric the status bar displays; the snapshot is refreshed by the LLM adapter ([runner-driven.ts](packages/coding/src/task-engine/runner-driven.ts) `buildRunnerLlmAdapter`) after every stream completion via the new `contextTokenSnapshotRef` shared between adapter (writer) and hook (reader); (Phase 3) `gracefulCompactDegradation` deterministic prune fallback when LLM compact failed / partial / circuit-breaker-tripped — the gate's "still over" check also uses snapshot-aware accounting for symmetry with Phase 2. Snapshot rebases to `createEstimatedContextTokenSnapshot(compacted)` after any compaction (LLM or graceful) so subsequent delta corrections start from the compacted baseline rather than the stale pre-compaction API total. New test file [packages/coding/src/task-engine/_internal/managed-task/compaction.test.ts](packages/coding/src/task-engine/_internal/managed-task/compaction.test.ts) pins all three phases with 9 tests covering: microcompact-only return path (below trigger); microcompact identity (no diff returns undefined); snapshot-aware trigger fires when transcript-only estimate is sub-threshold but API-total crosses; no-snapshot cold-start falls back to transcript estimate (unchanged baseline); graceful fallback on LLM throw / `compacted: false` / partial success above gap-ratio; snapshot rebase after successful LLM compaction; circuit-breaker semantics (3 strikes skip LLM but graceful still fires). Side fix in [packages/coding/vitest.config.ts](packages/coding/vitest.config.ts) + [packages/repl/vitest.config.ts](packages/repl/vitest.config.ts): `@kodax-ai/llm` alias updated from stale `packages/ai/` to `packages/llm/` (rename leftover from v0.7.40 directory rename above). **Behaviour delta vs pre-fix**: AMA mode compaction now triggers at the same effective API-total threshold as SA mode; graceful fallback prevents zhipu/kimi/mimo provider-side LLM summary failures from leading to monotonic context growth; microcompact's per-turn pruning recovers ~5–15% of context per long-running Worker turn even without crossing the LLM trigger threshold. Full coding suite green: **2446 tests pass** (239 test files, 23 todo; +9 new compaction hook tests).
+- **FEATURE_076 follow-up: round-boundary reshape now preserves tool_use / tool_result chains across rounds (cache + re-read regression fix)**. User-reported during v0.7.40 release validation: status bar dropped from `121k/200k` to `1.1k/200k` after an AMA round, and follow-up rounds were re-reading files the prior worker had already read. Root cause: FEATURE_076 (v0.7.25)'s `reshapeToUserConversation` replaced `result.messages` wholesale with a synthetic `[user, assistant]` dialog at every round exit, discarding the entire `tool_use` / `tool_result` chain. The replacement was the correct fix for v0.7.25's actual problem (cross-round role-prompt pollution — Evaluator role prompts leaking into the next round's worker context as user messages) but over-corrected by stripping structurally useful content, with two concrete production costs: (1) **cross-round file re-reads** — round 2 worker had no visible `tool_result` for files round 1 already read, so common follow-ups like "now modify that file you reviewed" forced re-reads; (2) **provider prompt-cache miss on the dialog prefix** — round 1's first LLM call prefix was `[system, user, assistant(tool_use), user(tool_result), …]`, round 2's is `[system, user, assistant(final), user_2, …]`; the prefixes diverge immediately after the first user message, so the dialog portion gets zero cache reuse across rounds. Fix in [packages/coding/src/task-engine/_internal/round-boundary.ts](packages/coding/src/task-engine/_internal/round-boundary.ts): new `preserveTranscriptForRoundExit` helper runs a 4-step pipeline — (Step 1) strip the leading stale role-prompt system message (Runner.run leaves the last-active agent's role prompt at `transcript[0]`; round 2's entry agent injects its own at position 0, so keeping the previous one would create two conflicting `system` instructions back-to-back); CompactionSummary system messages are preserved via the now-exported `COMPACTION_SUMMARY_PREFIX` discriminator from `@kodax-ai/session-lineage`. (Step 2) Apply `normalizeLoadedSessionMessages` to strip V1-legacy role-prompt-wrapped trailing `{user, assistant}` pairs (`"You are the Evaluator role..."` phrased as a user message); no-op for V2 AMA where role prompts are system-message-shaped. (Step 3) Ensure the round's user prompt is observable — V2 sessions retain it via `runnerInput`, V1 paths may have lost it when normalisation stripped the wrapper. (Step 4) Ensure the transcript ends with a plain-text assistant carrying the sanitised final answer; **replaces** (not appends) when the last message is an assistant with array content (typically `emit_verdict` / `emit_handoff` tool_use blocks — KodaX protocol machinery whose user-facing payload is captured in `result.lastText`). The replace avoids two consecutive `role: 'assistant'` messages which Anthropic's API rejects on the next request. Net behaviour: round 2 worker sees what round 1 read/edited, prompt-cache prefix stays continuous, status bar reflects the actual context size (e.g. ~50k instead of 1.1k after a heavy worker round) — a side-effect users will perceive as the bar "no longer collapsing after each turn". Updated [packages/coding/src/task-engine/_internal/round-boundary.test.ts](packages/coding/src/task-engine/_internal/round-boundary.test.ts) with 7 new cases covering: V2 worker-shape preservation, terminal `emit_verdict` tool_use replacement (no consecutive-assistant violation), `CompactionSummary` system message preservation, dedup of already-correct trailing assistant, empty `result.messages`, compaction-system-only transcript, terminal `thinking`-only assistant. `COMPACTION_SUMMARY_PREFIX` exported from [packages/session-lineage/src/compaction/compaction.ts](packages/session-lineage/src/compaction/compaction.ts) so the producer-consumer pair no longer maintains the literal in two places. Test guide: regression case lives in [docs/test-guides/FEATURE_134_v0.7.40_TEST_GUIDE.md](docs/test-guides/FEATURE_134_v0.7.40_TEST_GUIDE.md) once that doc's "round-boundary cache + re-read" section is appended. **Behaviour delta vs FEATURE_076 baseline**: the v0.7.25 cross-round-coherence guarantee is preserved (stale role prompts still get stripped); only the over-corrective full-transcript replacement is reverted. Full test suite green at HEAD: **5745 tests pass** (was 5742; +3 new edge-case tests).
+- **FEATURE_168 — AMA agent tool wiring (exclude-based, registry as source of truth)**. Commit `56330d1c` + doc `e902b194`. Root cause investigation surfaced from production trace 2026-05-15 (zhipu/glm51 Worker blocked at `emit_handoff` pending-children gate, told by the gate's own error message to call `task_stop`, model honestly responded that `task_stop` "is not registered as a callable tool" — and the model was correct). `runner-driven.ts::buildRunnerAgentChain` had used **include-mode hand-written `agent.tools` arrays** for all 5 AMA roles (Scout / Planner / Generator / Evaluator / Worker) since v0.7.26, while the SA path defaults to `listToolDefinitions()` minus `excludeTools`. The two paths drifted across three features: **FEATURE_120 v0.7.39** registered `send_message` + `task_stop`, taught them in the Worker prompt, gated `emit_handoff` with a "call `task_stop` first" error message — but never wired the `RunnableTool` instances into any AMA agent's tools array. **FEATURE_161 v0.7.40 prompt teaching** taught the Worker 8 repo-intel pull tools (`module_context` / `symbol_context` / `process_context` / `impact_estimate` + 4 shallow ones), `worker-role-prompt.ts:127` comment claimed "the 8 pull tools get stripped from the LLM-visible tool list (see agent-runtime/tool-resolution.ts)", but AMA path doesn't go through `tool-resolution.ts` — it reads `agent.tools` directly, so only 4 of the 8 ever landed in the schema. **Web tools / ask_user_question / worktree_create / worktree_remove / insert_after_anchor / undo**: registered + `permission` system gave them entries, but no AMA agent could see them. Total **17 registered tools silently dropped** from production AMA; no test layer caught it because no test asserted "agent.tools actually contains a schema entry with this name". Fix: switch AMA path to exclude-based wiring matching SA semantics. New `buildAgentToolsFromRegistry(role, ctx, budget, events, overrides)` helper enumerates `listToolDefinitions()` and applies `AMA_BASELINE_EXCLUDE ∪ <ROLE>_EXTRA_EXCLUDE`. Role-specific wraps (mutation-guarded bash/write/edit/multi_edit for Generator/Worker, read-only bash for Evaluator, `dispatch_child_task` per-role drain wrappers, FEATURE_097 throttle-aware `todo_update`) flow through the `overrides` map. Evaluator security boundary made **architectural, not prompt-dependent**: write/edit/multi_edit/insert_after_anchor/undo/dispatch_child_task/send_message/task_stop/worktree_create/worktree_remove/exit_plan_mode/todo_update/ask_user_question all hard-excluded from `EVALUATOR_EXTRA_EXCLUDE` — a prompt-jailbroken or tool-confused Evaluator is physically unable to mutate, dispatch, or change plan state. Planner kept as read-only inspection role (no bash, no mutation, no dispatch, no user interaction). Scout/Worker carry the full execution surface. Generator unchanged behaviorally; new tools (web/repo-intel/coordinator) added to schema. New contract test [packages/coding/src/task-engine/runner-driven-tool-wiring.test.ts](packages/coding/src/task-engine/runner-driven-tool-wiring.test.ts) pins each role's full tool-name set against `getAmaRoleExpectedToolNames(role)` derivation plus spot-checks for FEATURE_120 / FEATURE_161 coverage / Evaluator boundary / Planner boundary / no-orphan invariant — 50 assertions total. Any future EXCLUDE-set change or registry addition surfaces as a concrete test failure, not a silent production schema gap. **Test impact**: 2507/2507 coding-package tests pass (240 files, +50 new contract assertions). No regression in `runner-driven.test.ts` (130 tests including FEATURE_165 race-regression), `child-executor.test.ts` (30), `task-stop` / `send-message` handler tests (28), `worker-role-prompt.test.ts`. Design doc: [docs/features/v0.7.40.md#feature_168--ama-agent-tool-wiringexclude-based-registry-as-source-of-truth](docs/features/v0.7.40.md#feature_168--ama-agent-tool-wiringexclude-based-registry-as-source-of-truth).
+- **FEATURE_169 — Pull-Tool Prompt Adoption Hardening** (commit `519af4b9`). Production trace after FEATURE_161 wiring + FEATURE_168 schema fix surfaced 3 residual adoption gaps the wiring alone did not close: (1) **Worker hand-feeding bash in `dispatch_child_task.objective`** — 18% of production dispatch (3/17 in 2026-05-15 audit) embedded literal `git diff v0.7.39..HEAD`-style command directives, overriding the child's prompt-side tool teaching; 0/17 objectives recommended a pull-tool family. (2) **Child agent prompt stayed read+grep-first** — `CHILD_AGENT_SYSTEM_PROMPT` had taught "3-8 parallel tool calls (glob + grep + key file reads)" since v0.7.18 with no mention of pull-tools; children defaulted to grep/read in review and exploration tasks. (3) **Worker self-review still picked `bash git diff` first** — F7 taught tool existence but not "for review tasks, use which one". Three localized prompt strengthenings (F0a / F0b / F1v2 / F3) ship; F2 (3-tier order injection) **rejected** post-eval as zero-value churn. F0a teaches Worker to keep dispatch objective as data ("scope: v0.7.39..HEAD") not command ("使用 `git diff v0.7.39..HEAD`"); F0b adds an explicit pull-tool-family recommendation to the dispatch objective teaching; F1v2 adds child-agent reverse-steering toward pull-tools when the task is review/exploration-shaped; F3 reframes the Worker change-review surface from `bash git diff` first to `module_context` / `symbol_context` first. No wiring change, no new tool, no new permission. Prompt eval (`tests/feature-169-pull-tool-adoption.eval.ts`) clears SHIP gate (4/4 aliases, F0a/F0b/F1v2/F3 each ≥80% on 3 cases). Design doc: `docs/features/v0.7.40.md#feature_169--pull-tool-prompt-adoption-hardening-worker-dispatch-objective--child-reverse-steering--change-review-reframe`.
+- **FEATURE_134 follow-up: vision capability flag widened to 9 additional providers**. User empirically validated 2026-05-13 that `kimi-code` accepts and processes image input despite its v0.7.40 RC snapshot flag claiming `multimodalSupport: 'none'`. Root cause: AMA path (`runner-driven.ts` `provider.stream` direct call) bypasses the SA-path `applyProviderPolicyGate` (`run-substrate.ts:660`) where multimodal block enforcement lives, so the latent flag mismatch was never observable in production. The pre-v0.7.40 RC flag was over-conservative: every Anthropic-compat clone inherits the image-block forwarding serializer at `anthropic.ts:770`, and every OpenAI-compat clone inherits the `image_url` forwarding serializer at `openai.ts:904`. Flag widened in `packages/ai/src/providers/registry.ts` from `NATIVE_PROVIDER_CAPABILITY_PROFILE` to `IMAGE_INPUT_NATIVE_PROVIDER_CAPABILITY_PROFILE` for: **Anthropic-compat clones** — `kimi-code`, `zhipu-coding`, `mimo-coding`, `ark-coding`, `minimax-coding`; **OpenAI-compat clones** — `deepseek`, `kimi`, `qwen`, `zhipu`. Plus a separate follow-up commit (`71d45783`) wires **Gemini-CLI** vision via the new `serializeImageBlockToPromptToken` extension point on `KodaXAcpProvider` (Gemini CLI 2.x `@<path>` file-include syntax), bringing the final post-release total to **12 vision-capable providers (was 2)** — only `codex-cli` remains text-only because `codex exec --json --full-auto` has no image-input surface today. Production semantics: the flag controls only KodaX's SA-path artificial block; the actual model-level vision contract remains the upstream provider's responsibility — if a specific model alias is text-only, users now see the real API error from the provider instead of a KodaX-side `[Provider Policy] multimodal requests are unsupported` rejection. `capability-profile.test.ts` gains an explicit pin test asserting all 12 vision-capable providers report `multimodalSupport === 'image-input'` and `codex-cli` reports `'none'` so future regressions are caught.
+
+### Internal / architecture
+
+- **Directory rename: `packages/ai/` → `packages/llm/` for package-name parity**. Closes the directory-vs-npm-name discrepancy left behind by FEATURE_147 v0.7.37, which renamed the npm package from `@kodax/ai` to `@kodax-ai/llm` but left the on-disk directory at `packages/ai/`. New contributors had to mentally translate "ai is actually llm"; prompt examples and tool-description hints saying "Audit packages/ai" pointed to a path that disagreed with the package name. Scope: `git mv packages/ai packages/llm` + 8 tsconfig path/reference updates (root + agent + coding + mcp + session-lineage) + 2 real relative imports in `tests/feature-116-active-cache-control.eval.ts` + ~30 prompt/comment/eval-fixture string updates (`worker-role-prompt.ts` / `role-prompt.ts` / `tools/registry.ts` / `tools/todo-list.test.ts` / `paste/persist-image.ts` / 5 `tests/*.eval.ts` files / 5 `benchmark/datasets/*` fixture files including `h2-plan-execute-boundary/cases.ts` path-prefix `mustNotTouchFiles` assertions). Frozen historical artifacts NOT touched: `benchmark/results/**` snapshots, `.agent/**` runner cache, `.repointel/**` index cache, `docs/CHANGELOG_ARCHIVE.md` + `docs/features/v0.7.0-39.md` historical design records, and `.claude/settings.local.json` user-private allowlist. Active operational docs (CHANGELOG, README, README_CN, DD.md, KNOWN_ISSUES.md, root CLAUDE.md, docs/CLAUDE.md) all updated. No behavior change; git rename detection preserves blame across the move. Pure naming-consistency refactor.
+- **`@kodax/coding` blob summarizer module** (`packages/coding/src/tools/blob-summarizer.ts`, 187 lines). `createBlobSummarizer({provider, model, timeoutMs?})` returns a `SummarizeBlob` callback. Combines caller's `abortSignal` with a 30-second timeout via a fresh `AbortController` + listener fan-out. Throws `BlobSummarizerError` on empty input / empty output / provider error. System prompt + user prompt builder exported as constants (`SUMMARIZER_SYSTEM_PROMPT`, `buildSummarizerUserMessage`) so the Layer 2 eval can pin the EXACT production prompt text. 8 deterministic-shell unit tests cover the contract surface (timeout / abort / empty rejection / error wrapping / threshold constants).
+- **`KodaXToolExecutionContext.summarizeBlob?` field** (`packages/coding/src/types.ts`). Optional callback shape `(content: string, options?: {readonly maxChars?: number; readonly abortSignal?: AbortSignal}) => Promise<string>`. Wired in `runner-driven.ts` baseCtx with lazy-once memoization bound to the Worker's own provider/model. The dispatch tool calls it via `ctx.summarizeBlob` without owning provider construction — preserves `@kodax/agent` layer independence (agent stays unaware of LLM client).
+- **`GuardedToolResult.spillFailed?` flag** (`packages/coding/src/tools/tool-result-policy.ts`). Set when `persistToolOutput` throws and content was returned inline as the data-loss-guard fallback. Callers that need an LLM-summary follow-up (`dispatch-child-tasks` for `child_task_summary` >100KB) branch on this flag. The `console.warn` is intentionally NOT gated on `KODAX_DEBUG_TOOL_GUARDRAILS` — disk failure is a severe operational event an operator must see immediately.
+- **`applyChildSummaryGuardrailWithSummarizer` helper in `dispatch-child-tasks.ts`**. Unifies the 4 dispatch call sites (async success / async crash / sync success / sync failure) through a single helper that chains: `applyToolResultGuardrail` → if `spillFailed` + content > 100KB + `ctx.summarizeBlob` exists, attempt LLM summarize → on success, banner-wrap with LOSSY marker → on summarizer failure, console.warn + fall back to inline full content with emergency banner. Symmetric error handling across all 4 sites; banner strings consistent.
+- **3 new tracker entries marked Released**: FEATURE_121 (Envelope Spillover Gap-Fix), FEATURE_134 (Image / Screenshot Paste Input). Both originally `Planned` → now `v0.7.40 Released`. FEATURE_LIST.md `Current released version` bumped from v0.7.39 to v0.7.40.
+- **Layer 2 eval datasets** at `benchmark/datasets/feature-121-envelope-spillover/` (3 cases: `preview_sufficient` / `detail_required` / `inline_no_spillover`) and `benchmark/datasets/feature-121-blob-summarizer/` (2 cases: `audit_report` / `grep_findings`). Eval drivers at `tests/feature-121-envelope-spillover.eval.ts` and `tests/feature-121-blob-summarizer.eval.ts` (gated on `KODAX_EVAL_F121_SUMMARIZER=1`). Raw outputs preserved per `EVAL_GUIDELINES.md` §"Raw output preservation" under `os.tmpdir()/kodax-eval-dumps/feature-121-*/`.
+
+### Test coverage delta
+
+- New: 8 blob-summarizer unit tests + 25 blob-summarizer dataset shape tests + 27 envelope-spillover dataset shape tests + 9 tool-output-gc tests + 4 worker-role-prompt new tests (LARGE CHILD OUTPUT block + spill-path hint). Plus FEATURE_134's 52 new tests (45 foundation + 7 integration in `prompt-input-controller`).
+- Total green at HEAD: **5745 tests pass** (507 test files, 1 skipped, 23 todo).
+
 ## [0.7.39] - 2026-05-12
 
 ### Theme

package/README.md

@@ -81,6 +81,37 @@ If you need a custom base URL or an OpenAI/Anthropic-compatible endpoint, define
 
 `userAgentMode` defaults to `"compat"`, which sends `KodaX` instead of the official SDK User-Agent. Switch it to `"sdk"` only when your gateway expects the upstream SDK header.
 
+#### Opting a custom provider into image / vision input (FEATURE_134 v0.7.40)
+
+If your custom provider's underlying model supports image input (vision), add a `capabilityProfile.multimodalSupport: "image-input"` block so KodaX does not artificially block multimodal requests at the SA-path policy gate. The 12 built-in vision-capable providers (Anthropic, OpenAI, the 9 Anthropic-/OpenAI-compat clones — DeepSeek, Kimi, Kimi-code, Qwen, Zhipu, Zhipu-coding, MiniMax-coding, MiMo-coding, Ark-coding — plus Gemini-CLI via the CLI's `@<path>` file-include syntax) already ship with this flag enabled by default; only Codex-CLI and custom providers need to opt in.
+
+```json
+{
+  "customProviders": [
+    {
+      "name": "my-vision-provider",
+      "protocol": "openai",
+      "baseUrl": "https://example.com/v1",
+      "apiKeyEnv": "MY_LLM_API_KEY",
+      "model": "my-vision-model",
+      "capabilityProfile": {
+        "transport": "native-api",
+        "conversationSemantics": "full-history",
+        "mcpSupport": "none",
+        "contextFidelity": "full",
+        "toolCallingFidelity": "full",
+        "sessionSupport": "full",
+        "longRunningSupport": "full",
+        "multimodalSupport": "image-input",
+        "evidenceSupport": "full"
+      }
+    }
+  ]
+}
+```
+
+The serializer layer (`packages/llm/src/providers/anthropic.ts:770` for Anthropic-compat, `openai.ts:904` for OpenAI-compat) forwards image blocks automatically through base-class inheritance. The flag only gates whether KodaX's policy layer pre-rejects multimodal requests — the model-level vision contract remains your upstream provider's responsibility. If the model is actually text-only, you'll see the real upstream API error instead of a KodaX-side rejection.
+
 ### 3. Start in REPL or run a one-shot task
 
 ```bash
@@ -674,6 +705,33 @@ kodax -h team          # Multi-agent parallel execution
 kodax -h print         # Print configuration
 ```
 
+### Environment Variables
+
+KodaX recognizes a number of environment variables for tuning runtime behavior. The most commonly used ones are listed below; for the full list, search the repo for `process.env.KODAX_`.
+
+#### `KODAX_MAX_OUTPUT_TOKENS`
+
+Overrides the per-turn `max_tokens` value sent to **every** provider (Anthropic, OpenAI, Zhipu, Kimi, MiniMax, Qwen, DeepSeek, MiMo, Gemini, Codex, …). Set to a positive integer; unset or non-numeric values are ignored. This is an **explicit user intent**: when set, it wins over the provider's model descriptor cap, over the provider config default, and over the global `KODAX_MAX_TOKENS` fallback. The runtime's automatic safety caps (e.g. the v0.7.28 P2b RST-prone write-turn cap that limits write/edit turns to 8K tokens on Zhipu/Kimi/MiniMax) are **bypassed** when this variable is set, so the user override is also a way to opt out of those caps.
+
+```bash
+# Allow up to 48K output tokens per turn (use a higher cap when generating long files)
+export KODAX_MAX_OUTPUT_TOKENS=48000
+kodax "generate the full implementation"
+
+# Unset to restore default behavior
+unset KODAX_MAX_OUTPUT_TOKENS
+```
+
+Precedence used by every provider's `getEffectiveMaxOutputTokens()` (see `packages/llm/src/providers/base.ts`):
+
+1. One-shot per-request override (agent-loop escalation / context-overflow recovery — internal)
+2. **`KODAX_MAX_OUTPUT_TOKENS`** (this variable, explicit user intent)
+3. Active model descriptor's `maxOutputTokens` (FEATURE_098 per-model cap)
+4. Provider config default
+5. Global `KODAX_MAX_TOKENS` fallback
+
+Related variables: `KODAX_MAX_TOKENS` (global fallback when no provider/model cap applies), `KODAX_RST_PRONE_PROVIDERS` and `KODAX_WRITE_TURN_MAX_TOKENS` (v0.7.28 P2b write-turn safety cap configuration), `KODAX_ESCALATED_MAX_OUTPUT_TOKENS` (escalation budget used by the agent loop when a turn returns `stop_reason: max_tokens`).
+
 ## Advanced Library Usage
 
 #### Simple Mode (runKodaX)

package/README_CN.md

@@ -131,6 +131,37 @@ import { loadConfig } from '@kodax-ai/kodax/repl';     // REPL 配置 / session
 
 `userAgentMode` 默认 `"compat"`（发送 `KodaX` 而非上游 SDK 的 User-Agent）；如果你的网关要求原生 SDK header，再切到 `"sdk"`。
 
+#### 给自定义 provider 开图片 / vision 输入（FEATURE_134 v0.7.40）
+
+如果你的自定义 provider 后面的模型支持 vision，加 `capabilityProfile.multimodalSupport: "image-input"` 显式开启，KodaX 的 SA-path policy gate 就不会人为拦截多模态请求。内置的 12 个 vision-capable provider（Anthropic、OpenAI、9 个 Anthropic-/OpenAI-compat clone：DeepSeek / Kimi / Kimi-code / Qwen / Zhipu / Zhipu-coding / MiniMax-coding / MiMo-coding / Ark-coding，加 Gemini-CLI 通过 CLI 的 `@<path>` file-include 语法）已经默认开了这个 flag。只有 Codex-CLI 和自定义 provider 需要手动 opt-in。
+
+```json
+{
+  "customProviders": [
+    {
+      "name": "my-vision-provider",
+      "protocol": "openai",
+      "baseUrl": "https://example.com/v1",
+      "apiKeyEnv": "MY_LLM_API_KEY",
+      "model": "my-vision-model",
+      "capabilityProfile": {
+        "transport": "native-api",
+        "conversationSemantics": "full-history",
+        "mcpSupport": "none",
+        "contextFidelity": "full",
+        "toolCallingFidelity": "full",
+        "sessionSupport": "full",
+        "longRunningSupport": "full",
+        "multimodalSupport": "image-input",
+        "evidenceSupport": "full"
+      }
+    }
+  ]
+}
+```
+
+序列化层（Anthropic-compat 走 `packages/llm/src/providers/anthropic.ts:770`，OpenAI-compat 走 `openai.ts:904`）通过基类继承自动转发 image block。这个 flag 只控制 KodaX 自身是否预先拒绝多模态请求 —— 上游模型到底支不支持 vision 由 provider 自己决定。如果模型实际是 text-only，你会看到真实的上游 API 错误，而不是 KodaX 一侧的 `[Provider Policy] multimodal requests are unsupported` 预拦截。
+
 库模式下用 `registerCustomProviders()` 显式注册：
 
 ```typescript

package/dist/chunks/chunk-6QO6HWGU.js

@@ -0,0 +1,30 @@
+// @kodax-ai/kodax — bundled distribution. See docs/ADR.md ADR-022 + ADR-024.
+import{a as i}from"./chunk-V4WSBIXB.js";var H=class extends Error{static{i(this,"KodaXError")}code;constructor(e,n="KODAX_ERROR"){super(e),this.code=n,this.name="KodaXError"}},R=class extends H{static{i(this,"KodaXProviderError")}provider;constructor(e,n){super(e,"PROVIDER_ERROR"),this.provider=n,this.name="KodaXProviderError"}},de=class extends H{static{i(this,"KodaXRateLimitError")}retryAfter;constructor(e,n){super(e,"RATE_LIMIT_ERROR"),this.retryAfter=n,this.name="KodaXRateLimitError"}},Fe=class extends H{static{i(this,"KodaXNetworkError")}isTimeout;constructor(e,n=!1){super(e,"NETWORK_ERROR"),this.isTimeout=n,this.name="KodaXNetworkError"}},Ue=class extends H{static{i(this,"KodaXToolCallIdError")}constructor(e){super(e,"TOOL_CALL_ID_ERROR"),this.name="KodaXToolCallIdError"}};var Xe=32768,J=32e3,We=64e3,un=.5;var pn=["off","auto","quick","balanced","deep"],Rt={low:6e3,medium:1e4,high:2e4},At=4096;function $e(t){return t.reasoningCapability?t.reasoningCapability:t.supportsThinking?"native-toggle":"prompt-only"}i($e,"getReasoningCapability");function Q(t){return ke(t).enabled}i(Q,"isReasoningEnabled");function ke(t){if(typeof t=="boolean")return{enabled:t,mode:t?"auto":"off",depth:t?"medium":"off",taskType:"unknown",executionMode:"implementation"};let e=t?.mode??"off",n=t?.depth??St(e),o=t?.enabled??(e!=="off"&&n!=="off");return{enabled:o&&e!=="off"&&n!=="off",mode:e,depth:o?n:"off",taskType:t?.taskType??"unknown",executionMode:t?.executionMode??"implementation"}}i(ke,"normalizeReasoningRequest");function St(t){switch(t){case"quick":return"low";case"balanced":case"auto":return"medium";case"deep":return"high";default:return"off"}}i(St,"getDefaultThinkingDepthForMode");function q(t,e,n="unknown"){if(e==="off")return 0;let o={...Rt,...t.defaultThinkingBudgets??{}},s=t.taskBudgetOverrides?.[n]?.[e]??o[e];return t.thinkingBudgetCap?Math.min(s,t.thinkingBudgetCap):s}i(q,"resolveThinkingBudget");function Z(t,e,n=At){let o=Math.max(1024,e-n);return Math.max(1024,Math.min(t,o))}i(Z,"clampThinkingBudget");function je(t){switch(t){case"low":return"low";case"medium":return"medium";case"high":return"high";default:return}}i(je,"mapDepthToOpenAIReasoningEffort");import xe from"fs";import dn from"os";import He from"path";var Ge=null;function mn(t){return t==="budget"||t==="effort"||t==="toggle"||t==="none"}i(mn,"isReasoningOverride");function fn(t){if(!t||typeof t!="object"||Array.isArray(t))return!1;let e=t.providerReasoningOverrides;return e===void 0?!0:!e||typeof e!="object"||Array.isArray(e)?!1:Object.values(e).every(mn)}i(fn,"isStoredConfig");function gn(){return process.env.KODAX_HOME??He.join(dn.homedir(),".kodax")}i(gn,"getKodaxDir");function Ot(){return process.env.KODAX_CONFIG_FILE??He.join(gn(),"config.json")}i(Ot,"getConfigFilePath");function ze(t,e){return Ge={filePath:t,config:e},e}i(ze,"updateStoredConfigCache");function hn(t){let e=JSON.parse(xe.readFileSync(t,"utf-8"));return fn(e)?e:{}}i(hn,"readStoredConfigFromDisk");function Ve(t){switch(t){case"native-budget":return"budget";case"native-effort":return"effort";case"native-toggle":return"toggle";case"none":return"none";default:return}}i(Ve,"reasoningCapabilityToOverride");function Ye(t){switch(t){case"budget":return"native-budget";case"effort":return"native-effort";case"toggle":return"native-toggle";default:return"none"}}i(Ye,"reasoningOverrideToCapability");function ee(t,e,n){return[t,e.baseUrl??"",n??e.model].join("|")}i(ee,"buildReasoningOverrideKey");function Je(){let t=Ot();if(Ge?.filePath===t)return Ge.config;try{if(xe.existsSync(t))return ze(t,hn(t))}catch{}return ze(t,{})}i(Je,"loadStoredConfig");function It(t){let e=Ot();try{xe.mkdirSync(He.dirname(e),{recursive:!0}),xe.writeFileSync(e,JSON.stringify(t,null,2)),ze(e,t)}catch(n){process.env.KODAX_DEBUG_OVERRIDES&&console.error("[ReasoningOverride] Failed to save config:",n)}}i(It,"saveStoredConfig");function be(t,e,n){let o=Je(),r=ee(t,e,n);return o.providerReasoningOverrides?.[r]}i(be,"loadReasoningOverride");function Qe(t,e,n,o){let r=Je(),s=ee(t,e,o);r.providerReasoningOverrides={...r.providerReasoningOverrides??{},[s]:n},It(r)}i(Qe,"saveReasoningOverride");function yn(t,e,n){let o=Je(),r=ee(t,e,n);if(!o.providerReasoningOverrides?.[r])return;let s={...o.providerReasoningOverrides};delete s[r],o.providerReasoningOverrides=Object.keys(s).length>0?s:void 0,It(o)}i(yn,"clearReasoningOverride");function Dt(t,e){if(!t)return;if(typeof t.get=="function")return t.get(e)??void 0;let n=t,o=e.toLowerCase(),r=n[e]??n[o]??n[o.replace(/\b\w/g,s=>s.toUpperCase())];if(r!==void 0)return Array.isArray(r)?r[0]:r}i(Dt,"readHeader");function Kt(t){if(!t)return;let e=t.trim();if(e.length===0)return;let n=Number(e);if(!(!Number.isFinite(n)||n<=0))return n}i(Kt,"parsePositiveNumber");function _n(t,e){if(!t)return;let n=t.trim();if(n.length===0)return;let o=Date.parse(n);if(!Number.isFinite(o))return;let r=o-e;return r>0?r:void 0}i(_n,"parseHttpDate");function ve(t,e){let n=e.baseBackoffMs??1e3,o=e.maxBackoffMs??3e4,r=e.maxHeaderWaitMs??12e4,s=e.now?e.now():Date.now(),a=Kt(Dt(t,"retry-after-ms"));if(a!==void 0){let _=Math.min(a,r);return{type:"header",waitMs:Math.round(_),source:"retry-after-ms",cappedFromHeader:_!==a}}let l=Dt(t,"retry-after");if(l!==void 0&&l.trim().length>0){let _=Kt(l);if(_!==void 0){let x=_*1e3,u=Math.min(x,r);return{type:"header",waitMs:Math.round(u),source:"retry-after-seconds",cappedFromHeader:u!==x}}let M=_n(l,s);if(M!==void 0){let x=Math.min(M,r);return{type:"header",waitMs:Math.round(x),source:"retry-after-date",cappedFromHeader:x!==M}}}let c=n*Math.pow(2,Math.max(0,e.attempt)),m=Math.min(c,o),f=e.withJitter!==!1?Math.random()*.25*m:0;return{type:"backoff",waitMs:Math.round(m+f),source:"exponential-backoff",attempt:e.attempt}}i(ve,"parseRetryAfter");function Me(t){if(!t||typeof t!="object")return;let e=t;return e.headers??e.response?.headers??e.cause?.headers??e.cause?.response?.headers}i(Me,"extractHeadersFromError");var me={transport:"native-api",conversationSemantics:"full-history",mcpSupport:"none",contextFidelity:"full",toolCallingFidelity:"full",sessionSupport:"full",longRunningSupport:"full",multimodalSupport:"none",evidenceSupport:"full"},L={...me,multimodalSupport:"image-input"},fe={transport:"cli-bridge",conversationSemantics:"last-user-message",mcpSupport:"none",contextFidelity:"lossy",toolCallingFidelity:"limited",sessionSupport:"stateless",longRunningSupport:"limited",multimodalSupport:"none",evidenceSupport:"limited"},Lt={...fe,multimodalSupport:"image-input"};function Ce(t){return{transport:t.transport,conversationSemantics:t.conversationSemantics,mcpSupport:t.mcpSupport,contextFidelity:t.contextFidelity??"full",toolCallingFidelity:t.toolCallingFidelity??"full",sessionSupport:t.sessionSupport??"full",longRunningSupport:t.longRunningSupport??"full",multimodalSupport:t.multimodalSupport??"none",evidenceSupport:t.evidenceSupport??"full"}}i(Ce,"normalizeCapabilityProfile");function F(t){return{...Ce(t)}}i(F,"cloneCapabilityProfile");function kn(t){if(!t)return;let e=parseInt(t,10);return Number.isFinite(e)&&e>0?e:void 0}i(kn,"parseEnvInt");var G=class{static{i(this,"KodaXBaseProvider")}maxOutputTokensOverride;setMaxOutputTokensOverride(e){this.maxOutputTokensOverride=e}getEffectiveMaxOutputTokens(e){if(this.maxOutputTokensOverride!==void 0)return this.maxOutputTokensOverride;let n=kn(process.env.KODAX_MAX_OUTPUT_TOKENS);if(n!==void 0)return n;let o=this.getModelDescriptor(e)?.maxOutputTokens;return o!==void 0?o:this.config.maxOutputTokens??32768}getStreamMaxDurationMs(){return this.config.streamMaxDurationMs}supportsNonStreamingFallback(){return!1}async complete(e,n,o,r,s,a){throw new R(`${this.name} does not support non-streaming fallback`)}isConfigured(){return!!process.env[this.config.apiKeyEnv]}getModel(){return this.config.model}getAvailableModels(){return this.config.models?.length?[...new Set([this.config.model,...this.config.models.map(e=>e.id)])]:[this.config.model]}getModelDescriptor(e){return!e||e===this.config.model?{id:this.config.model}:this.config.models?.find(n=>n.id===e)}getBaseUrl(){return this.config.baseUrl}getApiKeyEnv(){return this.config.apiKeyEnv}getCapabilityProfile(){return F(this.config.capabilityProfile??me)}getConfiguredReasoningCapability(e){let n=this.getModelDescriptor(e);return n?.reasoningCapability?n.reasoningCapability:$e(this.config)}getReasoningCapability(e){let n=be(this.name,this.config,e);return n?Ye(n):this.getConfiguredReasoningCapability(e)}getReasoningOverride(e){return be(this.name,this.config,e)}getReasoningOverrideKey(e){return ee(this.name,this.config,e)}persistReasoningCapabilityOverride(e,n){let o=Ve(e);o&&Qe(this.name,this.config,o,n)}shouldFallbackForReasoningError(e,...n){let o=e instanceof Error?e.message.toLowerCase():String(e).toLowerCase(),s=n.map(l=>l.toLowerCase()).some(l=>o.includes(l)),a=o.includes("parameter")||s;return o.includes("unknown parameter")||o.includes("invalid parameter")||o.includes("unsupported")&&a}shouldFallbackForSpecificReasoningError(e,...n){let o=e instanceof Error?e.message.toLowerCase():String(e).toLowerCase();return n.map(a=>a.toLowerCase()).some(a=>o.includes(a))?o.includes("unknown parameter")||o.includes("invalid parameter")||o.includes("unsupported"):!1}getReasoningFallbackChain(e){switch(e){case"native-budget":return["native-budget","native-toggle","none"];case"native-effort":return["native-effort","none"];case"native-toggle":return["native-toggle","none"];default:return["none"]}}getContextWindow(){return this.getEffectiveContextWindow()}getEffectiveContextWindow(e){let n=this.getModelDescriptor(e)?.contextWindow;return n!==void 0?n:this.config.contextWindow??2e5}getApiKey(){let e=process.env[this.config.apiKeyEnv];if(!e)throw new Error(`${this.config.apiKeyEnv} not set`);return e}shouldLogStreamDiagnostics(){return!!process.env.KODAX_DEBUG_STREAM}logStreamDiagnostic(...e){this.shouldLogStreamDiagnostics()&&console.error(...e)}normalizeReasoning(e){return ke(e)}onStaleConnection(){}isRateLimitError(e){if(!(e instanceof Error))return!1;let n=e.message.toLowerCase();return["rate","limit","\u901F\u7387","\u9891\u7387","1302","429","too many","overload","overwhelmed","503","529","busy"].some(o=>n.includes(o))}classifyRateLimitReason(e){let n=e instanceof Error?e.message.toLowerCase():String(e).toLowerCase();return n.includes("overload")||n.includes("overwhelmed")||n.includes("503")||n.includes("529")||n.includes("busy")?"overloaded":"rate-limit"}extractRetryAfterMs(e){let n=Me(e);if(!n)return;let o=ve(n,{attempt:0,withJitter:!1});return o.type==="header"?o.waitMs:void 0}parseContextOverflow(e){let n=String(e?.message??""),o=[/(\d[\d,]*)\s*tokens?.*?(\d[\d,]*)\s*(?:maximum|limit|context)/i,/maximum.*?(\d[\d,]*)\s*tokens?.*?requested.*?(\d[\d,]*)/i,/exceeds?\s+.*?(\d[\d,]*)\s*.*?(?:limit|max|上限).*?(\d[\d,]*)/i];for(let r of o){let s=n.match(r);if(s){let a=Number(s[1].replace(/,/g,"")),l=Number(s[2].replace(/,/g,"")),c=Math.min(a,l),m=Math.max(a,l);return Math.max(3e3,m-c-1e3)}}}isContextOverflowError(e){let n=String(e?.message??"").toLowerCase();return n.includes("prompt is too long")||n.includes("prompt too long")||n.includes("context length")||n.includes("context_length_exceeded")||n.includes("context window")||n.includes("\u4E0A\u4E0B\u6587\u957F\u5EA6")}async withRateLimit(e,n,o=3,r,s){for(let a=0;a<o;a++)try{let l=await e();return this.maxOutputTokensOverride=void 0,l}catch(l){if(this.isContextOverflowError(l)&&!this.maxOutputTokensOverride){let c=this.parseContextOverflow(l);if(c){this.maxOutputTokensOverride=c,r?.(a+1,o,0);continue}}if(this.isRateLimitError(l)){if(a===o-1)throw new de(`API rate limit exceeded after ${o} retries. Please wait and try again later.`,6e4);let c=Me(l)??{},m=ve(c,{attempt:a,baseBackoffMs:500,maxBackoffMs:32e3,withJitter:!0}),v=m.waitMs,f=this.classifyRateLimitReason(l);if(s?.({provider:this.name,waitMs:v,reason:f,source:m.source,attempt:a+1,maxAttempts:o}),r?r(a+1,o,v):s||console.log(`[Rate Limit] Retrying in ${v/1e3}s (${a+1}/${o})...`),n?.aborted)throw new DOMException("Request aborted","AbortError");if(await new Promise(_=>setTimeout(_,v)),n?.aborted)throw new DOMException("Request aborted","AbortError");continue}if(l instanceof Error){if((l.name==="AbortError"||l.name==="APIUserAbortError")&&n?.aborted)throw l.name==="AbortError"?l:new DOMException(l.message||"Request aborted","AbortError");let c=l?.cause?.code??l?.code??"";throw(c==="ECONNRESET"||c==="EPIPE")&&this.onStaleConnection(),new R(`${this.name} API error: ${l.message}`,this.name)}throw l}throw new H("Unexpected end of withRateLimit")}};import Cn from"@anthropic-ai/sdk";import{parse as xn}from"partial-json";function te(t){if(!t)return{};try{let e=JSON.parse(t);return e&&typeof e=="object"&&!Array.isArray(e)?e:{}}catch{}try{let e=xn(t);return process.env.KODAX_DEBUG_TOOL_STREAM&&console.warn("[Tool Block Salvaged] partial JSON recovered, rawLength=",t.length),e&&typeof e=="object"&&!Array.isArray(e)?e:{}}catch{return{}}}i(te,"parseToolInputWithSalvage");function qe(t,e){let n=t[t.length-1];if(n&&ne(n)&&n.hint===e)return t;let o=e?{type:"cache-boundary",hint:e}:{type:"cache-boundary"};return[...t,o]}i(qe,"insertCacheBoundary");function ne(t){if(typeof t!="object"||t===null||t.type!=="cache-boundary")return!1;let e=Object.keys(t);return e.length<=2&&e.every(n=>n==="type"||n==="hint")}i(ne,"isCacheBoundary");function Ze(t,e){let n=[];for(let o of t){if(ne(o)){if(e==="attach"&&n.length>0){let r=n[n.length-1];n[n.length-1]={...r,cache_control:{type:"ephemeral"}}}continue}n.push(o)}return n}i(Ze,"lowerCacheBoundaries");function ge(t){return t.filter(e=>!ne(e))}i(ge,"stripCacheBoundaries");import{readFile as bn}from"node:fs/promises";import vn from"node:path";var Mn={".gif":"image/gif",".jpeg":"image/jpeg",".jpg":"image/jpeg",".png":"image/png",".webp":"image/webp"};function et(t,e){return e??Mn[vn.extname(t).toLowerCase()]??"image/png"}i(et,"resolveImageMediaType");async function tt(t){return(await bn(t)).toString("base64")}i(tt,"readImageFileAsBase64");async function Nt(t,e){let n=et(t,e),o=await tt(t);return`data:${n};base64,${o}`}i(Nt,"buildImageDataUrl");var Tn="KodaX";function En(t){return t.userAgentMode==="sdk"?void 0:{"User-Agent":Tn}}i(En,"getAnthropicCompatDefaultHeaders");function nt(t,e){if(!t)return e;let n=t.input_tokens!==void 0&&t.input_tokens!==null||t.cache_creation_input_tokens!==void 0&&t.cache_creation_input_tokens!==null||t.cache_read_input_tokens!==void 0&&t.cache_read_input_tokens!==null,o=typeof t.input_tokens=="number"?t.input_tokens:n?0:e?.inputTokens??0,r=typeof t.cache_creation_input_tokens=="number"?t.cache_creation_input_tokens:n?0:e?.cachedWriteTokens??0,s=typeof t.cache_read_input_tokens=="number"?t.cache_read_input_tokens:n?0:e?.cachedReadTokens??0,a=typeof t.output_tokens=="number"?t.output_tokens:e?.outputTokens??0,l=n?o+r+s:e?.inputTokens??0;if(![l,a].some(c=>!Number.isFinite(c)||c<0))return{inputTokens:l,outputTokens:a,totalTokens:l+a,cachedReadTokens:s||void 0,cachedWriteTokens:r||void 0}}i(nt,"normalizeAnthropicUsage");var K=class extends G{static{i(this,"KodaXAnthropicCompatProvider")}supportsThinking=!0;client;initClient(){let e=En(this.config);this.client=new Cn({apiKey:this.getApiKey(),baseURL:this.config.baseUrl,...e?{defaultHeaders:e}:{}})}onStaleConnection(){this.initClient()}applyCacheControlToSystem(e){if(!e.trim()||process.env.KODAX_DISABLE_PROMPT_CACHE==="1")return e;let n=qe([{type:"text",text:e}],"system");return Ze(n,"attach")}applyCacheControlToTools(e){if(e.length===0||process.env.KODAX_DISABLE_PROMPT_CACHE==="1")return e;let n=e.slice(),o=n[n.length-1];return n[n.length-1]={...o,cache_control:{type:"ephemeral"}},n}async stream(e,n,o,r,s,a){return this.withRateLimit(async()=>{let l=this.normalizeReasoning(r),c=s?.modelOverride??this.config.model,m=this.getEffectiveMaxOutputTokens(c),v=await this.convertMessages(e),f=l.enabled?this.getReasoningCapability(c):"none",_=l.enabled?this.getReasoningFallbackChain(f).filter(k=>k==="native-budget"||k==="native-toggle"||k==="none"):["none"],M=i(k=>{let P={model:c,max_tokens:m,system:this.applyCacheControlToSystem(this.buildSystemPrompt(o,e)),messages:v,tools:this.applyCacheControlToTools(n),stream:!0};if(k==="native-budget"){let S=q(this.config,l.depth,l.taskType);P.thinking={type:"enabled",budget_tokens:Z(S,m)}}else k==="native-toggle"&&(P.thinking={type:"enabled"});return P},"buildRequest");if(a?.aborted)throw new DOMException("Request aborted","AbortError");let x=[],u=[],y=[],g,b=null,p="",d="",T="",w="",W="",O="",$="",E=!1,j,A=Date.now(),_e=Date.now(),le,ue;for(let k of _)try{le=await this.client.messages.create(M(k),a?{signal:a}:{}),k!==f&&this.persistReasoningCapabilityOverride(k,c);break}catch(P){ue=P;let S=k==="native-budget"?["budget_tokens","thinking"]:k==="native-toggle"?["thinking"]:[];if(!this.shouldFallbackForReasoningError(P,...S))throw P}if(!le)throw ue??new R("All reasoning capability attempts failed without a captured error",this.name);let pe=Date.now(),C=0,I=0,D=3e4;for await(let k of le){if(a?.aborted)throw new DOMException("Request aborted","AbortError");let P=Date.now(),S=P-pe;if(S>D&&(C++,I+=S,this.logStreamDiagnostic(`[Stream] stall detected: ${Math.round(S/1e3)}s gap before ${k.type}`,{stallCount:C,totalStallMs:I,eventType:k.type})),pe=P,k.type==="content_block_start"||k.type==="content_block_stop"?s?.onHeartbeat?.(!0):s?.onHeartbeat?.(),k.type==="content_block_start"){A=Date.now();let h=k.content_block;b=h.type,process.env.KODAX_DEBUG_TOOL_STREAM&&h.type==="tool_use"&&console.error("[ToolStream] content_block_start:",{type:h.type,id:h.id,name:h.name}),h.type==="thinking"?(d="",T=h.signature??""):h.type==="redacted_thinking"?(b="redacted_thinking",w=h.data??""):h.type==="text"?p="":h.type==="tool_use"&&(W=h.id,O=h.name,$="")}else if(k.type==="content_block_delta"){A=Date.now();let h=k.delta;h.type==="thinking_delta"?(d+=h.thinking??"",s?.onThinkingDelta?.(h.thinking??"")):h.type==="text_delta"?(p+=h.text??"",s?.onTextDelta?.(h.text??"")):h.type==="input_json_delta"&&($+=h.partial_json??"",s?.onToolInputDelta?.(O,h.partial_json??"",W?{toolId:W}:void 0))}else if(k.type==="content_block_stop")A=Date.now(),b==="thinking"?d&&(y.push({type:"thinking",thinking:d,signature:T}),s?.onThinkingEnd?.(d)):b==="redacted_thinking"?(w&&y.push({type:"redacted_thinking",data:w}),w=""):b==="text"?p&&x.push({type:"text",text:p}):b==="tool_use"&&(!W||!O?console.error("[Tool Block Invalid] Missing tool id or name:",{id:JSON.stringify(W),name:JSON.stringify(O),input:$.slice(0,100)}):u.push({type:"tool_use",id:W,name:O,input:te($)})),b=null;else if(k.type==="message_stop"){if(E=!0,A=Date.now(),process.env.KODAX_DEBUG_STREAM){let h=Date.now()-_e;this.logStreamDiagnostic(`[Stream] message_stop received after ${h}ms`)}}else if(k.type==="message_delta"){A=Date.now(),g=nt(k.usage,g);let h=k.delta;h?.stop_reason&&(j=h.stop_reason,process.env.KODAX_DEBUG_STREAM&&this.logStreamDiagnostic(`[Stream] message_delta with stop_reason: ${j}`))}else k.type==="message_start"&&(A=Date.now(),g=nt(k.message?.usage,g),process.env.KODAX_DEBUG_STREAM&&this.logStreamDiagnostic("[Stream] message_start received"))}if(!E){let k=Date.now()-_e,P=Date.now()-A;if(a?.aborted){let h=a.reason instanceof Error?a.reason.message:typeof a.reason=="string"?a.reason:"Request aborted";throw this.logStreamDiagnostic("[Stream] Stream ended after abort before message_stop:",{duration:k,lastEventAge:P,reason:h,textBlocks:x.length,toolBlocks:u.length,thinkingBlocks:y.length}),new DOMException(h,"AbortError")}let S=new Error(`Stream incomplete: message_stop event not received. Duration: ${k}ms, Last event: ${P}ms ago. This may indicate a network disconnection or API timeout.`);throw S.name="StreamIncompleteError",this.logStreamDiagnostic("[Stream] Incomplete stream detected:",{duration:k,lastEventAge:P,textBlocks:x.length,toolBlocks:u.length,thinkingBlocks:y.length}),S}return{textBlocks:x,toolBlocks:u,thinkingBlocks:y,usage:g,stopReason:j}},a,3,s?.onRateLimit,s?.onRetryAfter)}supportsNonStreamingFallback(){return!0}async complete(e,n,o,r,s,a){return this.withRateLimit(async()=>{let l=this.normalizeReasoning(r),c=s?.modelOverride??this.config.model,m=this.getEffectiveMaxOutputTokens(c),v=await this.convertMessages(e),f=l.enabled?this.getReasoningCapability(c):"none",_=l.enabled?this.getReasoningFallbackChain(f).filter(p=>p==="native-budget"||p==="native-toggle"||p==="none"):["none"],M=i(p=>{let d={model:c,max_tokens:m,system:this.applyCacheControlToSystem(this.buildSystemPrompt(o,e)),messages:v,tools:this.applyCacheControlToTools(n)};if(p==="native-budget"){let T=q(this.config,l.depth,l.taskType);d.thinking={type:"enabled",budget_tokens:Z(T,m)}}else p==="native-toggle"&&(d.thinking={type:"enabled"});return d},"buildRequest"),x,u;for(let p of _)try{x=await this.client.messages.create(M(p),a?{signal:a}:{}),p!==f&&this.persistReasoningCapabilityOverride(p,c);break}catch(d){u=d;let T=p==="native-budget"?["budget_tokens","thinking"]:p==="native-toggle"?["thinking"]:[];if(!this.shouldFallbackForReasoningError(d,...T))throw d}if(!x)throw u??new R("All reasoning capability attempts failed without a captured error",this.name);let y=[],g=[],b=[];for(let p of x.content)p.type==="text"?(y.push({type:"text",text:p.text}),s?.onTextDelta?.(p.text)):p.type==="thinking"?(b.push({type:"thinking",thinking:p.thinking,signature:p.signature??""}),s?.onThinkingDelta?.(p.thinking),s?.onThinkingEnd?.(p.thinking)):p.type==="redacted_thinking"?b.push({type:"redacted_thinking",data:p.data}):p.type==="tool_use"&&g.push({type:"tool_use",id:p.id,name:p.name,input:typeof p.input=="object"&&p.input!==null?p.input:{}});return{textBlocks:y,toolBlocks:g,thinkingBlocks:b,usage:nt(x.usage),stopReason:x.stop_reason??void 0}},a,3,s?.onRateLimit,s?.onRetryAfter)}serializeSystemMessageContent(e){if(typeof e=="string")return e.trim();for(let n of e)if(ne(n))throw new R("cache-boundary marker reached system message serialization unlowered. Provider base class lowering must run before any wire-level serialization.",this.name);return e.filter(n=>n.type==="text").map(n=>n.text.trim()).filter(Boolean).join(`
+`)}buildSystemPrompt(e,n){let o=n.filter(r=>r.role==="system").map(r=>this.serializeSystemMessageContent(r.content)).filter(Boolean);return[e,...o].filter(r=>typeof r=="string"&&r.trim().length>0).join(`
+
+`)}async convertMessages(e){let n=[];for(let o of e.filter(r=>r.role!=="system")){let r=o.role==="user"?"user":"assistant";if(typeof o.content=="string"){n.push({role:r,content:o.content});continue}let s=[],a=[];for(let c of o.content)c.type==="thinking"?!this.config.strictThinkingSignature||typeof c.signature=="string"&&c.signature.length>0?s.push({type:"thinking",thinking:c.thinking,signature:c.signature??""}):c.thinking&&a.push(c.thinking):c.type==="redacted_thinking"&&(this.config.strictThinkingSignature||s.push({type:"redacted_thinking",data:c.data}));a.length>0&&o.role==="assistant"&&s.push({type:"text",text:`<prior_reasoning>
+${a.join(`
+
+`)}
+</prior_reasoning>`});for(let c of o.content)c.type==="tool_result"&&o.role==="user"&&s.push({type:"tool_result",tool_use_id:c.tool_use_id,content:c.content,...c.is_error===!0?{is_error:!0}:{}});for(let c of o.content)c.type==="tool_use"&&o.role==="assistant"&&s.push({type:"tool_use",id:c.id,name:c.name,input:c.input});for(let c of o.content)c.type==="text"?s.push({type:"text",text:c.text}):c.type==="image"&&o.role==="user"&&s.push({type:"image",source:{type:"base64",media_type:et(c.path,c.mediaType),data:await tt(c.path)}});if(r==="assistant"&&this.config.supportsThinking&&!this.config.strictThinkingSignature){let c=s.some(v=>v.type==="tool_use"),m=s.some(v=>v.type==="thinking"||v.type==="redacted_thinking");c&&!m&&s.unshift({type:"thinking",thinking:"...",signature:""})}let l=s.length===0||r==="assistant"&&s.every(c=>{let m=c;return m.type==="thinking"&&!m.thinking||m.type==="text"&&!m.text});n.push({role:o.role,content:l?[{type:"text",text:"..."}]:s})}return n}};import Pn from"openai";var wn="KodaX";function Rn(t){return t.userAgentMode==="sdk"?void 0:{"User-Agent":wn}}i(Rn,"getOpenAICompatDefaultHeaders");function Bt(t){if(!t)return;let e=typeof t.prompt_tokens=="number"?t.prompt_tokens:0,n=typeof t.completion_tokens=="number"?t.completion_tokens:0,o=typeof t.total_tokens=="number"?t.total_tokens:e+n;if([e,n,o].some(s=>!Number.isFinite(s)||s<0)||o<e||o<n)return;let r=typeof t.prompt_tokens_details?.cached_tokens=="number"&&t.prompt_tokens_details.cached_tokens>=0?t.prompt_tokens_details.cached_tokens:typeof t.prompt_cache_hit_tokens=="number"&&t.prompt_cache_hit_tokens>=0?t.prompt_cache_hit_tokens:void 0;return{inputTokens:e,outputTokens:n,totalTokens:o,...r!==void 0?{cachedReadTokens:r}:{}}}i(Bt,"normalizeOpenAIUsage");function An(t){return t?t.type==="function"&&"function"in t:!1}i(An,"isOpenAIFunctionToolCall");function Sn(t){return typeof t=="string"?t:Array.isArray(t)?t.map(e=>{if(!e||typeof e!="object")return"";let n=Reflect.get(e,"text");if(typeof n=="string")return n;if(n&&typeof n=="object"){let o=Reflect.get(n,"value");return typeof o=="string"?o:""}return""}).filter(Boolean).join(""):""}i(Sn,"extractOpenAIMessageText");function On(t){if(!t||typeof t!="object")return"";let e=Reflect.get(t,"reasoning_content");return typeof e=="string"?e:Array.isArray(e)?e.map(n=>typeof n=="string"?n:n&&typeof n=="object"&&"text"in n&&typeof n.text=="string"?n.text:"").join(""):""}i(On,"extractOpenAIMessageReasoning");var N=class extends G{static{i(this,"KodaXOpenAICompatProvider")}supportsThinking=!0;client;initClient(){let e=Rn(this.config);this.client=new Pn({apiKey:this.getApiKey(),baseURL:this.config.baseUrl,...e?{defaultHeaders:e}:{}})}onStaleConnection(){this.initClient()}stripCacheBoundariesFromMessages(e){return e.map(n=>{if(typeof n.content=="string")return n;let o=ge(n.content);return o.length===n.content.length?n:{...n,content:o}})}normalizeSystemForWire(e,n){let o=[];e&&e.trim().length>0&&o.push(e);let r=[];for(let s of n){if(s.role!=="system"){r.push(s);continue}let a=typeof s.content=="string"?s.content:Array.isArray(s.content)?s.content.filter(l=>typeof l=="object"&&l!==null&&l.type==="text"&&typeof l.text=="string").map(l=>l.text).join(`
+`):"";a.trim().length>0&&o.push(a)}return{system:o.join(`
+
+`),rest:r}}appendExtraBody(e,n){let o=typeof e.extra_body=="object"&&e.extra_body!==null?e.extra_body:{};e.extra_body={...o,...n}}resetReasoningCapabilityParams(e){delete e.reasoning_effort,delete e.thinking;let n=typeof e.extra_body=="object"&&e.extra_body!==null?{...e.extra_body}:void 0;if(n){if(delete n.enable_thinking,delete n.thinking_budget,delete n.thinking,Object.keys(n).length===0){delete e.extra_body;return}e.extra_body=n}}applyReasoningCapability(e,n,o){let r=e,s=this.getEffectiveMaxOutputTokens(e.model),a=Z(q(this.config,o.depth,o.taskType),s);switch(n){case"native-effort":{let l=je(o.depth);l&&(r.reasoning_effort=l);break}case"native-budget":{this.name==="qwen"?this.appendExtraBody(r,{enable_thinking:!0,thinking_budget:a}):this.name==="zhipu"&&(r.thinking={type:"enabled",budget_tokens:a});break}case"native-toggle":{this.name==="qwen"?this.appendExtraBody(r,{enable_thinking:!0}):this.name==="zhipu"&&(r.thinking={type:"enabled"});break}default:break}}getFallbackTerms(e){switch(e){case"native-budget":return["thinking_budget","budget_tokens","thinking"];case"native-effort":return["reasoning_effort"];case"native-toggle":return["enable_thinking","thinking"];default:return[]}}async stream(e,n,o,r,s,a){return this.withRateLimit(async()=>{let l=this.stripCacheBoundariesFromMessages(e),{system:c,rest:m}=this.normalizeSystemForWire(o,l),v=[{role:"system",content:c},...await this.convertMessages(m)],f=n.map(C=>({type:"function",function:{name:C.name,description:C.description,parameters:C.input_schema}}));if(a?.aborted)throw new DOMException("Request aborted","AbortError");let _=new Map,M="",x="",u,y=!0,g=null,b=Date.now(),p=this.normalizeReasoning(r),d=s?.modelOverride??this.config.model,T=Q(p)?this.getReasoningCapability(d):"none",w=Q(p)?this.getReasoningFallbackChain(T).filter(C=>C==="native-budget"||C==="native-effort"||C==="native-toggle"||C==="none"):["none"],W={model:d,messages:v,tools:f,max_completion_tokens:this.getEffectiveMaxOutputTokens(d),stream:!0},O,$;for(let C of w){for(;!O;){let I={...W};y&&(I.stream_options={include_usage:!0}),this.resetReasoningCapabilityParams(I),this.applyReasoningCapability(I,C,p);try{O=await this.client.chat.completions.create(I,a?{signal:a}:{}),C!==T&&this.persistReasoningCapabilityOverride(C,d)}catch(D){if($=D,y&&this.shouldFallbackForSpecificReasoningError(D,"stream_options","include_usage")){y=!1;continue}if(!this.shouldFallbackForReasoningError(D,...this.getFallbackTerms(C)))throw D;break}}if(O)break}if(!O)throw $??new R("All reasoning capability attempts failed without a captured error",this.name);let E=Date.now(),j=0,A=0,_e=3e4;for await(let C of O){if(a?.aborted)throw new DOMException("Request aborted","AbortError");let I=Date.now(),D=I-E;D>_e&&(j++,A+=D,this.logStreamDiagnostic(`[Stream] stall detected: ${Math.round(D/1e3)}s gap`,{stallCount:j,totalStallMs:A})),E=I,s?.onHeartbeat?.(),u=Bt(C.usage)??u;let k=C.choices[0],P=k?.delta;if(k?.finish_reason&&(g=k.finish_reason,process.env.KODAX_DEBUG_STREAM)){let h=Date.now()-b;this.logStreamDiagnostic(`[Stream] finish_reason: ${g} after ${h}ms`)}P?.content&&(M+=P.content,s?.onTextDelta?.(P.content));let S=this.extractReasoningDelta(P);if(S&&(x+=S,s?.onThinkingDelta?.(S)),P?.tool_calls)for(let h of P.tool_calls){let Y=_.get(h.index)??{id:"",name:"",arguments:""};h.id&&(Y.id=h.id),h.function?.name&&(Y.name=h.function.name),h.function?.arguments&&(Y.arguments+=h.function.arguments,s?.onToolInputDelta?.(Y.name,h.function.arguments,Y.id?{toolId:Y.id}:void 0)),_.set(h.index,Y)}}if(!g){let C=Date.now()-b;if(a?.aborted){let D=a.reason instanceof Error?a.reason.message:typeof a.reason=="string"?a.reason:"Request aborted";throw this.logStreamDiagnostic("[Stream] Stream ended after abort before finish_reason:",{duration:C,reason:D,textContentLength:M.length,toolCallsCount:_.size}),new DOMException(D,"AbortError")}let I=new Error(`Stream incomplete: finish_reason not received. Duration: ${C}ms. This may indicate a network disconnection or API timeout.`);throw I.name="StreamIncompleteError",this.logStreamDiagnostic("[Stream] Incomplete stream detected:",{duration:C,textContentLength:M.length,toolCallsCount:_.size}),I}let le=M?[{type:"text",text:M}]:[],ue=[],pe=[];x&&(pe.push({type:"thinking",thinking:x}),s?.onThinkingEnd?.(x));for(let[,C]of _)C.id&&C.name&&ue.push({type:"tool_use",id:C.id,name:C.name,input:te(C.arguments)});return{textBlocks:le,toolBlocks:ue,thinkingBlocks:pe,usage:u,stopReason:g??void 0}},a,3,s?.onRateLimit,s?.onRetryAfter)}supportsNonStreamingFallback(){return!0}async complete(e,n,o,r,s,a){return this.withRateLimit(async()=>{let l=this.stripCacheBoundariesFromMessages(e),{system:c,rest:m}=this.normalizeSystemForWire(o,l),v=[{role:"system",content:c},...await this.convertMessages(m)],f=n.map(E=>({type:"function",function:{name:E.name,description:E.description,parameters:E.input_schema}})),_=this.normalizeReasoning(r),M=s?.modelOverride??this.config.model,x=Q(_)?this.getReasoningCapability(M):"none",u=Q(_)?this.getReasoningFallbackChain(x).filter(E=>E==="native-budget"||E==="native-effort"||E==="native-toggle"||E==="none"):["none"],y={model:M,messages:v,tools:f,max_completion_tokens:this.getEffectiveMaxOutputTokens(M)},g,b;for(let E of u){let j={...y};this.resetReasoningCapabilityParams(j),this.applyReasoningCapability(j,E,_);try{g=await this.client.chat.completions.create(j,a?{signal:a}:{}),E!==x&&this.persistReasoningCapabilityOverride(E,M);break}catch(A){if(b=A,!this.shouldFallbackForReasoningError(A,...this.getFallbackTerms(E)))throw A}}if(!g)throw b??new R("All reasoning capability attempts failed without a captured error",this.name);let p=g.choices[0],d=p?.message,T=Sn(d?.content),w=On(d),W=(d?.tool_calls??[]).filter(An).map(E=>({type:"tool_use",id:E.id,name:E.function.name,input:te(E.function.arguments)}));T&&s?.onTextDelta?.(T);let O=T?[{type:"text",text:T}]:[],$=[];return w&&($.push({type:"thinking",thinking:w}),s?.onThinkingDelta?.(w),s?.onThinkingEnd?.(w)),{textBlocks:O,toolBlocks:W,thinkingBlocks:$,usage:Bt(g.usage),stopReason:p?.finish_reason??void 0}},a,3,s?.onRateLimit,s?.onRetryAfter)}extractReasoningDelta(e){let n=e?.reasoning_content;return typeof n=="string"?n:Array.isArray(n)?n.map(o=>typeof o=="string"?o:typeof o=="object"&&o!==null&&"text"in o&&typeof o.text=="string"?o.text:"").join(""):""}serializeAssistantMessage(e){let n=e.filter(c=>c.type==="text").map(c=>c.text).join(`
+`),o=e.filter(c=>c.type==="tool_use").map(c=>({id:c.id,type:"function",function:{name:c.name,arguments:JSON.stringify(c.input??{})}})),r=e.filter(c=>c.type==="thinking"||c.type==="redacted_thinking").map(c=>c.type==="thinking"?c.thinking:"").filter(Boolean).join(`
+
+`),s=e.some(c=>c.type==="thinking"||c.type==="redacted_thinking");if(!n&&o.length===0&&!s)return[];let a;n?a=n:o.length>0?a=null:a="...";let l={role:"assistant",content:a};return o.length>0&&(l.tool_calls=o),this.config.replayReasoningContent&&(l.reasoning_content=r||""),[l]}async serializeUserMessage(e){let n=[],o=e.filter(a=>a.type==="text").map(a=>a.text).join(`
+`),r=e.filter(a=>a.type==="image");for(let a of e)a.type==="tool_result"&&n.push({role:"tool",tool_call_id:a.tool_use_id,content:a.content});if(r.length===0)return o&&n.push({role:"user",content:o}),n;let s=[];o&&s.push({type:"text",text:o});for(let a of r)s.push({type:"image_url",image_url:{url:await Nt(a.path,a.mediaType)}});return n.push({role:"user",content:s}),n}serializeSystemMessage(e){if(typeof e=="string")return[{role:"system",content:e}];let n=e.filter(o=>o.type==="text").map(o=>o.text).join(`
+`);return n?[{role:"system",content:n}]:[]}async convertMessages(e){let n=[];for(let o of e){if(o.role==="system"){n.push(...this.serializeSystemMessage(o.content));continue}if(typeof o.content=="string"){n.push({role:o.role,content:o.content});continue}if(o.role==="assistant"){n.push(...this.serializeAssistantMessage(o.content));continue}n.push(...await this.serializeUserMessage(o.content))}return n}};import{spawn as In}from"node:child_process";import{Readable as Dn,Writable as Kn}from"node:stream";import ot from"node:process";import{ClientSideConnection as Ln,PROTOCOL_VERSION as Nn,ndJsonStream as Bn}from"@agentclientprotocol/sdk";var Te=class{static{i(this,"AcpClient")}client=null;agentProcess=null;options;constructor(e){this.options=e}async connect(){let e,n;if(this.options.inputStream&&this.options.outputStream)e=this.options.inputStream,n=this.options.outputStream;else if(this.options.command){let s=ot.platform==="win32"&&!this.options.command.endsWith(".cmd")?`${this.options.command}.cmd`:this.options.command;if(this.agentProcess=In(s,this.options.args??[],{cwd:this.options.cwd??ot.cwd(),stdio:["pipe","pipe","inherit"]}),!this.agentProcess.stdin||!this.agentProcess.stdout)throw new Error("Failed to create ACP stdio pipes");n=Kn.toWeb(this.agentProcess.stdin),e=Dn.toWeb(this.agentProcess.stdout)}else throw new Error("AcpClient requires either a command or I/O streams");let o=Bn(n,e);this.client=new Ln(()=>({sessionUpdate:i(async r=>{this.options.onSessionUpdate?.(r)},"sessionUpdate"),requestPermission:i(async r=>{let s=r.options??[],a=s.find(l=>l.kind==="allow_once"||l.kind==="allow_always")??s[0];return a?{outcome:{outcome:"selected",optionId:a.optionId}}:{outcome:{outcome:"cancelled"}}},"requestPermission")}),o),await this.client.initialize({protocolVersion:Nn,clientCapabilities:{},clientInfo:{name:"kodax-ai-acp-client",version:"1.0.0"}})}async createNewSession(){if(!this.client)throw new Error("Client not connected");return(await this.client.newSession({cwd:this.options.cwd??ot.cwd(),mcpServers:[]})).sessionId}async prompt(e,n,o,r){if(!this.client)throw new Error("Client not connected");let s={sessionId:n,prompt:[{type:"text",text:e}]};r?.model&&(s.model=r.model);let a=this.client.prompt(s);if(o){let l=i(()=>{this.client?.cancel({sessionId:n}).catch(()=>{})},"onAbort");o.addEventListener("abort",l),a=a.finally(()=>{o.removeEventListener("abort",l)})}return await a}disconnect(){this.agentProcess?.kill(),this.options.abort?.();try{this.client?.close?.()}catch{}this.client=null}};function Fn(t){if(!t||typeof t!="object")return;let e=t,n=typeof e.inputTokens=="number"?e.inputTokens:0,o=typeof e.outputTokens=="number"?e.outputTokens:0,r=typeof e.totalTokens=="number"?e.totalTokens:n+o;if(![n,o,r].some(s=>!Number.isFinite(s)||s<0)&&!(r<n||r<o))return{inputTokens:n,outputTokens:o,totalTokens:r,cachedReadTokens:typeof e.cachedReadTokens=="number"?e.cachedReadTokens:void 0,cachedWriteTokens:typeof e.cachedWriteTokens=="number"?e.cachedWriteTokens:void 0,thoughtTokens:typeof e.thoughtTokens=="number"?e.thoughtTokens:void 0}}i(Fn,"normalizeAcpUsage");var oe=class extends G{static{i(this,"KodaXAcpProvider")}_client=null;_sessionMap=new Map;_activeStreams=new Map;isConfigured(){return!0}getCapabilityProfile(){return F(fe)}serializeImageBlockToPromptToken(e){return null}stripCacheBoundariesFromMessages(e){return e.map(n=>{if(typeof n.content=="string")return n;let o=ge(n.content);return o.length===n.content.length?n:{...n,content:o}})}async stream(e,n,o,r,s,a){if(e=this.stripCacheBoundariesFromMessages(e),this.acpClientOptions.executor&&typeof this.acpClientOptions.executor.isInstalled=="function"&&!await this.acpClientOptions.executor.isInstalled())throw new Error(`${this.name} requires a local CLI environment, but the CLI was not found or is not configured correctly.`);let l=[],c=[],m=e[e.length-1],v="";if(m&&typeof m.content=="string")v=m.content;else if(m&&Array.isArray(m.content)){let y=[];for(let g of m.content)if(g.type==="text")y.push(g.text);else if(g.type==="image"){let b=this.serializeImageBlockToPromptToken(g);b&&y.push(b)}v=y.join(`
+`)}let f={...this.acpClientOptions,onSessionUpdate:i(y=>{let g=y.update,b=y.sessionId;if(!("sessionUpdate"in g))return;let p=b?this._activeStreams.get(b):void 0;if(p)switch(g.sessionUpdate){case"agent_message_chunk":if(g.content?.type==="text"){let d=g.content.text;p.output.text+=d,p.streamOptions?.onTextDelta?.(d)}break;case"tool_call":{let d="{}",T=g.arguments??g.parameters;T&&(d=typeof T=="string"?T:JSON.stringify(T)),p.streamOptions?.onToolInputDelta?.(g.title,d);let w=`
+> [Tool Use] ${g.title}: ${d}
+`;p.output.text+=w,p.streamOptions?.onTextDelta?.(w);break}case"tool_call_update":if(g.status){let d=`> [Tool Result] ${g.status}
+
+`;p.output.text+=d,p.streamOptions?.onTextDelta?.(d)}break}},"onSessionUpdate")},_=s?.sessionId??"default";this._client||(this._client=new Te(f),await this._client.connect());let M=this._sessionMap.get(_);M||(M=await this._client.createNewSession(),this._sessionMap.set(_,M));let x={text:""};this._activeStreams.set(M,{streamOptions:s,output:x});let u;try{u=await this._client.prompt(v,M,a,{model:s?.modelOverride})}catch(y){if(!(y instanceof Error&&y.name==="AbortError"))throw y}finally{this._activeStreams.delete(M)}return x.text&&l.push({type:"text",text:x.text}),{textBlocks:l,toolBlocks:c,thinkingBlocks:[],usage:Fn(u?.usage)}}disconnect(){this._client&&(this._client.disconnect(),this._client=null),this._activeStreams.clear(),this._sessionMap.clear()}};import{spawn as Un}from"node:child_process";import it from"node:process";var ie=class{static{i(this,"CLIExecutor")}config;_installedCache=null;constructor(e){this.config=e}async isInstalled(){return this._installedCache!==null?this._installedCache:(this._installedCache=await this.checkInstalled(),this._installedCache)}async*execute(e){let n=this.buildArgs(e),o={...it.env,...this.config.env},s=it.platform==="win32"&&!this.config.command.endsWith(".cmd")?`${this.config.command}.cmd`:this.config.command,a=Un(s,n,{cwd:this.config.cwd??it.cwd(),env:o,stdio:["ignore","pipe","pipe"]}),l="";a.stderr?.on("data",v=>{l+=v.toString()});let c=!1,m=i(()=>{c||a.kill("SIGTERM")},"abortHandler");e.signal?.addEventListener("abort",m),a.on("exit",()=>{c=!0});try{yield*this.parseOutputStream(a.stdout,e.signal),l.trim()&&console.error(`[CLIExecutor] stderr: ${l.trim()}`)}finally{e.signal?.removeEventListener("abort",m),c||a.kill()}}async*parseOutputStream(e,n){let o="";for await(let r of e){if(n?.aborted)break;o+=r.toString();let s=o.split(`
+`);o=s.pop()??"";for(let a of s){if(!a.trim())continue;let l=this.parseLine(a.trim());l&&(yield l)}}if(o.trim()&&!n?.aborted){let r=this.parseLine(o.trim());r&&(yield r)}}};import{spawn as Xn}from"node:child_process";import Wn from"node:process";async function Ee(t){try{let e=Wn.platform==="win32",n=Xn(e?`${t}.cmd`:t,["--version"]);return await new Promise(o=>{n.on("close",r=>o(r===0)),n.on("error",()=>o(!1))})}catch{return!1}}i(Ee,"checkCliCommandInstalled");var Pe=class extends ie{static{i(this,"GeminiCLIExecutor")}model;constructor(e){super({command:"gemini",baseArgs:["--output-format","stream-json","--approval-mode","yolo"],timeout:3e5,...e}),this.model=e?.model??"gemini-2.5-pro"}async checkInstalled(){return Ee("gemini")}buildArgs(e){let n=["-m",e.model??this.model];return e.sessionId?(n.push("-r",e.sessionId),n.push(e.prompt)):n.push("-p",e.prompt),[...n,...this.config.baseArgs]}parseLine(e){if(!e.startsWith("{"))return null;try{let n=JSON.parse(e);return this.convertEvent(n)}catch{return null}}convertEvent(e){let n=e.timestamp?Date.parse(e.timestamp):Date.now();switch(e.type){case"init":return{type:"session_start",timestamp:n,sessionId:e.session_id??"",model:e.model??this.model,raw:e};case"message":return{type:"message",timestamp:n,role:e.role,content:e.content??"",delta:e.delta,raw:e};case"tool_use":return{type:"tool_use",timestamp:n,toolId:e.tool_id??"",toolName:e.tool_name??"",parameters:e.parameters??{},raw:e};case"tool_result":return{type:"tool_result",timestamp:n,toolId:e.tool_id??"",status:e.status==="success"?"success":"error",output:e.output??"",raw:e};case"error":return{type:"error",timestamp:n,errorType:"error",message:e.message??"Unknown error",raw:e};case"result":return{type:"complete",timestamp:n,status:e.status==="success"?"success":"failed",usage:e.stats?{inputTokens:e.stats.input_tokens??0,outputTokens:e.stats.output_tokens??0,totalTokens:e.stats.total_tokens??0}:void 0,raw:e};default:return null}}};import{TransformStream as Ft}from"node:stream/web";import{randomUUID as rt}from"node:crypto";function we(t){let e=new Ft,n=new Ft,o=e.readable.getReader(),r=n.writable.getWriter(),s=new TextDecoder,a=new TextEncoder,l="",c=rt(),m=new AbortController,v=new Map;(async()=>{try{for(;;){let{done:u,value:y}=await o.read();if(u)break;l+=s.decode(y,{stream:!0});let g=l.split(`
+`);l=g.pop()??"";for(let b of g)if(b.trim())try{let p=JSON.parse(b);_(p).catch(d=>{console.error("[PseudoAcpServer] Failed to handle request:",d)})}catch{console.error("[PseudoAcpServer] Failed to parse message:",b)}}}catch(u){console.error("[PseudoAcpServer] Stream read error:",u)}})();let f=i(async u=>{let y=JSON.stringify(u)+`
+`;await r.write(a.encode(y))},"sendMsg");async function _(u){if(u.method==="initialize")await f({jsonrpc:"2.0",id:u.id,result:{protocolVersion:u.params.protocolVersion,serverInfo:{name:"pseudo-acp-server",version:"1.0.0"},serverCapabilities:{}}});else if(u.method==="session/new"||u.method==="sessions/new")c=u.params?.sessionId??rt(),await f({jsonrpc:"2.0",id:u.id,result:{sessionId:c}});else if(u.method==="session/prompt"||u.method==="chat/prompt"){let y=new AbortController,g=u.params.sessionId;v.set(g,y);let b=i(()=>y.abort(),"onGlobalAbort");m.signal.addEventListener("abort",b);let d=await M(g,u.params.prompt,typeof u.params.model=="string"?u.params.model:void 0,y.signal).finally(()=>{v.delete(g),m.signal.removeEventListener("abort",b)});await f({jsonrpc:"2.0",id:u.id,result:d.usage?{stopReason:d.stopReason,usage:d.usage}:{stopReason:d.stopReason}})}else if(u.method==="session/cancel"||u.method==="chat/cancel"){let y=v.get(u.params.sessionId);y&&y.abort(),u.id!==void 0&&await f({jsonrpc:"2.0",id:u.id,result:{}})}else u.id!==void 0&&await f({jsonrpc:"2.0",id:u.id,result:{}})}i(_,"handleRequest");let M=i(async(u,y,g,b)=>{let p=y.find(d=>d.type==="text")?.text??"";try{let d=t.execute({prompt:p,model:g,sessionId:u==="default"?void 0:u,signal:b});for await(let T of d){let w=x(T);if(w&&await f({jsonrpc:"2.0",method:"session/update",params:{sessionId:u,update:w}}),T.type==="complete")return{stopReason:b.aborted?"cancelled":"end_turn",usage:T.usage}}return{stopReason:b.aborted?"cancelled":"end_turn"}}catch(d){return b.aborted||d instanceof Error&&d.name==="AbortError"?{stopReason:"cancelled"}:(console.error("[PseudoAcpServer] Error executing prompt:",d),await f({jsonrpc:"2.0",method:"session/update",params:{sessionId:u,update:{sessionUpdate:"agent_message_chunk",content:{type:"text",text:`
+[Fatal Error: ${d}]
+`}}}}),{stopReason:"end_turn"})}},"executePrompt"),x=i(u=>{switch(u.type){case"message":if(u.role==="assistant"&&u.content)return{sessionUpdate:"agent_message_chunk",content:{type:"text",text:u.content}};break;case"tool_use":return{sessionUpdate:"tool_call",title:u.toolName,arguments:u.parameters,status:"running",toolCallId:u.toolId||rt()};case"tool_result":return{sessionUpdate:"tool_call_update",toolCallId:u.toolId,status:u.status};case"error":return{sessionUpdate:"agent_message_chunk",content:{type:"text",text:`
+[Error: ${u.message}]
+`}};case"complete":break}return null},"mapToAcpNotification");return{inputStream:n.readable,outputStream:e.writable,abort:i(()=>{m.abort(),e.readable.cancel().catch(()=>{}),n.writable.abort().catch(()=>{})},"abort"),executor:t}}i(we,"createPseudoAcpServer");import Ut from"node:fs";import Xt from"node:os";import Wt from"node:path";var $n="gpt-5.4",jn="auto-gemini-3",Gn=["gpt-5.4","gpt-5.3-codex","gpt-5.3-codex-spark"],zn=["auto-gemini-3","gemini-3.1-pro-preview","gemini-3-flash-preview","gemini-2.5-pro","gemini-2.5-flash"];function $t(t){let e=new Set,n=[];for(let o of t){let r=o.trim();if(!r)continue;let s=r.toLowerCase();e.has(s)||(e.add(s),n.push(r))}return n}i($t,"dedupePreserveOrder");function Hn(){let t=Wt.join(Xt.homedir(),".codex","config.toml");try{return Ut.readFileSync(t,"utf8").match(/^\s*model\s*=\s*"([^"]+)"/m)?.[1]?.trim()||null}catch{return null}}i(Hn,"readCodexConfiguredModel");function Vn(){let t=Wt.join(Xt.homedir(),".gemini","settings.json");try{let e=JSON.parse(Ut.readFileSync(t,"utf8")),n=typeof e.model=="string"?e.model.trim():"";return n||(typeof e.general?.model=="string"?e.general.model.trim():"")||null}catch{return null}}i(Vn,"readGeminiConfiguredModel");function re(){return Hn()||$n}i(re,"getCodexCliDefaultModel");function se(){return Vn()||jn}i(se,"getGeminiCliDefaultModel");function he(){return $t([re(),...Gn])}i(he,"getCodexCliKnownModels");function ye(){return $t([se(),...zn])}i(ye,"getGeminiCliKnownModels");var st=se(),Yn=ye(),Jn={transport:"cli-bridge",conversationSemantics:"last-user-message",mcpSupport:"none",contextFidelity:"lossy",toolCallingFidelity:"limited",sessionSupport:"stateless",longRunningSupport:"limited",multimodalSupport:"image-input",evidenceSupport:"limited"},Re=class extends oe{static{i(this,"KodaXGeminiCliProvider")}name="gemini-cli";supportsThinking=!1;config={apiKeyEnv:"GEMINI_CLI_API_KEY",model:st,models:Yn.filter(e=>e!==st).map(e=>({id:e,displayName:e})),supportsThinking:!1,reasoningCapability:"prompt-only",contextWindow:1048576};acpClientOptions;constructor(){super();let e=new Pe({model:st});this.acpClientOptions=we(e)}getCapabilityProfile(){return F(Jn)}serializeImageBlockToPromptToken(e){return/\s/.test(e.path)?(console.warn(`[gemini-cli] Image path contains whitespace and cannot be safely passed via @<path> syntax \u2014 dropping image block. Path: ${JSON.stringify(e.path)}`),null):`@${e.path}`}};var Ae=class extends ie{static{i(this,"CodexCLIExecutor")}model;constructor(e){super({command:"codex",baseArgs:["exec","--json","--full-auto"],timeout:3e5,...e}),this.model=e?.model??"gpt-5.4"}async checkInstalled(){return Ee("codex")}buildArgs(e){this.model=e.model??this.model;let n=e.model?["-m",e.model]:[];return e.sessionId?["exec","resume",e.sessionId,...n,e.prompt,...this.config.baseArgs.filter(o=>o!=="exec")]:[...this.config.baseArgs,...n,e.prompt]}parseLine(e){if(!e.startsWith("{"))return null;try{let n=JSON.parse(e);return this.convertEvent(n)}catch{return null}}convertEvent(e){let n=Date.now();switch(e.type){case"thread.started":return{type:"session_start",timestamp:n,sessionId:e.thread_id??"",model:this.model,raw:e};case"item.completed":return e.item?.type==="agent_message"?{type:"message",timestamp:n,role:"assistant",content:e.item.text??"",raw:e}:e.item?.type==="command_execution"?{type:"tool_use",timestamp:n,toolId:e.item.id,toolName:"Bash",parameters:{command:e.item.command},raw:e}:null;case"turn.completed":return{type:"complete",timestamp:n,status:"success",usage:e.usage?{inputTokens:e.usage.input_tokens,outputTokens:e.usage.output_tokens,totalTokens:e.usage.input_tokens+e.usage.output_tokens}:void 0,raw:e};case"error":case"turn.failed":return{type:"error",timestamp:n,errorType:e.type,message:e.message??"Unknown error",raw:e};default:return null}}};var at=re(),Qn=he(),Se=class extends oe{static{i(this,"KodaXCodexCliProvider")}name="codex-cli";supportsThinking=!1;config={apiKeyEnv:"CODEX_CLI_API_KEY",model:at,models:Qn.filter(e=>e!==at).map(e=>({id:e,displayName:e})),supportsThinking:!1,reasoningCapability:"prompt-only",contextWindow:128e3};acpClientOptions;constructor(){super();let e=new Ae({model:at});this.acpClientOptions=we(e)}};import qn from"@anthropic-ai/sdk";var jt=se(),Zn=ye(),Gt=re(),eo=he(),z={anthropic:{apiKeyEnv:"ANTHROPIC_API_KEY",model:"claude-sonnet-4-6",models:["claude-opus-4-6","claude-haiku-4-5"],reasoningCapability:"native-budget",capabilityProfile:L},openai:{apiKeyEnv:"OPENAI_API_KEY",model:"gpt-5.3-codex",models:["gpt-5.4","gpt-5.3-codex-spark"],reasoningCapability:"native-effort",capabilityProfile:L},deepseek:{apiKeyEnv:"DEEPSEEK_API_KEY",model:"deepseek-v4-flash",models:["deepseek-v4-pro"],reasoningCapability:"native-effort",capabilityProfile:L},kimi:{apiKeyEnv:"KIMI_API_KEY",model:"kimi-k2.6",models:["k2.5"],reasoningCapability:"native-effort",capabilityProfile:L},"kimi-code":{apiKeyEnv:"KIMI_API_KEY",model:"kimi-for-coding",reasoningCapability:"native-budget",capabilityProfile:L},qwen:{apiKeyEnv:"QWEN_API_KEY",model:"qwen3.5-plus",reasoningCapability:"native-budget",capabilityProfile:L},zhipu:{apiKeyEnv:"ZHIPU_API_KEY",model:"glm-5",models:["glm-5.1","glm-5-turbo"],reasoningCapability:"native-budget",capabilityProfile:L},"zhipu-coding":{apiKeyEnv:"ZHIPU_API_KEY",model:"glm-5",models:["glm-5.1","glm-5-turbo"],reasoningCapability:"native-budget",capabilityProfile:L},"minimax-coding":{apiKeyEnv:"MINIMAX_API_KEY",model:"MiniMax-M2.7",models:["MiniMax-M2.7-highspeed","MiniMax-M2.5","MiniMax-M2.5-highspeed","MiniMax-M2.1","MiniMax-M2.1-highspeed","MiniMax-M2"],reasoningCapability:"native-budget",capabilityProfile:L},"mimo-coding":{apiKeyEnv:"MIMO_API_KEY",model:"mimo-v2.5-pro",models:["mimo-v2.5"],reasoningCapability:"native-budget",capabilityProfile:L},"ark-coding":{apiKeyEnv:"ARK_API_KEY",model:"glm-5.1",models:["glm-4.7","kimi-k2.6","kimi-k2.5","minimax-latest","deepseek-v3.2","doubao-seed-2.0-code","doubao-seed-2.0-pro","doubao-seed-2.0-lite"],reasoningCapability:"native-budget",capabilityProfile:L},"gemini-cli":{apiKeyEnv:"GEMINI_API_KEY",model:jt,models:Zn.filter(t=>t!==jt),reasoningCapability:"prompt-only",capabilityProfile:Lt},"codex-cli":{apiKeyEnv:"OPENAI_API_KEY",model:Gt,models:eo.filter(t=>t!==Gt),reasoningCapability:"prompt-only",capabilityProfile:fe}};function X(t,e){let n=z[t];return{apiKeyEnv:n.apiKeyEnv,model:n.model,reasoningCapability:n.reasoningCapability,...e}}i(X,"buildProviderConfig");var ct=class extends K{static{i(this,"AnthropicProvider")}name="anthropic";config=X("anthropic",{models:[{id:"claude-opus-4-6",displayName:"Opus 4.6",thinkingBudgetCap:28e3},{id:"claude-haiku-4-5",displayName:"Haiku 4.5",thinkingBudgetCap:1e4}],supportsThinking:!0,contextWindow:2e5,maxOutputTokens:64e3,thinkingBudgetCap:28e3,strictThinkingSignature:!0});constructor(){super(),this.client=new qn({apiKey:this.getApiKey()})}},lt=class extends K{static{i(this,"ZhipuCodingProvider")}name="zhipu-coding";config=X("zhipu-coding",{baseUrl:"https://open.bigmodel.cn/api/anthropic",models:[{id:"glm-5.1",displayName:"GLM-5.1"},{id:"glm-5-turbo",displayName:"GLM-5 Turbo",contextWindow:128e3}],supportsThinking:!0,contextWindow:2e5,maxOutputTokens:16e3,thinkingBudgetCap:16e3,streamMaxDurationMs:3e5});constructor(){super(),this.initClient()}},ut=class extends K{static{i(this,"KimiCodeProvider")}name="kimi-code";config=X("kimi-code",{baseUrl:"https://api.kimi.com/coding/",supportsThinking:!0,contextWindow:256e3,maxOutputTokens:32e3});constructor(){super(),this.initClient()}},pt=class extends K{static{i(this,"MiniMaxCodingProvider")}name="minimax-coding";config=X("minimax-coding",{baseUrl:"https://api.minimaxi.com/anthropic",models:[{id:"MiniMax-M2.7",displayName:"MiniMax M2.7"},{id:"MiniMax-M2.7-highspeed",displayName:"MiniMax M2.7 Highspeed (higher-tier plan)"},{id:"MiniMax-M2.5",displayName:"MiniMax M2.5"},{id:"MiniMax-M2.5-highspeed",displayName:"MiniMax M2.5 Highspeed (higher-tier plan)"},{id:"MiniMax-M2.1",displayName:"MiniMax M2.1"},{id:"MiniMax-M2.1-highspeed",displayName:"MiniMax M2.1 Highspeed (higher-tier plan)"},{id:"MiniMax-M2",displayName:"MiniMax M2"}],supportsThinking:!0,contextWindow:204800,maxOutputTokens:32e3});constructor(){super(),this.initClient()}},dt=class extends K{static{i(this,"MimoCodingProvider")}name="mimo-coding";config=X("mimo-coding",{baseUrl:"https://token-plan-cn.xiaomimimo.com/anthropic",models:[{id:"mimo-v2.5",displayName:"MiMo V2.5"}],supportsThinking:!0,contextWindow:1e6,maxOutputTokens:32e3,thinkingBudgetCap:16e3});constructor(){super(),this.initClient()}},mt=class extends K{static{i(this,"ArkCodingProvider")}name="ark-coding";config=X("ark-coding",{baseUrl:"https://ark.cn-beijing.volces.com/api/coding",models:[{id:"glm-4.7",displayName:"GLM-4.7"},{id:"kimi-k2.6",displayName:"Kimi K2.6",contextWindow:256e3},{id:"kimi-k2.5",displayName:"Kimi K2.5",contextWindow:256e3},{id:"minimax-latest",displayName:"MiniMax Latest",contextWindow:204800},{id:"deepseek-v3.2",displayName:"DeepSeek V3.2",contextWindow:128e3},{id:"doubao-seed-2.0-code",displayName:"Doubao Seed 2.0 Code",contextWindow:256e3},{id:"doubao-seed-2.0-pro",displayName:"Doubao Seed 2.0 Pro",contextWindow:256e3},{id:"doubao-seed-2.0-lite",displayName:"Doubao Seed 2.0 Lite",contextWindow:256e3}],supportsThinking:!0,contextWindow:2e5,maxOutputTokens:32e3});constructor(){super(),this.initClient()}},ft=class extends N{static{i(this,"OpenAIProvider")}name="openai";config=X("openai",{models:[{id:"gpt-5.4",displayName:"GPT-5.4"},{id:"gpt-5.3-codex-spark",displayName:"GPT-5.3 Codex Spark"}],supportsThinking:!0,contextWindow:4e5,maxOutputTokens:32768});constructor(){super(),this.initClient()}},gt=class extends N{static{i(this,"DeepSeekProvider")}name="deepseek";config=X("deepseek",{baseUrl:"https://api.deepseek.com",models:[{id:"deepseek-v4-pro",displayName:"DeepSeek V4 Pro"}],supportsThinking:!0,contextWindow:1e6,maxOutputTokens:64e3,replayReasoningContent:!0});constructor(){super(),this.initClient()}},ht=class extends N{static{i(this,"KimiProvider")}name="kimi";config=X("kimi",{baseUrl:"https://api.moonshot.cn/v1",models:[{id:"k2.5",displayName:"K2.5"}],supportsThinking:!0,contextWindow:256e3,maxOutputTokens:32768,replayReasoningContent:!0});constructor(){super(),this.initClient()}},yt=class extends N{static{i(this,"QwenProvider")}name="qwen";config=X("qwen",{baseUrl:"https://dashscope.aliyuncs.com/compatible-mode/v1",supportsThinking:!0,contextWindow:256e3,maxOutputTokens:32768,replayReasoningContent:!0});constructor(){super(),this.initClient()}},_t=class extends N{static{i(this,"ZhipuProvider")}name="zhipu";config=X("zhipu",{baseUrl:"https://open.bigmodel.cn/api/paas/v4",models:[{id:"glm-5.1",displayName:"GLM-5.1"},{id:"glm-5-turbo",displayName:"GLM-5 Turbo",contextWindow:128e3}],supportsThinking:!0,contextWindow:2e5,maxOutputTokens:32768,replayReasoningContent:!0});constructor(){super(),this.initClient()}},U={anthropic:i(()=>new ct,"anthropic"),openai:i(()=>new ft,"openai"),deepseek:i(()=>new gt,"deepseek"),kimi:i(()=>new ht,"kimi"),"kimi-code":i(()=>new ut,"kimi-code"),qwen:i(()=>new yt,"qwen"),zhipu:i(()=>new _t,"zhipu"),"zhipu-coding":i(()=>new lt,"zhipu-coding"),"minimax-coding":i(()=>new pt,"minimax-coding"),"mimo-coding":i(()=>new dt,"mimo-coding"),"ark-coding":i(()=>new mt,"ark-coding"),"gemini-cli":i(()=>new Re,"gemini-cli"),"codex-cli":i(()=>new Se,"codex-cli")},kt=process.env.KODAX_PROVIDER??"zhipu-coding",zt=new Map;function to(t){if(B(t))return z[t].apiKeyEnv}i(to,"resolveApiKeyEnvForProvider");function Ht(t){let e=t??kt,n=U[e];if(!n)throw new R(`Unknown provider: ${e}. Available: ${Object.keys(U).join(", ")}`,e);let o=to(e),r=o?process.env[o]:void 0,s=zt.get(e);if(s&&s.apiKey===r)return s.instance;let a=n();return zt.set(e,{apiKey:r,instance:a}),a}i(Ht,"getProvider");function Vt(t){return B(t)?!!process.env[z[t].apiKeyEnv]:!1}i(Vt,"isProviderConfigured");function Yt(t){return B(t)?z[t].model:null}i(Yt,"getProviderModel");function Jt(t,e){if(!B(t))return"unknown";let n=z[t],o=e??n.model;return n.modelReasoningCapabilities?.[o]??n.reasoningCapability}i(Jt,"getProviderConfiguredReasoningCapability");function Qt(t){return B(t)?F(z[t].capabilityProfile):null}i(Qt,"getProviderConfiguredCapabilityProfile");function qt(){let t=[];for(let e of Object.keys(U)){let n=z[e];t.push({name:e,model:n.model,models:n.models?[n.model,...n.models]:[n.model],configured:!!process.env[n.apiKeyEnv],reasoningCapability:n.reasoningCapability,capabilityProfile:F(n.capabilityProfile)})}return t}i(qt,"getProviderList");function Zt(t){let e=z[t];return e?e.models?[e.model,...e.models]:[e.model]:[]}i(Zt,"getProviderModels");function B(t){return t in U}i(B,"isProviderName");var no=new Set(["compat","sdk"]);function vt(t){if(!t.name||!t.baseUrl||!t.apiKeyEnv||!t.model)throw new Error(`Custom provider requires name, baseUrl, apiKeyEnv, and model. Got: ${JSON.stringify({name:t.name,baseUrl:t.baseUrl,apiKeyEnv:t.apiKeyEnv,model:t.model})}`);if(t.protocol!=="anthropic"&&t.protocol!=="openai")throw new Error(`Unknown protocol "${t.protocol}" for custom provider "${t.name}". Must be "anthropic" or "openai".`);if(t.userAgentMode!==void 0&&!no.has(t.userAgentMode))throw new Error(`Unknown userAgentMode "${t.userAgentMode}" for custom provider "${t.name}". Must be "compat" or "sdk".`)}i(vt,"validateCustomProviderConfig");function oo(t){let e=t.models?.length?t.models.map(n=>typeof n=="string"?{id:n}:n):void 0;return{apiKeyEnv:t.apiKeyEnv,model:t.model,baseUrl:t.baseUrl,models:e,userAgentMode:t.userAgentMode,supportsThinking:t.supportsThinking??!1,reasoningCapability:t.reasoningCapability??"none",capabilityProfile:t.capabilityProfile,contextWindow:t.contextWindow,maxOutputTokens:t.maxOutputTokens,thinkingBudgetCap:t.thinkingBudgetCap}}i(oo,"buildProviderConfig");function Oe(t){vt(t);let e=oo(t);return t.protocol==="anthropic"?new xt(t.name,e):new bt(t.name,e)}i(Oe,"createCustomProvider");var xt=class extends K{static{i(this,"DynamicAnthropicProvider")}name;config;constructor(e,n){super(),this.name=e,this.config=n,this.initClient()}},bt=class extends N{static{i(this,"DynamicOpenAIProvider")}name;config;constructor(e,n){super(),this.name=e,this.config=n,this.initClient()}};var ae=new Map,Mt=new Map;function en(t){let e=new Set,n=new Map,o=new Map;for(let r of t){if(vt(r),e.has(r.name))throw new Error(`Duplicate custom provider name: "${r.name}". Each custom provider must have a unique name.`);r.name in U&&console.warn(`[kodax] Custom provider "${r.name}" shadows a built-in provider. The built-in provider will be used. Choose a different name to use your custom provider.`),e.add(r.name),n.set(r.name,r),o.set(r.name,()=>Oe(r))}ae.clear(),Mt.clear();for(let[r,s]of n)ae.set(r,s);for(let[r,s]of o)Mt.set(r,s)}i(en,"registerCustomProviders");function Ie(t){let e=Mt.get(t);return e?e():void 0}i(Ie,"getCustomProvider");function ce(t){return ae.has(t)}i(ce,"isCustomProviderName");function De(){return[...ae.keys()]}i(De,"getCustomProviderNames");function tn(){let t=[];for(let[e,n]of ae){let o=!!process.env[n.apiKeyEnv],r=(n.models??[]).map(a=>typeof a=="string"?a:a.id),s=n.model&&r.length?[...new Set([n.model,...r])]:[n.model];t.push({name:e,model:n.model,models:s,configured:o,reasoningCapability:n.reasoningCapability??"none",capabilityProfile:F(n.capabilityProfile??me),custom:!0})}return t}i(tn,"getCustomProviderList");function nn(t){let e=ae.get(t);if(!e)return;let n=(e.models??[]).map(o=>typeof o=="string"?o:o.id);return e.model&&n.length?[...new Set([e.model,...n])]:[e.model]}i(nn,"getCustomProviderModels");var V=new Map,io=0;function Ct(t){let e=V.get(t);if(!(!e||e.length===0))return e[e.length-1]}i(Ct,"getActiveRuntimeProviderRegistration");function ro(t){for(let[e,n]of V){let o=n.filter(r=>r.id!==t);if(o.length!==n.length){o.length===0?V.delete(e):V.set(e,o);return}}}i(ro,"removeRuntimeProviderRegistration");function on(t,e){let n=t.trim();if(!n)throw new Error("Model provider name cannot be empty.");if(B(n))throw new Error(`Runtime model provider "${n}" conflicts with a built-in provider.`);if(ce(n))throw new Error(`Runtime model provider "${n}" conflicts with an existing config-defined custom provider.`);let o={id:`runtime-provider:${++io}`,name:n,factory:e},r=V.get(n)??[];return V.set(n,[...r,o]),()=>{ro(o.id)}}i(on,"registerModelProvider");function Ke(t){let e=Ct(t);return e?e.factory():void 0}i(Ke,"getRuntimeModelProvider");function Le(t){return Ct(t)!==void 0}i(Le,"isRuntimeModelProviderName");function Ne(){return Array.from(V.keys()).filter(t=>Ct(t)!==void 0)}i(Ne,"getRuntimeModelProviderNames");function rn(){V.clear()}i(rn,"clearRuntimeModelProviders");function sn(t){if(B(t))return U[t]();let e=Ke(t);if(e)return e;let n=Ie(t);if(n)return n;let o=Tt();throw new Error(`Unknown provider: ${t}. Available: ${o.join(", ")}`)}i(sn,"resolveProvider");function an(t){return B(t)||Le(t)||ce(t)}i(an,"isKnownProvider");function Tt(){let t=Object.keys(U),e=Ne(),n=De();return[...new Set([...t,...e,...n])]}i(Tt,"getAvailableProviderNames");var cn={anthropic:{"claude-opus-4-6":{inputPer1M:15,outputPer1M:75,cachePer1M:1.875},"claude-sonnet-4-6":{inputPer1M:3,outputPer1M:15,cachePer1M:.375},"claude-haiku-4-5":{inputPer1M:.8,outputPer1M:4,cachePer1M:.08}},openai:{"gpt-5.4":{inputPer1M:30,outputPer1M:120},"gpt-5.3-codex-spark":{inputPer1M:10,outputPer1M:40}},deepseek:{"deepseek-v4-flash":{inputPer1M:.14,outputPer1M:.28,cachePer1M:.028},"deepseek-v4-pro":{inputPer1M:1.68,outputPer1M:3.36,cachePer1M:.14}},kimi:{"k2.5":{inputPer1M:.005,outputPer1M:.015},"kimi-k2.6":{inputPer1M:.005,outputPer1M:.015}},"kimi-code":{"kimi-for-coding":{inputPer1M:.005,outputPer1M:.015}},qwen:{"qwen3.5-plus":{inputPer1M:.003,outputPer1M:.006}},zhipu:{"glm-5":{inputPer1M:.05,outputPer1M:.1},"glm-5.1":{inputPer1M:.05,outputPer1M:.1},"glm-5-turbo":{inputPer1M:.01,outputPer1M:.03}},"zhipu-coding":{"glm-5":{inputPer1M:.05,outputPer1M:.1},"glm-5.1":{inputPer1M:.05,outputPer1M:.1},"glm-5-turbo":{inputPer1M:.01,outputPer1M:.03}},"minimax-coding":{"MiniMax-M2.7":{inputPer1M:.01,outputPer1M:.03},"MiniMax-M2.7-highspeed":{inputPer1M:.01,outputPer1M:.03},"MiniMax-M2.5":{inputPer1M:.01,outputPer1M:.03},"MiniMax-M2.5-highspeed":{inputPer1M:.01,outputPer1M:.03},"MiniMax-M2.1":{inputPer1M:.01,outputPer1M:.03},"MiniMax-M2.1-highspeed":{inputPer1M:.01,outputPer1M:.03},"MiniMax-M2":{inputPer1M:.01,outputPer1M:.03}},"mimo-coding":{"mimo-v2.5-pro":{inputPer1M:.01,outputPer1M:.03},"mimo-v2.5":{inputPer1M:.01,outputPer1M:.03}},"ark-coding":{"glm-5.1":{inputPer1M:.005,outputPer1M:.015},"glm-4.7":{inputPer1M:.005,outputPer1M:.015},"kimi-k2.6":{inputPer1M:.005,outputPer1M:.015},"kimi-k2.5":{inputPer1M:.005,outputPer1M:.015},"minimax-latest":{inputPer1M:.005,outputPer1M:.015},"deepseek-v3.2":{inputPer1M:.005,outputPer1M:.015},"doubao-seed-2.0-code":{inputPer1M:.005,outputPer1M:.015},"doubao-seed-2.0-pro":{inputPer1M:.005,outputPer1M:.015},"doubao-seed-2.0-lite":{inputPer1M:.005,outputPer1M:.015}},"gemini-cli":{},"codex-cli":{}};function Et(t,e,n){let o=n?.[t]?.[e];return o||cn[t]?.[e]}i(Et,"getCostRate");function Pt(t,e,n,o=0){let r=e/1e6*t.inputPer1M,s=n/1e6*t.outputPer1M,a=t.cachePer1M?o/1e6*t.cachePer1M:0;return r+s+a}i(Pt,"calculateCost");function so(){return{records:[],retries:[]}}i(so,"createCostTracker");function ao(t,e){let n={timestamp:Date.now(),provider:e.provider,waitMs:e.waitMs,reason:e.reason,source:e.source};return{records:t.records,retries:[...t.retries,n]}}i(ao,"recordRetry");function wt(t,e,n){let o=Et(e.provider,e.model,n),r=(e.cacheReadTokens??0)+(e.cacheWriteTokens??0),s=o?Pt(o,e.inputTokens,e.outputTokens,r):0,a={timestamp:Date.now(),provider:e.provider,model:e.model,inputTokens:e.inputTokens,outputTokens:e.outputTokens,cacheReadTokens:e.cacheReadTokens??0,cacheWriteTokens:e.cacheWriteTokens??0,cost:s,role:e.role};return{records:[...t.records,a],retries:t.retries}}i(wt,"recordUsage");function co(t){let e=0,n=0,o=0,r=0,s=0,a=0,l={},c={};for(let f of t.records){e+=f.cost,n+=f.inputTokens,o+=f.outputTokens,r+=f.cacheReadTokens+f.cacheWriteTokens,s+=f.cacheReadTokens,a+=f.cacheWriteTokens;let _=l[f.provider];l[f.provider]={cost:(_?.cost??0)+f.cost,calls:(_?.calls??0)+1,inputTokens:(_?.inputTokens??0)+f.inputTokens,outputTokens:(_?.outputTokens??0)+f.outputTokens};let M=f.role??"default",x=c[M];c[M]={cost:(x?.cost??0)+f.cost,calls:(x?.calls??0)+1,inputTokens:(x?.inputTokens??0)+f.inputTokens,outputTokens:(x?.outputTokens??0)+f.outputTokens}}let m=0;for(let f of t.retries)m+=f.waitMs;let v=r>0?s/r:0;return{totalCost:e,totalInputTokens:n,totalOutputTokens:o,totalCacheTokens:r,totalCacheReadTokens:s,totalCacheWriteTokens:a,cacheHitRate:v,callCount:t.records.length,retryCount:t.retries.length,retryWaitMs:m,byProvider:l,byRole:c}}i(co,"getSummary");function Be(t){return t<.01?`$${t.toFixed(4)}`:t<1?`$${t.toFixed(3)}`:`$${t.toFixed(2)}`}i(Be,"formatCost");function lo(t){let e=[];if(e.push(`Session Cost: ${Be(t.totalCost)} (${t.callCount} calls)`),e.push(`Tokens: ${t.totalInputTokens.toLocaleString()} in / ${t.totalOutputTokens.toLocaleString()} out`),t.totalCacheTokens>0){let r=(t.cacheHitRate*100).toFixed(0);e.push(`Cache: ${t.totalCacheTokens.toLocaleString()} tokens (${t.totalCacheReadTokens.toLocaleString()} read / ${t.totalCacheWriteTokens.toLocaleString()} write, ${r}% hit rate)`)}if(t.retryCount>0){let r=(t.retryWaitMs/1e3).toFixed(1);e.push(`Retries: ${t.retryCount} (${r}s total wait)`)}e.push("");let n=Object.entries(t.byProvider).sort((r,s)=>s[1].cost-r[1].cost);if(n.length>0){e.push("By Provider:");for(let[r,s]of n)e.push(`  ${r}: ${Be(s.cost)} (${s.calls} calls, ${s.inputTokens.toLocaleString()} in / ${s.outputTokens.toLocaleString()} out)`);e.push("")}let o=Object.entries(t.byRole).sort((r,s)=>s[1].cost-r[1].cost);if(o.length>1){e.push("By Role:");for(let[r,s]of o)e.push(`  ${r}: ${Be(s.cost)} (${s.calls} calls)`)}return e.join(`
+`)}i(lo,"formatCostReport");var ln={inputTokens:0,outputTokens:0,totalTokens:0},uo=3e4;async function po(t){let e=new AbortController,n=t.timeoutMs??uo,o,r=i(l=>{o||(o=l),e.abort()},"recordAbort"),s=setTimeout(()=>r("timeout"),n),a=i(()=>r("parent"),"onParentAbort");t.abortSignal&&(t.abortSignal.aborted?r("parent"):t.abortSignal.addEventListener("abort",a,{once:!0}));try{let l=await t.provider.stream([...t.messages],[],t.system,t.reasoning??{mode:"off"},{modelOverride:t.model},e.signal),c=l.usage??ln,m=l.textBlocks??[],v=l.toolBlocks??[],f=m.map(M=>M.text).join("");if(v.length>0)return{text:f,usage:c,costTracker:t.costTracker,stopReason:"error",error:new Error(`sideQuery: provider returned ${v.length} tool_use block(s); sideQuery expects text-only output`)};let _=t.costTracker;return _&&(_=wt(_,{provider:t.provider.name,model:t.model,inputTokens:c.inputTokens,outputTokens:c.outputTokens,cacheReadTokens:c.cachedReadTokens,cacheWriteTokens:c.cachedWriteTokens,role:t.querySource})),{text:f,usage:c,costTracker:_,stopReason:mo(l.stopReason)}}catch(l){let c=l instanceof Error?l:new Error(String(l)),m="error";return e.signal.aborted&&(m=o==="timeout"?"timeout":"aborted"),{text:"",usage:ln,costTracker:t.costTracker,stopReason:m,error:c}}finally{clearTimeout(s),t.abortSignal&&t.abortSignal.removeEventListener("abort",a)}}i(po,"sideQuery");function mo(t){return t==="max_tokens"?"max_tokens":"end_turn"}i(mo,"mapStopReason");export{H as a,R as b,de as c,Fe as d,Ue as e,Xe as f,J as g,We as h,un as i,pn as j,Rt as k,At as l,$e as m,Q as n,ke as o,St as p,q,Z as r,je as s,Ve as t,Ye as u,ee as v,be as w,Qe as x,yn as y,ve as z,Me as A,Ce as B,G as C,te as D,qe as E,ne as F,Ze as G,ge as H,K as I,N as J,re as K,se as L,he as M,ye as N,z as O,U as P,kt as Q,Ht as R,Vt as S,Yt as T,Jt as U,Qt as V,qt as W,Zt as X,B as Y,Oe as Z,en as _,Ie as $,ce as aa,De as ba,tn as ca,nn as da,on as ea,Ke as fa,Le as ga,Ne as ha,rn as ia,sn as ja,an as ka,Tt as la,cn as ma,Et as na,Pt as oa,so as pa,ao as qa,wt as ra,co as sa,Be as ta,lo as ua,po as va};