@bitkyc08/opencodex 2.81.0 → 2.82.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/bin/ocx.mjs +166 -73
- package/gui/dist/assets/App-Ds_qL7o8.js +54 -0
- package/gui/dist/assets/{Tray-EwAOsP0S.js → Tray-BqHF5Trd.js} +1 -1
- package/gui/dist/assets/{index-Brk6F3fn.css → index-BAbcj51J.css} +1 -1
- package/gui/dist/assets/{index-RsoaAxXY.js → index-D3FBCNiP.js} +13 -13
- package/gui/dist/assets/{quota-summary-D2lN8tDm.js → quota-summary-DTBC8n7-.js} +1 -1
- package/gui/dist/index.html +2 -2
- package/package.json +1 -1
- package/src/adapters/anthropic.ts +1 -1
- package/src/adapters/openai-responses/compaction-search-history.ts +109 -0
- package/src/adapters/openai-responses/passthrough.ts +9 -4
- package/src/adapters/openai-responses/tool-output-recovery.ts +20 -0
- package/src/claude/agents-inject.ts +14 -3
- package/src/claude/context-windows.ts +19 -17
- package/src/claude/desktop-3p.ts +88 -12
- package/src/claude/desktop-profile.ts +2 -1
- package/src/claude/gateway-cache.ts +18 -1
- package/src/claude/inbound.ts +77 -3
- package/src/claude/intercept/cli-catalog.ts +3 -0
- package/src/claude/intercept/picker-ca-persistence.ts +7 -6
- package/src/claude/intercept/picker-models.ts +14 -6
- package/src/claude/intercept/settings.ts +3 -0
- package/src/claude/long-context.ts +66 -0
- package/src/claude/model-info.ts +19 -16
- package/src/claude/outbound.ts +94 -9
- package/src/claude/subagent-model.ts +40 -11
- package/src/cli/agent.ts +9 -5
- package/src/cli/aside-profile-recovery.ts +33 -0
- package/src/cli/capabilities-base.ts +5 -5
- package/src/cli/capabilities-integrations.ts +2 -2
- package/src/cli/cli-path-diagnostics.ts +132 -0
- package/src/cli/desktop-runtime-guidance.ts +65 -0
- package/src/cli/dispatch.ts +5 -0
- package/src/cli/doctor.ts +16 -3
- package/src/cli/index.ts +14 -9
- package/src/cli/integration-aside-sync.ts +2 -1
- package/src/cli/integrations.ts +16 -3
- package/src/cli/resolve.ts +25 -2
- package/src/cli/root.ts +4 -2
- package/src/cli/status.ts +64 -12
- package/src/cli/update-restart-child.ts +4 -2
- package/src/cli/update-restart-home.ts +28 -10
- package/src/cli/update-restart-service-record.ts +116 -0
- package/src/cli/update-restart-supervision.ts +96 -0
- package/src/cli/update-restart.ts +59 -13
- package/src/cli/v2.ts +27 -20
- package/src/cli/version-skew-notice.ts +77 -0
- package/src/clients/lazycodex.ts +1 -1
- package/src/clients/omo-role-models.ts +32 -7
- package/src/codex/auth-api/account-selection.ts +107 -0
- package/src/codex/auth-api/routes.ts +10 -49
- package/src/codex/auth-context.ts +20 -1
- package/src/codex/autostart-health.ts +12 -3
- package/src/codex/catalog/build-entries.ts +18 -0
- package/src/codex/catalog/bundled.ts +155 -26
- package/src/codex/catalog/combo-member.ts +3 -2
- package/src/codex/catalog/effort.ts +26 -5
- package/src/codex/catalog/gather-capture.ts +5 -0
- package/src/codex/catalog/model-hints.ts +15 -8
- package/src/codex/catalog/model-visibility.ts +3 -1
- package/src/codex/catalog/operator-tiers.ts +38 -0
- package/src/codex/catalog/provider-models.ts +43 -35
- package/src/codex/catalog/retained-sync.ts +7 -5
- package/src/codex/catalog/routed-gather.ts +15 -1
- package/src/codex/catalog-auto-refresh-sources.ts +3 -4
- package/src/codex/catalog-auto-refresh.ts +19 -0
- package/src/codex/catalog.ts +1 -1
- package/src/codex/codex-home-owner.ts +3 -0
- package/src/codex/config-write-lock.ts +333 -0
- package/src/codex/convergence.ts +2 -1
- package/src/codex/features.ts +257 -200
- package/src/codex/inject/config-write-section.ts +34 -0
- package/src/codex/inject/multi-agent-v2.ts +14 -6
- package/src/codex/inject/remove.ts +36 -11
- package/src/codex/inject/restore.ts +85 -20
- package/src/codex/inject-coordination.ts +15 -6
- package/src/codex/inject.ts +62 -20
- package/src/codex/journal.ts +62 -17
- package/src/codex/main-account-cache.ts +43 -0
- package/src/codex/paths.ts +2 -0
- package/src/codex/prepared-home.ts +18 -0
- package/src/codex/prompt-journal.ts +25 -12
- package/src/codex/prompt-layers.ts +49 -26
- package/src/codex/prompt-lock-claim.ts +134 -0
- package/src/codex/prompt-lock-evidence.ts +26 -0
- package/src/codex/prompt-lock-io.ts +16 -0
- package/src/codex/prompt-lock-owner.ts +130 -0
- package/src/codex/prompt-lock.ts +84 -53
- package/src/codex/routing/failure-window.ts +146 -0
- package/src/codex/routing/selection.ts +4 -1
- package/src/codex/routing.ts +13 -17
- package/src/codex/runtime.ts +106 -18
- package/src/codex/shim.ts +45 -19
- package/src/combos/jev-service-exchange.ts +281 -0
- package/src/combos/jev.ts +34 -220
- package/src/config/atomic-write.ts +1 -1
- package/src/config/codex-ws-reuse-setting.ts +10 -0
- package/src/config/live-reconcile.ts +17 -0
- package/src/config/load-degrade.ts +8 -0
- package/src/config/rebase-provenance.ts +11 -0
- package/src/config/schema/config-schema.ts +4 -0
- package/src/generated/compatibility-version.json +303 -171
- package/src/generated/model-metadata.ts +2 -2
- package/src/grok/inject.ts +50 -4
- package/src/grok/sync.ts +2 -1
- package/src/images/loop.ts +29 -11
- package/src/integrations/aside-profiles.ts +10 -3
- package/src/integrations/catalog-refresh.ts +18 -2
- package/src/integrations/writer.ts +9 -3
- package/src/lib/bun-path-runtime.d.mts +32 -0
- package/src/lib/bun-path-runtime.mjs +95 -0
- package/src/lib/bun-runtime-preflight.ts +140 -0
- package/src/lib/desktop-cli-handoff.d.mts +20 -0
- package/src/lib/desktop-cli-handoff.mjs +92 -0
- package/src/lib/desktop-cli-record.d.mts +29 -0
- package/src/lib/desktop-cli-record.mjs +102 -0
- package/src/lib/request-execution-budget.ts +8 -0
- package/src/lib/synced-state-location.ts +46 -6
- package/src/lib/test-home-guard.ts +87 -1
- package/src/lib/translator-budget.ts +17 -4
- package/src/lib/windows-atomic-replace.ts +38 -16
- package/src/lib/windows-elevation.ts +6 -0
- package/src/lib/windows-secret-acl.ts +92 -25
- package/src/lib/winsw.ts +2 -3
- package/src/oauth/anthropic-account-refusal.ts +39 -1
- package/src/oauth/anthropic-routing.ts +1 -1
- package/src/oauth/anthropic-send-ownership.ts +3 -3
- package/src/oauth/chatgpt.ts +113 -41
- package/src/oauth/index.ts +6 -2
- package/src/oauth/store.ts +1 -1
- package/src/providers/azure-model-metadata.ts +115 -0
- package/src/providers/derive.ts +21 -0
- package/src/providers/quota/vendor-probes-key.ts +127 -17
- package/src/providers/quota.ts +1 -1
- package/src/responses/reasoning-envelope.ts +29 -1
- package/src/responses/reasoning-replay-cache.ts +17 -1
- package/src/responses/spill-store.ts +15 -2
- package/src/responses/state/metrics.ts +8 -0
- package/src/responses/state/spill-failure.ts +12 -0
- package/src/responses/state/spill-queue.ts +12 -5
- package/src/responses/state.ts +18 -10
- package/src/responses/task-input.ts +17 -0
- package/src/server/background-lifecycle.ts +27 -0
- package/src/server/chat-completions.ts +1 -0
- package/src/server/chat-native.ts +28 -12
- package/src/server/claude-messages.ts +43 -6
- package/src/server/inbound-body-admission.ts +1 -0
- package/src/server/index/claude-intercept-lifecycle.ts +3 -1
- package/src/server/index/serve-options.ts +10 -2
- package/src/server/index/spend-ledger-lifecycle.ts +2 -1
- package/src/server/index/startup-warnings.ts +19 -0
- package/src/server/index.ts +1 -0
- package/src/server/management/agent-settings-routes.ts +93 -32
- package/src/server/management/codex-agent-role-routes.ts +4 -2
- package/src/server/management/codex-prompt-routes.ts +1 -0
- package/src/server/management/codex-role-auto-assign.ts +13 -2
- package/src/server/management/config-routes.ts +59 -8
- package/src/server/management/context.ts +2 -2
- package/src/server/management/shared.ts +3 -3
- package/src/server/messages-native-oauth.ts +9 -8
- package/src/server/messages-native.ts +49 -13
- package/src/server/messages-response-headers.ts +14 -0
- package/src/server/request-log-conversation.ts +48 -5
- package/src/server/request-log.ts +31 -2
- package/src/server/responses/adapter-continuation.ts +2 -2
- package/src/server/responses/adapter-dispatch.ts +34 -13
- package/src/server/responses/codex-ws-exchange.ts +17 -2
- package/src/server/responses/codex-ws-pool.ts +100 -13
- package/src/server/responses/codex-ws-wire.ts +9 -0
- package/src/server/responses/combo-requested-effort.ts +3 -1
- package/src/server/responses/core-codex-account.ts +22 -1
- package/src/server/responses/core-combo-native.ts +20 -3
- package/src/server/responses/core-combo.ts +22 -14
- package/src/server/responses/core-lifetime.ts +3 -36
- package/src/server/responses/core-options.ts +5 -0
- package/src/server/responses/core-replay.ts +73 -0
- package/src/server/responses/core.ts +3 -3
- package/src/server/responses/fetch-helpers.ts +26 -3
- package/src/server/responses/passthrough-delivery.ts +24 -2
- package/src/server/responses/passthrough-dispatch.ts +19 -2
- package/src/server/responses/request-prepare.ts +4 -0
- package/src/server/responses/request-send-budget.ts +112 -49
- package/src/server/responses/response-effects.ts +1 -0
- package/src/server/responses/run-turn-execution.ts +2 -0
- package/src/server/responses/sidecar-execution.ts +25 -16
- package/src/server/responses/sidecar-send-budget.ts +58 -0
- package/src/server/responses/ws-upstream.ts +19 -6
- package/src/server/startup-health-cache.ts +2 -1
- package/src/server/workflow-refusal.ts +3 -0
- package/src/service/cli.ts +20 -4
- package/src/service/desktop-command-guard.ts +32 -0
- package/src/service/desktop-startup.ts +66 -40
- package/src/service/desktop-supervision.d.mts +26 -0
- package/src/service/desktop-supervision.mjs +140 -0
- package/src/service/managing-cli.ts +1 -0
- package/src/service/orchestration.ts +46 -20
- package/src/service/repair.ts +25 -9
- package/src/service/state.ts +4 -3
- package/src/service/systemd.ts +17 -8
- package/src/service/windows-ops.ts +61 -22
- package/src/storage/policy-job.ts +5 -1
- package/src/storage/policy-worker.ts +3 -1
- package/src/storage/policy.ts +9 -0
- package/src/types/config.ts +21 -0
- package/src/types/request.ts +5 -0
- package/src/types/wire.ts +1 -0
- package/src/update/index.ts +57 -9
- package/src/update/restart-ownership.ts +10 -3
- package/src/update/runtime-ownership.d.mts +6 -0
- package/src/update/runtime-ownership.mjs +23 -7
- package/src/web-search/loop.ts +5 -2
- package/gui/dist/assets/App-7yhoys8a.js +0 -54
package/src/claude/desktop-3p.ts
CHANGED
|
@@ -10,9 +10,10 @@ import {
|
|
|
10
10
|
} from "./desktop-3p-library";
|
|
11
11
|
export { resolveDesktop3pConfigLibraryPath, type Desktop3pConfigLibraryOptions } from "./desktop-3p-library";
|
|
12
12
|
import { createHash, randomUUID } from "node:crypto";
|
|
13
|
-
import { copyFileSync, existsSync, mkdirSync, readFileSync, unlinkSync } from "node:fs";
|
|
13
|
+
import { copyFileSync, existsSync, mkdirSync, readFileSync, lstatSync, unlinkSync } from "node:fs";
|
|
14
14
|
import { join } from "node:path";
|
|
15
15
|
import { atomicWriteFile, readConfigDiagnostics, withConfigMutationLockSync } from "../config";
|
|
16
|
+
import { atomicWriteFileNoFollow } from "../config/atomic-write";
|
|
16
17
|
import { claudeDesktopIntegrationEnabled } from "../codex/desired-state";
|
|
17
18
|
import type { OcxClaudeDesktopProfile } from "../types";
|
|
18
19
|
import {
|
|
@@ -24,6 +25,7 @@ import {
|
|
|
24
25
|
import { nativeOpenAiContextWindow, type NativeContextLimitsInput } from "../codex/catalog";
|
|
25
26
|
import { localAdmissionToken, localInferenceDestination } from "../lib/local-destinations";
|
|
26
27
|
import { assertDesktop3pModelsValid } from "./desktop-3p-guard";
|
|
28
|
+
import { claudeSurfaceSupportsOneMillion, ONE_MILLION } from "./long-context";
|
|
27
29
|
|
|
28
30
|
export interface Desktop3pModelEntry {
|
|
29
31
|
name: string;
|
|
@@ -55,12 +57,12 @@ export interface Desktop3pRoutedModel {
|
|
|
55
57
|
}
|
|
56
58
|
|
|
57
59
|
/**
|
|
58
|
-
* 1M
|
|
59
|
-
*
|
|
60
|
-
*
|
|
61
|
-
*
|
|
60
|
+
* The genuine-1M threshold real Anthropic rows still need. Every other row follows the
|
|
61
|
+
* long-window rule; the writer, the profile renderer and the dashboard DTO all decide through
|
|
62
|
+
* `claudeSurfaceSupportsOneMillion` (long-context.ts), so the 1M chip cannot disagree with the
|
|
63
|
+
* written config.
|
|
62
64
|
*/
|
|
63
|
-
export const DESKTOP_SUPPORTS_1M_THRESHOLD =
|
|
65
|
+
export const DESKTOP_SUPPORTS_1M_THRESHOLD = ONE_MILLION;
|
|
64
66
|
|
|
65
67
|
/** CLI arg parsing for `ocx claude desktop` mode flags (mutually exclusive). */
|
|
66
68
|
export function parseDesktop3pModeArgs(flags: string[]): { mode: Desktop3pConfigMode } | { error: string } {
|
|
@@ -251,7 +253,7 @@ function collectDesktop3pModels(
|
|
|
251
253
|
for (const { provider, id, contextWindow } of candidates) {
|
|
252
254
|
const route = `${provider}/${id}`;
|
|
253
255
|
const alias = desktop3pAlias(provider, id);
|
|
254
|
-
const supports1m =
|
|
256
|
+
const supports1m = claudeSurfaceSupportsOneMillion(provider, id, contextWindow)
|
|
255
257
|
? { supports1m: true as const }
|
|
256
258
|
: {};
|
|
257
259
|
if (alias === id) {
|
|
@@ -317,16 +319,30 @@ export function buildDesktop3pRegistry(
|
|
|
317
319
|
}
|
|
318
320
|
|
|
319
321
|
/** Generate Claude Desktop 3P model entries from the proxy's available models. */
|
|
322
|
+
export interface Desktop3pModelOptions {
|
|
323
|
+
/**
|
|
324
|
+
* false under claudeCode.contextAccounting "200k": a long-window model stays selectable at 1M
|
|
325
|
+
* (`supports1m`) but no longer defaults to it (`prefer1m` dropped).
|
|
326
|
+
*/
|
|
327
|
+
preferOneMillion?: boolean;
|
|
328
|
+
}
|
|
329
|
+
|
|
320
330
|
export function generateDesktop3pModels(
|
|
321
331
|
nativeSlugs: string[],
|
|
322
332
|
routedModels: Array<Desktop3pRoutedModel>,
|
|
323
333
|
profile?: OcxClaudeDesktopProfile,
|
|
324
334
|
nativeContextCap?: NativeContextLimitsInput,
|
|
335
|
+
options: Desktop3pModelOptions = {},
|
|
325
336
|
): Desktop3pModelEntry[] {
|
|
326
337
|
const { models, registry, realAnthropicIds } = collectDesktop3pModels(nativeSlugs, routedModels, profile, nativeContextCap);
|
|
327
338
|
desktop3pRegistry = registry;
|
|
328
339
|
desktop3pRealAnthropicIds = realAnthropicIds;
|
|
329
|
-
return models;
|
|
340
|
+
return options.preferOneMillion === false ? models.map(({ prefer1m: _prefer1m, ...model }) => model) : models;
|
|
341
|
+
}
|
|
342
|
+
|
|
343
|
+
/** The Desktop model options a config asks for (one place, so writer and export agree). */
|
|
344
|
+
export function desktop3pModelOptions(claudeCode: { contextAccounting?: string } | undefined): Desktop3pModelOptions {
|
|
345
|
+
return { preferOneMillion: claudeCode?.contextAccounting !== "200k" };
|
|
330
346
|
}
|
|
331
347
|
|
|
332
348
|
/** Resolve an alias using the most recently generated Desktop model registry. */
|
|
@@ -382,6 +398,7 @@ export function generateDesktop3pConfig(
|
|
|
382
398
|
mode: Desktop3pConfigMode = "static",
|
|
383
399
|
profile?: OcxClaudeDesktopProfile,
|
|
384
400
|
nativeContextCap?: NativeContextLimitsInput,
|
|
401
|
+
options: Desktop3pModelOptions = {},
|
|
385
402
|
): object {
|
|
386
403
|
const base = {
|
|
387
404
|
inferenceProvider: "gateway",
|
|
@@ -398,7 +415,7 @@ export function generateDesktop3pConfig(
|
|
|
398
415
|
...base,
|
|
399
416
|
modelDiscoveryEnabled: mode === "hybrid",
|
|
400
417
|
inferenceModels: (() => {
|
|
401
|
-
const models = generateDesktop3pModels(nativeSlugs, routedModels, profile, nativeContextCap);
|
|
418
|
+
const models = generateDesktop3pModels(nativeSlugs, routedModels, profile, nativeContextCap, options);
|
|
402
419
|
// Fail loud at the write boundary rather than ship a config Desktop rejects:
|
|
403
420
|
// the output counterpart of the request-path guards.
|
|
404
421
|
assertDesktop3pModelsValid(models);
|
|
@@ -662,6 +679,7 @@ export function writeDesktop3pConfig(
|
|
|
662
679
|
profile?: OcxClaudeDesktopProfile,
|
|
663
680
|
nativeContextCap?: NativeContextLimitsInput,
|
|
664
681
|
lifecycleLockDeps?: ClientLifecycleLockDeps,
|
|
682
|
+
refreshOnly?: { appliedFingerprint: string; admit: () => boolean },
|
|
665
683
|
): { written: boolean; path: string; reason?: string; fingerprint?: string } {
|
|
666
684
|
try {
|
|
667
685
|
return withClientLifecycleSync(() => withConfigMutationLockSync(() => {
|
|
@@ -674,9 +692,29 @@ export function writeDesktop3pConfig(
|
|
|
674
692
|
if (!claudeDesktopIntegrationEnabled(latest.config)) {
|
|
675
693
|
return { written: false, path: resolveDesktop3pConfigLibraryPath(), reason: "desired_state_changed" };
|
|
676
694
|
}
|
|
695
|
+
if (refreshOnly && !desktopRefreshPathsSafe(resolveDesktop3pConfigLibraryPath())) {
|
|
696
|
+
return { written: false, path: resolveDesktop3pConfigLibraryPath(), reason: "desktop_refresh_only_skipped" };
|
|
697
|
+
}
|
|
677
698
|
if (connection.kind === "connected" || inspectRemoteDesktopCleanup().kind !== "absent") {
|
|
678
699
|
return { written: false, path: resolveDesktop3pConfigLibraryPath(), reason: "desktop_remote_store_active" };
|
|
679
700
|
}
|
|
701
|
+
let requiredId: string | undefined;
|
|
702
|
+
if (refreshOnly) {
|
|
703
|
+
const refused = { written: false, path: resolveDesktop3pConfigLibraryPath(), reason: "desktop_refresh_only_skipped" };
|
|
704
|
+
// Ownership and bytes must still match under L→C; discovery outside these locks is not authority.
|
|
705
|
+
if (!refreshOnly.admit() || latest.config.claudeCode?.desktopProfile?.appliedFingerprint !== refreshOnly.appliedFingerprint) return refused;
|
|
706
|
+
const inspected = inspectDesktop3pConfigLibrary({ appliedFingerprint: refreshOnly.appliedFingerprint });
|
|
707
|
+
if (inspected.kind !== "gateway_ours" || !inspected.appliedId || !inspected.selectedProfilePath) return refused;
|
|
708
|
+
if (!desktopRefreshPathsSafe(inspected.libraryPath, inspected.selectedProfilePath)) return refused;
|
|
709
|
+
const metadata = parseMetadata(join(inspected.libraryPath, "_meta.json"));
|
|
710
|
+
if (!isOwnedDesktopGatewayEntry(metadata.entries.find(entry => entry.id === inspected.appliedId))) return refused;
|
|
711
|
+
try {
|
|
712
|
+
const selected = JSON.parse(readFileSync(inspected.selectedProfilePath, "utf8")) as Record<string, unknown>;
|
|
713
|
+
mode = selected.modelDiscoveryEnabled !== true ? "static"
|
|
714
|
+
: Array.isArray(selected.inferenceModels) ? "hybrid" : "discovery";
|
|
715
|
+
} catch { return refused; }
|
|
716
|
+
requiredId = inspected.appliedId;
|
|
717
|
+
}
|
|
680
718
|
// Claude Desktop runs on this machine, so it dials the unauthenticated loopback listener
|
|
681
719
|
// when one is enabled — on such a hub that is the only credential-free local socket
|
|
682
720
|
// (#4236) — and otherwise the bind address, which answers but demands data-plane
|
|
@@ -700,8 +738,9 @@ export function writeDesktop3pConfig(
|
|
|
700
738
|
);
|
|
701
739
|
}
|
|
702
740
|
return writeDesktop3pConfigWithGenerator(() => (
|
|
703
|
-
generateDesktop3pConfig(destination.origin, nativeSlugs, routedModels, gatewayKey, mode, profile, nativeContextCap
|
|
704
|
-
|
|
741
|
+
generateDesktop3pConfig(destination.origin, nativeSlugs, routedModels, gatewayKey, mode, profile, nativeContextCap,
|
|
742
|
+
desktop3pModelOptions(latest.config.claudeCode))
|
|
743
|
+
), requiredId && refreshOnly ? { requiredId, expectedFingerprint: refreshOnly.appliedFingerprint, admit: refreshOnly.admit } : undefined);
|
|
705
744
|
}), lifecycleLockDeps);
|
|
706
745
|
} catch { return { written: false, path: resolveDesktop3pConfigLibraryPath(), reason: "desktop_lifecycle_busy_or_unsafe" }; }
|
|
707
746
|
}
|
|
@@ -734,17 +773,38 @@ export function writeRemoteDesktop3pConfig(options: {
|
|
|
734
773
|
} catch { return { written: false, path: "", reason: "desktop_lifecycle_busy_or_unsafe" }; }
|
|
735
774
|
}
|
|
736
775
|
|
|
776
|
+
/** Refresh-only writes never follow links, including a dangling backup link. */
|
|
777
|
+
function desktopRefreshPathsSafe(libraryPath: string, configPath?: string): boolean {
|
|
778
|
+
try {
|
|
779
|
+
if (!lstatSync(libraryPath).isDirectory() || !lstatSync(join(libraryPath, "_meta.json")).isFile()) return false;
|
|
780
|
+
if (configPath) {
|
|
781
|
+
if (!lstatSync(configPath).isFile()) return false;
|
|
782
|
+
try { if (lstatSync(`${configPath}.bak`).isSymbolicLink()) return false; }
|
|
783
|
+
catch (error) { if ((error as NodeJS.ErrnoException).code !== "ENOENT") return false; }
|
|
784
|
+
}
|
|
785
|
+
return true;
|
|
786
|
+
} catch { return false; }
|
|
787
|
+
}
|
|
788
|
+
|
|
737
789
|
function writeDesktop3pConfigWithGenerator(
|
|
738
790
|
generate: () => object,
|
|
791
|
+
refreshOnly?: { requiredId: string; expectedFingerprint: string; admit: () => boolean },
|
|
739
792
|
): { written: boolean; path: string; reason?: string; fingerprint?: string } {
|
|
740
793
|
const libraryPath = resolveDesktop3pConfigLibraryPath();
|
|
741
794
|
const metadataPath = join(libraryPath, "_meta.json");
|
|
742
795
|
let configPath = libraryPath;
|
|
743
796
|
|
|
744
797
|
try {
|
|
745
|
-
|
|
798
|
+
if (refreshOnly && !desktopRefreshPathsSafe(libraryPath)) {
|
|
799
|
+
return { written: false, path: libraryPath, reason: "desktop_refresh_only_skipped" };
|
|
800
|
+
}
|
|
801
|
+
if (!refreshOnly) mkdirSync(libraryPath, { recursive: true, mode: 0o700 });
|
|
746
802
|
const metadata = parseMetadata(metadataPath);
|
|
747
803
|
const selected = metadata.entries.find(entry => entry?.id === metadata.appliedId && isOwnedDesktopGatewayEntry(entry));
|
|
804
|
+
// A refresh may reuse only the applied entry; fallback allocation would silently enroll again.
|
|
805
|
+
if (refreshOnly && selected?.id !== refreshOnly.requiredId) {
|
|
806
|
+
return { written: false, path: libraryPath, reason: "desktop_refresh_only_skipped" };
|
|
807
|
+
}
|
|
748
808
|
const existing = selected ?? metadata.entries.find(entry => isOwnedDesktopGatewayEntry(entry) && typeof entry.id === "string");
|
|
749
809
|
const id = existing?.id ?? randomUUID();
|
|
750
810
|
configPath = profilePath(libraryPath, id);
|
|
@@ -757,6 +817,22 @@ function writeDesktop3pConfigWithGenerator(
|
|
|
757
817
|
const preserved = readDesktopProfileForeignKeys(configPath);
|
|
758
818
|
const configJson = JSON.stringify({ ...preserved, ...generated }, null, 2) + "\n";
|
|
759
819
|
const fingerprint = createHash("sha256").update(configJson).digest("hex").slice(0, 16);
|
|
820
|
+
if (refreshOnly) {
|
|
821
|
+
atomicWriteFileNoFollow(configPath, configJson, undefined, {
|
|
822
|
+
validateBeforeRename: () => {
|
|
823
|
+
try {
|
|
824
|
+
if (!refreshOnly.admit() || !desktopRefreshPathsSafe(libraryPath, configPath)) throw new Error();
|
|
825
|
+
const freshMetadata = parseMetadata(metadataPath);
|
|
826
|
+
if (freshMetadata.appliedId !== refreshOnly.requiredId
|
|
827
|
+
|| !isOwnedDesktopGatewayEntry(freshMetadata.entries.find(entry => entry.id === refreshOnly.requiredId))
|
|
828
|
+
|| createHash("sha256").update(readFileSync(configPath)).digest("hex").slice(0, 16) !== refreshOnly.expectedFingerprint) {
|
|
829
|
+
throw new Error();
|
|
830
|
+
}
|
|
831
|
+
} catch { throw new Error("desktop_refresh_only_skipped"); }
|
|
832
|
+
},
|
|
833
|
+
});
|
|
834
|
+
return { written: true, path: configPath, fingerprint };
|
|
835
|
+
}
|
|
760
836
|
const { backupPath } = atomicReplaceDesktopConfig(configPath, configJson);
|
|
761
837
|
try {
|
|
762
838
|
atomicWriteFile(metadataPath, JSON.stringify({ ...metadata, appliedId: id, entries }, null, 2) + "\n");
|
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
import { createHash } from "node:crypto";
|
|
2
|
+
import { routeSupportsOneMillion } from "./long-context";
|
|
2
3
|
import type {
|
|
3
4
|
OcxClaudeDesktopAssignment,
|
|
4
5
|
OcxClaudeDesktopFamily,
|
|
@@ -359,7 +360,7 @@ export function renderDesktopProfile(
|
|
|
359
360
|
name: isRealAnthropicRoute(route) ? assignment.alias : desktopProfileWireAlias(assignment.alias),
|
|
360
361
|
family: assignment.family,
|
|
361
362
|
isFamilyDefault: effectiveDefaults[assignment.family] === route,
|
|
362
|
-
supports1m:
|
|
363
|
+
supports1m: routeSupportsOneMillion(route, model.contextWindow),
|
|
363
364
|
};
|
|
364
365
|
});
|
|
365
366
|
}
|
|
@@ -17,6 +17,7 @@ import { mkdirSync, writeFileSync } from "node:fs";
|
|
|
17
17
|
import { homedir } from "node:os";
|
|
18
18
|
import { join } from "node:path";
|
|
19
19
|
import { localAdmissionToken, localInferenceDestination } from "../lib/local-destinations";
|
|
20
|
+
import { assertNotRealClaudeConfigUnderTest } from "../lib/test-home-guard";
|
|
20
21
|
import type { OcxConfig } from "../types";
|
|
21
22
|
|
|
22
23
|
export interface GatewayModelRow {
|
|
@@ -46,11 +47,27 @@ export interface GatewayModelTarget {
|
|
|
46
47
|
/** Claude Code config dir (CLAUDE_CONFIG_DIR override honored, like the CLI). */
|
|
47
48
|
export function claudeConfigDir(): string {
|
|
48
49
|
const custom = process.env.CLAUDE_CONFIG_DIR;
|
|
49
|
-
return custom && custom.length > 0 ? custom : join(
|
|
50
|
+
return custom && custom.length > 0 ? custom : join(currentUserHome(), ".claude");
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
/**
|
|
54
|
+
* The home Claude Code itself resolves. Claude Code runs on Node, whose `os.homedir()`
|
|
55
|
+
* consults HOME (POSIX) or USERPROFILE (Windows) at call time when it is set; Bun's
|
|
56
|
+
* `os.homedir()` keeps the value it read at startup. In production the two agree. Under
|
|
57
|
+
* the test preload, which rewrites HOME after Bun has started, only this one follows the
|
|
58
|
+
* sandbox — the cached value is the developer's real home (#6775). Exported for tests.
|
|
59
|
+
*/
|
|
60
|
+
export function currentUserHome(env: NodeJS.ProcessEnv = process.env, platform: NodeJS.Platform = process.platform): string {
|
|
61
|
+
const fromEnv = platform === "win32" ? env.USERPROFILE : env.HOME;
|
|
62
|
+
return fromEnv !== undefined && fromEnv !== "" ? fromEnv : homedir();
|
|
50
63
|
}
|
|
51
64
|
|
|
52
65
|
/** Write the cache file; returns its path or null (best-effort, never throws). */
|
|
53
66
|
export function writeGatewayModelCache(baseUrl: string, models: readonly GatewayModelRow[], configDir = claudeConfigDir()): string | null {
|
|
67
|
+
// Outside the best-effort catch: an armed test process must not write here, and must
|
|
68
|
+
// fail loudly rather than degrade to "returned null".
|
|
69
|
+
// The file itself too: writeFileSync follows a link at gateway-models.json.
|
|
70
|
+
assertNotRealClaudeConfigUnderTest(configDir, join(configDir, "cache"), join(configDir, "cache", "gateway-models.json"));
|
|
54
71
|
try {
|
|
55
72
|
// Mirror the CLI's usable-id filter so our file matches what it would cache.
|
|
56
73
|
const usable = models.filter(m => /(claude|anthropic)/i.test(m.id));
|
package/src/claude/inbound.ts
CHANGED
|
@@ -281,7 +281,7 @@ function userMessageToItems(content: unknown, input: Rec[], elide: SkillElisionC
|
|
|
281
281
|
pushUserMessage(input, pending);
|
|
282
282
|
}
|
|
283
283
|
|
|
284
|
-
function assistantMessageToItems(content: unknown, input: Rec[], budget: TranslatorBudget): void {
|
|
284
|
+
function assistantMessageToItems(content: unknown, input: Rec[], budget: TranslatorBudget, requestedModel: string, nativeReasoningReplay: Map<string, string>): void {
|
|
285
285
|
if (typeof content === "string") {
|
|
286
286
|
if (content.length > 0) input.push({ type: "message", role: "assistant", content: [{ type: "output_text", text: content }] });
|
|
287
287
|
return;
|
|
@@ -314,6 +314,24 @@ function assistantMessageToItems(content: unknown, input: Rec[], budget: Transla
|
|
|
314
314
|
const owned = decodeReasoningEnvelope(signature, budget);
|
|
315
315
|
if (!owned) throw new AnthropicRequestError("malformed ocxr1 reasoning signature");
|
|
316
316
|
if (Object.hasOwn(owned, "sig")) throw new AnthropicRequestError("OpenCodex reasoning continuity cannot be replayed as an Anthropic signature");
|
|
317
|
+
if (owned.nat) {
|
|
318
|
+
// The provider's own blob goes back only to the model that minted it; any other
|
|
319
|
+
// model gets the visible text alone, as before this envelope field existed.
|
|
320
|
+
const nat = owned.nat.model === requestedModel ? owned.nat : undefined;
|
|
321
|
+
if (nat) {
|
|
322
|
+
const previous = nativeReasoningReplay.get(nat.enc);
|
|
323
|
+
nativeReasoningReplay.set(nat.enc, previous === undefined || previous === nat.tag ? nat.tag : "");
|
|
324
|
+
}
|
|
325
|
+
if (nat) budget.chargeRetained(2 * nat.enc.length, { kind: "reasoning" });
|
|
326
|
+
if (thinking.length === 0 && !nat) break;
|
|
327
|
+
input.push({
|
|
328
|
+
type: "reasoning",
|
|
329
|
+
id: nat?.id ?? `rs_${crypto.randomUUID().replace(/-/g, "")}`,
|
|
330
|
+
summary: thinking.length > 0 ? [{ type: "summary_text", text: thinking }] : [],
|
|
331
|
+
...(nat ? { encrypted_content: nat.enc } : {}),
|
|
332
|
+
});
|
|
333
|
+
break;
|
|
334
|
+
}
|
|
317
335
|
}
|
|
318
336
|
const encrypted = signature.length === 0 ? undefined : signature.startsWith(OCX_REASONING_PREFIX) ? signature : encodeReasoningEnvelope({ sig: signature }, budget);
|
|
319
337
|
if (encrypted) budget.chargeRetained(2 * encrypted.length, { kind: "reasoning" });
|
|
@@ -355,6 +373,7 @@ export type ClaudeCacheKeySource = "metadata" | "system" | null;
|
|
|
355
373
|
export interface ClaudeInboundTranslation {
|
|
356
374
|
body: Rec;
|
|
357
375
|
cacheKeySource: ClaudeCacheKeySource;
|
|
376
|
+
nativeReasoningReplay: ReadonlyMap<string, string>;
|
|
358
377
|
}
|
|
359
378
|
|
|
360
379
|
/**
|
|
@@ -400,6 +419,7 @@ function translateAnthropicRequest(
|
|
|
400
419
|
}
|
|
401
420
|
|
|
402
421
|
const input: Rec[] = [];
|
|
422
|
+
const nativeReasoningReplay = new Map<string, string>();
|
|
403
423
|
const systemParts: string[] = [];
|
|
404
424
|
const topLevelSystem = systemToInstructions(raw.system);
|
|
405
425
|
if (topLevelSystem !== undefined) systemParts.push(topLevelSystem);
|
|
@@ -411,7 +431,7 @@ function translateAnthropicRequest(
|
|
|
411
431
|
for (const msg of raw.messages) {
|
|
412
432
|
if (!isRec(msg)) throw new AnthropicRequestError("each message must be an object");
|
|
413
433
|
if (msg.role === "user") userMessageToItems(msg.content, input, elide);
|
|
414
|
-
else if (msg.role === "assistant") assistantMessageToItems(msg.content, input, budget);
|
|
434
|
+
else if (msg.role === "assistant") assistantMessageToItems(msg.content, input, budget, raw.model, nativeReasoningReplay);
|
|
415
435
|
else if (msg.role === "system") {
|
|
416
436
|
const text = systemMessageText(msg.content);
|
|
417
437
|
// Keep it where the client put it. `developer` is first-class in the Responses
|
|
@@ -531,7 +551,61 @@ function translateAnthropicRequest(
|
|
|
531
551
|
reasoning.effort = effortForThinkingBudget(thinking.budget_tokens);
|
|
532
552
|
}
|
|
533
553
|
body.reasoning = reasoning;
|
|
554
|
+
// store:false returns a provider's encrypted reasoning only on request. The outbound
|
|
555
|
+
// envelope (`nat`) carries it to the client so the next turn can hand it back.
|
|
556
|
+
body.include = ["reasoning.encrypted_content"];
|
|
534
557
|
}
|
|
535
558
|
|
|
536
|
-
return { body, cacheKeySource };
|
|
559
|
+
return { body, cacheKeySource, nativeReasoningReplay };
|
|
560
|
+
}
|
|
561
|
+
|
|
562
|
+
/**
|
|
563
|
+
* Remove proxy-owned native blobs before sending a Messages body to native Anthropic.
|
|
564
|
+
*
|
|
565
|
+
* Every decode, and every re-encode on top of it, is reserved against the request's translator
|
|
566
|
+
* budget first (eight bytes per code unit, as `decodeReasoningEnvelope` does), and every new
|
|
567
|
+
* signature is charged as a retained copy, so a body of near-limit envelopes cannot allocate past the per-turn bound.
|
|
568
|
+
* Throws TranslatorBudgetExceededError when it would; callers answer 413.
|
|
569
|
+
*/
|
|
570
|
+
export function nativeAnthropicProjection(body: Rec, budget: TranslatorBudget): Rec {
|
|
571
|
+
if (!Array.isArray(body.messages)) return body;
|
|
572
|
+
let changed = false;
|
|
573
|
+
const messages = body.messages.flatMap(message => {
|
|
574
|
+
if (!isRec(message) || message.role !== "assistant" || !Array.isArray(message.content)) return [message];
|
|
575
|
+
let localChanged = false;
|
|
576
|
+
const content = message.content.flatMap(block => {
|
|
577
|
+
if (!isRec(block) || block.type !== "thinking" || typeof block.signature !== "string"
|
|
578
|
+
|| !block.signature.startsWith(OCX_REASONING_PREFIX)) return [block];
|
|
579
|
+
// The decoded object stays alive while it is re-encoded, so its reservation is held until
|
|
580
|
+
// the new signature exists; the encode copies get their own reservation on top of it.
|
|
581
|
+
const decodeReservation = budget.reserveTransient(8 * block.signature.length, { kind: "reasoning" });
|
|
582
|
+
try {
|
|
583
|
+
let decoded: unknown;
|
|
584
|
+
try { decoded = JSON.parse(Buffer.from(block.signature.slice(OCX_REASONING_PREFIX.length), "base64").toString("utf8")); }
|
|
585
|
+
catch { decoded = undefined; }
|
|
586
|
+
if (!isRec(decoded)) { localChanged = true; return []; }
|
|
587
|
+
if (!Object.hasOwn(decoded, "nat")) return [block];
|
|
588
|
+
localChanged = true;
|
|
589
|
+
const { nat: _nat, ...rest } = decoded;
|
|
590
|
+
if (!rest.sig && !rest.red && !rest.krc && !(typeof rest.txt === "string" && rest.txt.length > 0)) return [];
|
|
591
|
+
// The re-encoded envelope is never larger than the decoded one (a field was removed), so a
|
|
592
|
+
// decode-sized reservation bounds the stringify, buffer and base64 copies before they exist.
|
|
593
|
+
const encodeReservation = budget.reserveTransient(8 * block.signature.length, { kind: "reasoning" });
|
|
594
|
+
let signature: string;
|
|
595
|
+
try {
|
|
596
|
+
signature = OCX_REASONING_PREFIX + Buffer.from(JSON.stringify(rest), "utf8").toString("base64");
|
|
597
|
+
} finally {
|
|
598
|
+
encodeReservation.release();
|
|
599
|
+
}
|
|
600
|
+
budget.chargeRetained(2 * signature.length, { kind: "reasoning" });
|
|
601
|
+
return [{ ...block, signature }];
|
|
602
|
+
} finally {
|
|
603
|
+
decodeReservation.release();
|
|
604
|
+
}
|
|
605
|
+
});
|
|
606
|
+
if (!localChanged) return [message];
|
|
607
|
+
changed = true;
|
|
608
|
+
return content.length ? [{ ...message, content }] : [];
|
|
609
|
+
});
|
|
610
|
+
return changed ? { ...body, messages } : body;
|
|
537
611
|
}
|
|
@@ -14,6 +14,7 @@
|
|
|
14
14
|
*/
|
|
15
15
|
import { readdirSync, unlinkSync } from "node:fs";
|
|
16
16
|
import { join } from "node:path";
|
|
17
|
+
import { assertNotRealClaudeConfigUnderTest } from "../../lib/test-home-guard";
|
|
17
18
|
import type { ClaudeFirstPartyDesired } from "../first-party-settings";
|
|
18
19
|
import { classifyInterceptClient } from "./client-class";
|
|
19
20
|
import { BOOTSTRAP_MAX_DECODED_BYTES, injectPickerModels, type PickerModelEntry } from "./picker-bootstrap";
|
|
@@ -150,6 +151,8 @@ export async function rewriteCliCatalogResponse(
|
|
|
150
151
|
*/
|
|
151
152
|
export function invalidateClaudeCodeServedCatalog(claudeDir: string): number {
|
|
152
153
|
const dir = join(claudeDir, "cache", "model-catalog");
|
|
154
|
+
// Outside the catch below: an armed test process must not prune the real Claude cache (#6775).
|
|
155
|
+
assertNotRealClaudeConfigUnderTest(claudeDir, dir);
|
|
153
156
|
let names: string[];
|
|
154
157
|
try { names = readdirSync(dir); } catch { return 0; } // no-excuse-ok: catch -- no cache directory means nothing to drop.
|
|
155
158
|
let removed = 0;
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { X509Certificate } from "node:crypto";
|
|
2
|
-
import { constants, closeSync, fstatSync, lstatSync, openSync, readSync, realpathSync, rmSync, writeFileSync, type
|
|
2
|
+
import { constants, closeSync, fstatSync, lstatSync, openSync, readSync, realpathSync, rmSync, writeFileSync, type BigIntStats } from "node:fs";
|
|
3
3
|
import { join } from "node:path";
|
|
4
4
|
import {
|
|
5
5
|
decodePickerCaCredential, openPickerCaCredential, pickerCaConfigId, readPickerCaCredential,
|
|
@@ -21,18 +21,19 @@ export function canonicalPickerConfigDir(configDir: string): string {
|
|
|
21
21
|
|
|
22
22
|
/** Public records still cannot follow symlinks or accept oversized/malformed recovery data. */
|
|
23
23
|
function readState(path: string, configId: string, initialization: boolean): Initialization | AuthorityMetadata | null {
|
|
24
|
-
|
|
25
|
-
|
|
24
|
+
// BigInt stats: a Windows file ID exceeds 2^53, so Number ino loses low bits and two files can compare equal.
|
|
25
|
+
let expected: BigIntStats;
|
|
26
|
+
try { expected = lstatSync(path, { bigint: true }); }
|
|
26
27
|
catch (error) { if ((error as NodeJS.ErrnoException).code === "ENOENT") return null; throw new Error("picker_ca_metadata_unsafe"); }
|
|
27
28
|
if (expected.isSymbolicLink()) throw new Error("picker_ca_metadata_unsafe");
|
|
28
29
|
let fd: number;
|
|
29
30
|
try { fd = openSync(path, constants.O_RDONLY | constants.O_NOFOLLOW | constants.O_NONBLOCK); }
|
|
30
31
|
catch (error) { if ((error as NodeJS.ErrnoException).code === "ENOENT") return null; throw new Error("picker_ca_metadata_unsafe"); }
|
|
31
32
|
try {
|
|
32
|
-
const stat = fstatSync(fd);
|
|
33
|
+
const stat = fstatSync(fd, { bigint: true });
|
|
33
34
|
if (stat.dev !== expected.dev || stat.ino !== expected.ino) throw new Error("picker_ca_metadata_unsafe");
|
|
34
|
-
if (!stat.isFile() || stat.nlink !==
|
|
35
|
-
|| (process.platform !== "win32" && stat.uid !== process.getuid!())) throw new Error();
|
|
35
|
+
if (!stat.isFile() || stat.nlink !== 1n || stat.size > BigInt(MAX_PUBLIC_STATE_BYTES)
|
|
36
|
+
|| (process.platform !== "win32" && stat.uid !== BigInt(process.getuid!()))) throw new Error();
|
|
36
37
|
const bytes = Buffer.alloc(MAX_PUBLIC_STATE_BYTES + 1);
|
|
37
38
|
let used = 0;
|
|
38
39
|
while (used < bytes.length) {
|
|
@@ -7,7 +7,8 @@ import { dirname } from "node:path";
|
|
|
7
7
|
import { nativeOpenAiContextWindow, type NativeContextLimitsInput } from "../../codex/catalog";
|
|
8
8
|
import type { OcxClaudeDesktopProfile } from "../../types";
|
|
9
9
|
import { aliasForRoute, claudeCodeNativeAlias } from "../alias";
|
|
10
|
-
import { AUTO_CONTEXT_OFF, stripOneMillionMarker, withOneMillionMarker } from "../context-windows";
|
|
10
|
+
import { AUTO_CONTEXT_OFF, stripOneMillionMarker, UNPAIRED_AUTO_CONTEXT, withOneMillionMarker, type AutoContextMode } from "../context-windows";
|
|
11
|
+
import { isAnthropicClaudeRoute } from "../long-context";
|
|
11
12
|
import { activeDesktop3pAlias, displayModelId, resolveDesktop3pAlias, type Desktop3pRoutedModel } from "../desktop-3p";
|
|
12
13
|
import { reconcileDesktopProfile, renderDesktopProfile, type DesktopProfileModel } from "../desktop-profile";
|
|
13
14
|
import type { PickerModelEntry } from "./picker-bootstrap";
|
|
@@ -17,6 +18,8 @@ export interface PickerRouteInput {
|
|
|
17
18
|
routedModels: Desktop3pRoutedModel[];
|
|
18
19
|
profile?: OcxClaudeDesktopProfile;
|
|
19
20
|
nativeContextCap?: NativeContextLimitsInput;
|
|
21
|
+
/** resolveAutoContext(config.claudeCode); only its contextAccounting "200k" opt-in matters here. */
|
|
22
|
+
auto?: AutoContextMode;
|
|
20
23
|
}
|
|
21
24
|
|
|
22
25
|
/** Candidate routes in the gateway profile's order and with its labels. */
|
|
@@ -46,9 +49,14 @@ function splitRoute(route: string): { provider: string; id: string } {
|
|
|
46
49
|
return { provider: route.slice(0, slash), id: route.slice(slash + 1) };
|
|
47
50
|
}
|
|
48
51
|
|
|
49
|
-
/**
|
|
50
|
-
|
|
51
|
-
|
|
52
|
+
/**
|
|
53
|
+
* Desktop runners do not inherit the proxy's compaction env, so a long window (>= the default
|
|
54
|
+
* compact window) is marked on the strength of the `prompt is too long` recovery instead
|
|
55
|
+
* (devlog/_plan/261009_claude_1m_default/020). Real Anthropic models need a genuine 1M.
|
|
56
|
+
*/
|
|
57
|
+
function pickerSelector(alias: string, contextWindow: number | undefined, provider: string, id: string, auto?: AutoContextMode): string {
|
|
58
|
+
const mode = auto?.accounting200k ? auto : isAnthropicClaudeRoute(provider, id) ? AUTO_CONTEXT_OFF : UNPAIRED_AUTO_CONTEXT;
|
|
59
|
+
return withOneMillionMarker(alias, contextWindow === undefined ? {} : { [alias]: contextWindow }, mode)!;
|
|
52
60
|
}
|
|
53
61
|
|
|
54
62
|
export function buildPickerModels(input: PickerRouteInput): PickerModelEntry[] {
|
|
@@ -61,7 +69,7 @@ export function buildPickerModels(input: PickerRouteInput): PickerModelEntry[] {
|
|
|
61
69
|
const alias = provider === "native" ? claudeCodeNativeAlias(id) : aliasForRoute(provider, id);
|
|
62
70
|
if (!alias || seen.has(alias)) continue;
|
|
63
71
|
seen.add(alias);
|
|
64
|
-
out.push({ id: pickerSelector(alias, model.contextWindow), name: model.label,
|
|
72
|
+
out.push({ id: pickerSelector(alias, model.contextWindow, provider, id, input.auto), name: model.label,
|
|
65
73
|
...(model.contextWindow === undefined ? {} : { contextWindow: model.contextWindow }) });
|
|
66
74
|
}
|
|
67
75
|
return out;
|
|
@@ -81,7 +89,7 @@ export function buildCliPickerModels(input: PickerRouteInput): PickerModelEntry[
|
|
|
81
89
|
const alias = activeDesktop3pAlias(provider, id);
|
|
82
90
|
if (seen.has(alias) || resolveDesktop3pAlias(alias) !== model.route) continue;
|
|
83
91
|
seen.add(alias);
|
|
84
|
-
out.push({ id: pickerSelector(alias, model.contextWindow), name: model.label,
|
|
92
|
+
out.push({ id: pickerSelector(alias, model.contextWindow, provider, id, input.auto), name: model.label,
|
|
85
93
|
description: `opencodex · ${model.route}`, route: model.route,
|
|
86
94
|
...(model.contextWindow === undefined ? {} : { contextWindow: model.contextWindow }) });
|
|
87
95
|
}
|
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import { mkdirSync, readFileSync } from "node:fs";
|
|
2
2
|
import { dirname, join } from "node:path";
|
|
3
3
|
import { atomicWriteFile } from "../../config/atomic-write";
|
|
4
|
+
import { assertNotRealClaudeConfigUnderTest } from "../../lib/test-home-guard";
|
|
4
5
|
import { claudeConfigDir } from "../auth-detect";
|
|
5
6
|
import { invalidateClaudeCodeServedCatalog } from "./cli-catalog";
|
|
6
7
|
|
|
@@ -112,6 +113,8 @@ export function inspectClaudeInterceptSettings(
|
|
|
112
113
|
}
|
|
113
114
|
|
|
114
115
|
function writeSettings(path: string, doc: SettingsDoc): void {
|
|
116
|
+
// An armed test process must not rewrite the real Claude settings.json (#6775).
|
|
117
|
+
assertNotRealClaudeConfigUnderTest(dirname(path), path);
|
|
115
118
|
mkdirSync(dirname(path), { recursive: true });
|
|
116
119
|
// The managed env embeds the proxy token, so the file must stay owner-only:
|
|
117
120
|
// atomicWriteFile applies the real NTFS ACL on Windows where chmod is a no-op.
|
|
@@ -0,0 +1,66 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Long-context thresholds shared by every Claude surface
|
|
3
|
+
* (devlog/_plan/261009_claude_1m_default/020_long_context_eligibility.md).
|
|
4
|
+
*
|
|
5
|
+
* A leaf module: desktop-3p.ts and desktop-profile.ts need the predicate, and context-windows.ts
|
|
6
|
+
* already imports desktop-3p.ts, so the predicate cannot live there.
|
|
7
|
+
*/
|
|
8
|
+
import { isAnthropicInstanceId } from "../providers/anthropic-instance-id";
|
|
9
|
+
|
|
10
|
+
export const ONE_MILLION = 1_000_000;
|
|
11
|
+
|
|
12
|
+
/** A window at or below this is Claude Code's own default accounting; widening it gains nothing. */
|
|
13
|
+
export const AUTO_CONTEXT_FLOOR = 200_000;
|
|
14
|
+
|
|
15
|
+
/**
|
|
16
|
+
* Auto-context defaults (devlog 260712 020, user-approved).
|
|
17
|
+
*
|
|
18
|
+
* The compact window is the token count at which Claude Code starts compacting, and it is
|
|
19
|
+
* also the floor `shouldMarkOneMillion` uses — a model may only carry the marker if it can
|
|
20
|
+
* host this window. 350,000 was chosen when the widest native row advertised 372,000.
|
|
21
|
+
*
|
|
22
|
+
* It now matches the auto-compaction limit the Codex catalog ships for the same models
|
|
23
|
+
* (`nativeAutoCompactLimit`: 829,800 against the 922,000 native window). Leaving the two
|
|
24
|
+
* apart meant one model compacting at 350k under Claude Code and at 829,800 under Codex.
|
|
25
|
+
* The value stays clear of the measured 922,000 ceiling by ~92k, so compaction still has
|
|
26
|
+
* room to run before the upstream refuses.
|
|
27
|
+
*/
|
|
28
|
+
export const AUTO_COMPACT_WINDOW_DEFAULT = 829_800;
|
|
29
|
+
|
|
30
|
+
/**
|
|
31
|
+
* Whether a window earns the 1M selection on a surface whose Claude Code runner may not inherit
|
|
32
|
+
* CLAUDE_CODE_AUTO_COMPACT_WINDOW (Desktop pickers, discovery, Desktop 3P, generated subagents).
|
|
33
|
+
* A window >= 1M always does. A smaller one does when it can host the default compact window, so
|
|
34
|
+
* a runner that has the variable compacts in time, and one that lacks it overflows into the
|
|
35
|
+
* `prompt is too long` envelope (anthropicErrorBody) and compacts reactively. The floor is fixed
|
|
36
|
+
* on purpose: a custom compact window must not re-admit a 372k route (#854).
|
|
37
|
+
*/
|
|
38
|
+
export function isLongContextWindow(window: number | undefined): boolean {
|
|
39
|
+
if (typeof window !== "number" || window <= 0) return false;
|
|
40
|
+
return window >= ONE_MILLION || (window > AUTO_CONTEXT_FLOOR && window >= AUTO_COMPACT_WINDOW_DEFAULT);
|
|
41
|
+
}
|
|
42
|
+
|
|
43
|
+
/**
|
|
44
|
+
* A Claude model on an Anthropic-named route. Claude windows are 200k or a genuine 1M and a bare
|
|
45
|
+
* id rides the native passthrough, so the long-window rule never widens one: a Claude row listed
|
|
46
|
+
* under 1M is a capped one, and on a runner without the compaction env it keeps the 200k
|
|
47
|
+
* accounting it had before long windows were widened. Keyed on the model id as well, so a
|
|
48
|
+
* configured gateway that only shares the name `anthropic2` and serves other models keeps the
|
|
49
|
+
* long-window rule. A custom `anthropic2` gateway serving Claude below 1M is treated like a pool
|
|
50
|
+
* row on these surfaces, which is the pre-existing (unwidened) behavior.
|
|
51
|
+
*/
|
|
52
|
+
export function isAnthropicClaudeRoute(provider: string, modelId: string): boolean {
|
|
53
|
+
return isAnthropicInstanceId(provider) && modelId.startsWith("claude-");
|
|
54
|
+
}
|
|
55
|
+
|
|
56
|
+
/** Desktop 3P / picker / discovery / dashboard 1M eligibility for one provider route. */
|
|
57
|
+
export function claudeSurfaceSupportsOneMillion(provider: string, modelId: string, window: number | undefined): boolean {
|
|
58
|
+
if (isAnthropicClaudeRoute(provider, modelId)) return typeof window === "number" && window >= ONE_MILLION;
|
|
59
|
+
return isLongContextWindow(window);
|
|
60
|
+
}
|
|
61
|
+
|
|
62
|
+
/** Same, for a `provider/id` route string. */
|
|
63
|
+
export function routeSupportsOneMillion(route: string, window: number | undefined): boolean {
|
|
64
|
+
const slash = route.indexOf("/");
|
|
65
|
+
return claudeSurfaceSupportsOneMillion(route.slice(0, slash), route.slice(slash + 1), window);
|
|
66
|
+
}
|
package/src/claude/model-info.ts
CHANGED
|
@@ -19,7 +19,8 @@ import { orderForModelPicker, catalogModelEfforts, nativeEffortClamp, nativeOpen
|
|
|
19
19
|
import { claudeCodeAlias, claudeCodeNativeAlias } from "./alias";
|
|
20
20
|
import { cursorFastIdFor } from "../adapters/cursor/catalog";
|
|
21
21
|
import { desktop3pAlias } from "./desktop-3p";
|
|
22
|
-
import { AUTO_CONTEXT_OFF, type AutoContextMode } from "./context-windows";
|
|
22
|
+
import { AUTO_CONTEXT_OFF, shouldMarkOneMillion, UNPAIRED_AUTO_CONTEXT, type AutoContextMode } from "./context-windows";
|
|
23
|
+
import { isAnthropicClaudeRoute } from "./long-context";
|
|
23
24
|
|
|
24
25
|
const MODEL_INFO_CREATED_AT = "2026-01-01T00:00:00Z";
|
|
25
26
|
const ANTHROPIC_EFFORT_RUNGS = new Set(["low", "medium", "high", "xhigh", "max"]);
|
|
@@ -144,23 +145,25 @@ export function buildAnthropicModelInfos(
|
|
|
144
145
|
: aliasForRoute(m.provider, m.id);
|
|
145
146
|
}),
|
|
146
147
|
]);
|
|
147
|
-
// [1m] picker variant (devlog 260712 B1
|
|
148
|
-
// carrying the marker (2.1.207 binary: /\[1m\]/i
|
|
149
|
-
//
|
|
150
|
-
//
|
|
151
|
-
//
|
|
152
|
-
//
|
|
148
|
+
// [1m] picker variant (devlog 260712 B1; widened in devlog/_plan/261009_claude_1m_default/020):
|
|
149
|
+
// Claude Code accounts exactly 1M for ids carrying the marker (2.1.207 binary: /\[1m\]/i -> 1e6,
|
|
150
|
+
// compaction preserved). A discovery client may lack CLAUDE_CODE_AUTO_COMPACT_WINDOW, so the
|
|
151
|
+
// rule is the fixed unpaired one (long-context.ts), not the configurable launch compact window:
|
|
152
|
+
// a custom 350k window must not re-admit a 372k route (#854). The server's auto mode only turns
|
|
153
|
+
// widening off (autoContext: false / maxContextTokens), falling back to >= 1M. Base and Fast
|
|
154
|
+
// rows stay unmarked: this list is a choice list. Guards (audit R1#11): same dedupe set, never
|
|
155
|
+
// double-suffix.
|
|
156
|
+
const variantMode = auto.enabled ? UNPAIRED_AUTO_CONTEXT : AUTO_CONTEXT_OFF;
|
|
153
157
|
const push1mVariant = (
|
|
154
158
|
base: AnthropicModelInfo,
|
|
155
159
|
contextWindow: number | undefined,
|
|
156
160
|
maxInputTokens?: number,
|
|
157
161
|
selectorId?: string,
|
|
162
|
+
mode: AutoContextMode = variantMode,
|
|
158
163
|
) => {
|
|
159
|
-
//
|
|
160
|
-
//
|
|
161
|
-
|
|
162
|
-
// Claude Code over-fill it (the #854 defect).
|
|
163
|
-
if (contextWindow === undefined || contextWindow < ONE_MILLION) return;
|
|
164
|
+
// A window >= 1M always earns the row; a long window earns it under the unpaired rule,
|
|
165
|
+
// because an overflow is answered as `prompt is too long`, which Claude Code compacts on.
|
|
166
|
+
if (!shouldMarkOneMillion(contextWindow, mode)) return;
|
|
164
167
|
if (base.id.includes("[1m]")) return;
|
|
165
168
|
const id = selectorId ?? `${base.id}[1m]`;
|
|
166
169
|
if (seen.has(id)) return;
|
|
@@ -170,9 +173,9 @@ export function buildAnthropicModelInfos(
|
|
|
170
173
|
// (measured — see devlog/_plan/260817_native_gpt56_1m_context/001_measurement_evidence.md).
|
|
171
174
|
// Advertising the flat 1e6 there would invite mid-session context_length_exceeded, so the
|
|
172
175
|
// variant reports whichever of the two is smaller.
|
|
173
|
-
|
|
174
|
-
|
|
175
|
-
|
|
176
|
+
// A long window under 1M must not be advertised as 1e6 input either.
|
|
177
|
+
const ceiling = typeof maxInputTokens === "number" && maxInputTokens > 0 ? maxInputTokens : contextWindow;
|
|
178
|
+
const advertised = typeof ceiling === "number" && ceiling > 0 ? Math.min(ONE_MILLION, ceiling) : ONE_MILLION;
|
|
176
179
|
out.push({ ...base, id, display_name: `${base.display_name} · 1M`, max_input_tokens: advertised });
|
|
177
180
|
};
|
|
178
181
|
/**
|
|
@@ -251,7 +254,7 @@ export function buildAnthropicModelInfos(
|
|
|
251
254
|
&& listedModelId.startsWith("claude-fable-")
|
|
252
255
|
? `${claudeCodeNativeAlias(listedModelId)}[1m]`
|
|
253
256
|
: undefined;
|
|
254
|
-
push1mVariant(info, m.contextWindow, routedMaxInput, oneMillionSelector);
|
|
257
|
+
push1mVariant(info, m.contextWindow, routedMaxInput, oneMillionSelector, isAnthropicClaudeRoute(m.provider, m.id) ? AUTO_CONTEXT_OFF : variantMode);
|
|
255
258
|
// The whole model is passed, not a (provider, id) pair: a combo row lives in its own
|
|
256
259
|
// namespace with no config.providers entry, so the caller classifies it from the
|
|
257
260
|
// aggregated supportsServiceTier the row already carries.
|