open-multi-agent-kit 0.96.2 → 0.97.0

package/dist/core/provider-resilience.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider-resilience.js","sourceRoot":"","sources":["../../src/core/provider-resilience.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAOH,oFAAoF;AACpF,MAAM,CAAC,MAAM,sBAAsB,GAAG,QAAQ,CAAC;AAE/C,8EAA8E;AAC9E,MAAM,CAAC,MAAM,2BAA2B,GAAG,oCAAoC,CAAC;AAEhF,wEAAwE;AACxE,MAAM,CAAC,MAAM,kCAAkC,GAAiC;IAC/E,EAAE,QAAQ,EAAE,aAAa,EAAE,EAAE,EAAE,IAAI,EAAE;IACrC,EAAE,QAAQ,EAAE,kBAAkB,EAAE,EAAE,EAAE,qBAAqB,EAAE;IAC3D,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE,EAAE,UAAU,EAAE;IACnC,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE,iBAAiB,EAAE;IAC/C,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE,mBAAmB,EAAE;IACjD,EAAE,QAAQ,EAAE,kBAAkB,EAAE,EAAE,EAAE,iBAAiB,EAAE;IACvD,EAAE,QAAQ,EAAE,aAAa,EAAE,EAAE,EAAE,iBAAiB,EAAE;CACzC,CAAC;AAWX,MAAM,CAAC,MAAM,2BAA2B,GAIpC;IACH,uBAAuB,EAAE,IAAI;IAC7B,wBAAwB,EAAE,IAAI;IAC9B,kBAAkB,EAAE,kCAAkC;CACtD,CAAC;AAEF,MAAM,UAAU,mBAAmB,CAAC,OAA2B,EAAE,QAA6B,EAAW;IACxG,MAAM,EAAE,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAClC,IAAI,CAAC,EAAE;QAAE,OAAO,KAAK,CAAC;IACtB,IAAI,sBAAsB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IACjD,MAAM,CAAC,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IACzC,OAAO,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI,sBAAsB,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CAClE;AAED,0EAA0E;AAC1E,MAAM,UAAU,uBAAuB,CAAC,OAA2B,EAAE,QAA6B,EAAW;IAC5G,MAAM,EAAE,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAClC,IAAI,CAAC,EAAE;QAAE,OAAO,KAAK,CAAC;IACtB,IAAI,2BAA2B,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IACtD,MAAM,CAAC,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IACzC,OAAO,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI,2BAA2B,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CACvE;AAED,MAAM,UAAU,0BAA0B,CAAC,IAAwB,EAAW;IAC7E,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,iHAAiH,CAAC,IAAI,CAC5H,IAAI,CACJ,CAAC;AAAA,CACF;AAED;;;;;GAKG;AACH,MAAM,UAAU,wBAAwB,CAAC,IAAwB,EAAW;IAC3E,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,8HAA8H,CAAC,IAAI,CACzI,IAAI,CACJ,CAAC;AAAA,CACF;AAED,0FAA0F;AAC1F,MAAM,UAAU,uBAAuB,CAAC,IAAwB,EAAW;IAC1E,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,oFAAoF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACvG;AAED;;;GAGG;AACH,MAAM,UAAU,+BAA+B,CAAC,IAAwB,EAAW;IAClF,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,0pBAA0pB,CAAC,IAAI,CACrqB,IAAI,CACJ,CAAC;AAAA,CACF;AAED,MAAM,UAAU,yBAAyB,CACxC,QAAgD,EACjB;IAC/B,MAAM,MAAM,GAAG,QAAQ,EAAE,kBAAkB,CAAC;IAC5C,IAAI,MAAM,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,MAAM,CAAC;IAC/C,OAAO,kCAAkC,CAAC;AAAA,CAC1C;AAED,0EAA0E;AAC1E,MAAM,UAAU,yBAAyB,CAAC,OAKzC,EAAW;IACX,IAAI,CAAC,OAAO,CAAC,wBAAwB,IAAI,OAAO,CAAC,WAAW;QAAE,OAAO,KAAK,CAAC;IAC3E,OAAO,CAAC,uBAAuB,CAAC,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;AAAA,CACnE;AAED,0EAA0E;AAC1E,MAAM,UAAU,4BAA4B,CAAC,OAG5C,EAAW;IACX,OAAO,OAAO,CAAC,uBAAuB,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC;AAAA,CAC/D;AAED,MAAM,UAAU,yBAAyB,CAAC,QAAgD,EAIxF;IACD,OAAO;QACN,uBAAuB,EAAE,QAAQ,EAAE,uBAAuB,IAAI,2BAA2B,CAAC,uBAAuB;QACjH,wBAAwB,EACvB,QAAQ,EAAE,wBAAwB,IAAI,2BAA2B,CAAC,wBAAwB;QAC3F,kBAAkB,EAAE,yBAAyB,CAAC,QAAQ,CAAC;KACvD,CAAC;AAAA,CACF;AAED;;;GAGG;AACH,MAAM,UAAU,qBAAqB,CACpC,UAAwC,EACxC,OAAyE,EACzE,SAAoD,EACpB;IAChC,KAAK,MAAM,CAAC,IAAI,UAAU,EAAE,CAAC;QAC5B,IAAI,OAAO,IAAI,CAAC,CAAC,QAAQ,KAAK,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC,EAAE,KAAK,OAAO,CAAC,EAAE;YAAE,SAAS;QAChF,IAAI,mBAAmB,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC;YAAE,SAAS;QACpD,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC;YAAE,SAAS;QAC5B,OAAO,CAAC,CAAC;IACV,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,MAAM,UAAU,wBAAwB,CAAC,OAAe,EAAE,QAAgB,EAAU;IACnF,OAAO,CACN,+BAA+B,QAAQ,IAAI,OAAO,iDAAiD;QACnG,mFAAmF;QACnF,0DAA0D,CAC1D,CAAC;AAAA,CACF","sourcesContent":["/**\n * Provider resilience — root-level recovery for sticky safety models and\n * sanitize-and-retry protocol failures (orphan tool_call_id, terminated streams).\n *\n * This is not a jailbreak layer. It keeps sessions alive when a provider\n * false-positives or returns a sticky transcript-shape error.\n */\n\nexport type FailoverCandidate = Readonly<{\n\treadonly provider: string;\n\treadonly id: string;\n}>;\n\n/** Models known to emit high false-positive content/safety stops on coding work. */\nexport const STICKY_SAFETY_MODEL_RE = /fable/i;\n\n/** Claude lines that must stay on-model when they refuse. No DeepSeek hop. */\nexport const NO_SAFETY_FAILOVER_MODEL_RE = /(?:claude-)?(?:fable|opus|sonnet)/i;\n\n/** Default failover chain when a sticky safety model refuses a turn. */\nexport const DEFAULT_SAFETY_FAILOVER_CANDIDATES: readonly FailoverCandidate[] = [\n\t{ provider: \"kimi-coding\", id: \"k3\" },\n\t{ provider: \"modelstudio-maas\", id: \"qwen3.8-max-preview\" },\n\t{ provider: \"xai\", id: \"grok-4.5\" },\n\t{ provider: \"deepseek\", id: \"deepseek-v4-pro\" },\n\t{ provider: \"deepseek\", id: \"deepseek-v4-flash\" },\n\t{ provider: \"modelstudio-maas\", id: \"deepseek-v4-pro\" },\n\t{ provider: \"kimi-coding\", id: \"kimi-for-coding\" },\n] as const;\n\nexport type ProviderResilienceSettings = Readonly<{\n\t/** When true (default), refuse selecting sticky safety models (e.g. claude-fable-5). */\n\treadonly blockStickySafetyModels?: boolean;\n\t/** When true (default), auto-switch model on content/safety stop before retry. */\n\treadonly autoFailoverOnSafetyStop?: boolean;\n\t/** Ordered failover targets. Falls back to DEFAULT_SAFETY_FAILOVER_CANDIDATES. */\n\treadonly failoverCandidates?: readonly FailoverCandidate[];\n}>;\n\nexport const DEFAULT_PROVIDER_RESILIENCE: Required<\n\tPick<ProviderResilienceSettings, \"blockStickySafetyModels\" | \"autoFailoverOnSafetyStop\">\n> & {\n\treadonly failoverCandidates: readonly FailoverCandidate[];\n} = {\n\tblockStickySafetyModels: true,\n\tautoFailoverOnSafetyStop: true,\n\tfailoverCandidates: DEFAULT_SAFETY_FAILOVER_CANDIDATES,\n};\n\nexport function isStickySafetyModel(modelId: string | undefined, provider?: string | undefined): boolean {\n\tconst id = (modelId ?? \"\").trim();\n\tif (!id) return false;\n\tif (STICKY_SAFETY_MODEL_RE.test(id)) return true;\n\tconst p = (provider ?? \"\").toLowerCase();\n\treturn p.includes(\"anthropic\") && STICKY_SAFETY_MODEL_RE.test(id);\n}\n\n/** Fable / Opus / Sonnet stay on the requested id after a safety stop. */\nexport function isNoSafetyFailoverModel(modelId: string | undefined, provider?: string | undefined): boolean {\n\tconst id = (modelId ?? \"\").trim();\n\tif (!id) return false;\n\tif (NO_SAFETY_FAILOVER_MODEL_RE.test(id)) return true;\n\tconst p = (provider ?? \"\").toLowerCase();\n\treturn p.includes(\"anthropic\") && NO_SAFETY_FAILOVER_MODEL_RE.test(id);\n}\n\nexport function isContentSafetyStopMessage(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /content\\/safety stop|stop_reason\\s*=\\s*(refusal|sensitive)|safety stop|provider\\.refusal|kind=provider_refusal/i.test(\n\t\ttext,\n\t);\n}\n\n/**\n * Billing/quota exhaustion (\"usage limit for this billing cycle\",\n * \"insufficient_quota\", zero balance, ...). Same-model retry is useless until\n * the cycle resets, but the turn can still be saved by failing over to another\n * candidate — so these classify as retryable-with-failover, not auth-fatal.\n */\nexport function isQuotaExhaustionMessage(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /usage limit|GoUsageLimitError|FreeUsageLimitError|available balance|insufficient_quota|out of budget|quota exceeded|billing/i.test(\n\t\ttext,\n\t);\n}\n\n/** Orphan tool results / Kimi-K3 protocol shape errors that heal after sanitize+retry. */\nexport function isOrphanToolCallIdError(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /tool_call_id\\s+is\\s+not\\s+found|tool_call_id\\s+not\\s+found|unknown\\s+tool_call_id/i.test(text);\n}\n\n/**\n * Errors the agent loop may auto-retry (after optional failover / message sanitize).\n * Kept in one place so agent-session and tests share the same contract.\n */\nexport function isTransientProviderErrorMessage(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /overloaded|provider.?returned.?error|rate.?limit|too many requests|429|500|502|503|504|service.?unavailable|server.?error|internal.?error|network.?error|connection.?error|connection.?refused|connection.?lost|websocket.?closed|websocket.?error|other side closed|fetch failed|upstream.?connect|reset before headers|socket hang up|ended without|stream ended before message_stop|http2 request did not get a response|timed? out|timeout|\\bterminated\\b|retry delay|content\\/safety stop|stop_reason\\s*=\\s*(refusal|sensitive)|safety stop|tool_call_id\\s+is\\s+not\\s+found|tool_call_id\\s+not\\s+found|invalid_request_error|json error injected into sse stream|injected into sse/i.test(\n\t\ttext,\n\t);\n}\n\nexport function resolveFailoverCandidates(\n\tsettings: ProviderResilienceSettings | undefined,\n): readonly FailoverCandidate[] {\n\tconst custom = settings?.failoverCandidates;\n\tif (custom && custom.length > 0) return custom;\n\treturn DEFAULT_SAFETY_FAILOVER_CANDIDATES;\n}\n\n/** Explicit `--model` / `--provider` pins stay on the requested model. */\nexport function shouldHonorSafetyFailover(options: {\n\treadonly autoFailoverOnSafetyStop: boolean;\n\treadonly modelPinned: boolean;\n\treadonly modelId?: string;\n\treadonly provider?: string;\n}): boolean {\n\tif (!options.autoFailoverOnSafetyStop || options.modelPinned) return false;\n\treturn !isNoSafetyFailoverModel(options.modelId, options.provider);\n}\n\n/** Sticky-model eject still applies unless the user pinned that model. */\nexport function shouldEjectStickySafetyModel(options: {\n\treadonly blockStickySafetyModels: boolean;\n\treadonly modelPinned: boolean;\n}): boolean {\n\treturn options.blockStickySafetyModels && !options.modelPinned;\n}\n\nexport function resolveProviderResilience(settings: ProviderResilienceSettings | undefined): {\n\treadonly blockStickySafetyModels: boolean;\n\treadonly autoFailoverOnSafetyStop: boolean;\n\treadonly failoverCandidates: readonly FailoverCandidate[];\n} {\n\treturn {\n\t\tblockStickySafetyModels: settings?.blockStickySafetyModels ?? DEFAULT_PROVIDER_RESILIENCE.blockStickySafetyModels,\n\t\tautoFailoverOnSafetyStop:\n\t\t\tsettings?.autoFailoverOnSafetyStop ?? DEFAULT_PROVIDER_RESILIENCE.autoFailoverOnSafetyStop,\n\t\tfailoverCandidates: resolveFailoverCandidates(settings),\n\t};\n}\n\n/**\n * Pick first failover candidate that is not the current model and passes `isAllowed`.\n * Pure — caller performs auth checks and setModel.\n */\nexport function pickFailoverCandidate(\n\tcandidates: readonly FailoverCandidate[],\n\tcurrent: { readonly provider?: string; readonly id?: string } | undefined,\n\tisAllowed: (candidate: FailoverCandidate) => boolean,\n): FailoverCandidate | undefined {\n\tfor (const c of candidates) {\n\t\tif (current && c.provider === current.provider && c.id === current.id) continue;\n\t\tif (isStickySafetyModel(c.id, c.provider)) continue;\n\t\tif (!isAllowed(c)) continue;\n\t\treturn c;\n\t}\n\treturn undefined;\n}\n\nexport function stickySafetyBlockMessage(modelId: string, provider: string): string {\n\treturn (\n\t\t`Blocked sticky safety model ${provider}/${modelId} (providerResilience.blockStickySafetyModels). ` +\n\t\t`Use kimi-coding/k3, modelstudio-maas/qwen3.8-max-preview, grok-4.5, or deepseek; ` +\n\t\t`or set providerResilience.blockStickySafetyModels=false.`\n\t);\n}\n"]}
+{"version":3,"file":"provider-resilience.js","sourceRoot":"","sources":["../../src/core/provider-resilience.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAOH,oFAAoF;AACpF,MAAM,CAAC,MAAM,sBAAsB,GAAG,QAAQ,CAAC;AAE/C,8EAA8E;AAC9E,MAAM,CAAC,MAAM,2BAA2B,GAAG,oCAAoC,CAAC;AAEhF,wEAAwE;AACxE,MAAM,CAAC,MAAM,kCAAkC,GAAiC;IAC/E,EAAE,QAAQ,EAAE,aAAa,EAAE,EAAE,EAAE,IAAI,EAAE;IACrC,EAAE,QAAQ,EAAE,kBAAkB,EAAE,EAAE,EAAE,qBAAqB,EAAE;IAC3D,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE,EAAE,UAAU,EAAE;IACnC,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE,iBAAiB,EAAE;IAC/C,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE,mBAAmB,EAAE;IACjD,EAAE,QAAQ,EAAE,kBAAkB,EAAE,EAAE,EAAE,iBAAiB,EAAE;IACvD,EAAE,QAAQ,EAAE,aAAa,EAAE,EAAE,EAAE,iBAAiB,EAAE;CACzC,CAAC;AAWX,MAAM,CAAC,MAAM,2BAA2B,GAIpC;IACH,uBAAuB,EAAE,IAAI;IAC7B,wBAAwB,EAAE,IAAI;IAC9B,kBAAkB,EAAE,kCAAkC;CACtD,CAAC;AAEF,MAAM,UAAU,mBAAmB,CAAC,OAA2B,EAAE,QAA6B,EAAW;IACxG,MAAM,EAAE,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAClC,IAAI,CAAC,EAAE;QAAE,OAAO,KAAK,CAAC;IACtB,IAAI,sBAAsB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IACjD,MAAM,CAAC,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IACzC,OAAO,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI,sBAAsB,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CAClE;AAED,0EAA0E;AAC1E,MAAM,UAAU,uBAAuB,CAAC,OAA2B,EAAE,QAA6B,EAAW;IAC5G,MAAM,EAAE,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAClC,IAAI,CAAC,EAAE;QAAE,OAAO,KAAK,CAAC;IACtB,IAAI,2BAA2B,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IACtD,MAAM,CAAC,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IACzC,OAAO,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI,2BAA2B,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CACvE;AAED,MAAM,UAAU,0BAA0B,CAAC,IAAwB,EAAW;IAC7E,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,iHAAiH,CAAC,IAAI,CAC5H,IAAI,CACJ,CAAC;AAAA,CACF;AAED;;;;;GAKG;AACH,MAAM,UAAU,wBAAwB,CAAC,IAAwB,EAAW;IAC3E,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,8HAA8H,CAAC,IAAI,CACzI,IAAI,CACJ,CAAC;AAAA,CACF;AAED,0FAA0F;AAC1F,MAAM,UAAU,uBAAuB,CAAC,IAAwB,EAAW;IAC1E,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,oFAAoF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACvG;AAED;;;;;;GAMG;AACH,MAAM,UAAU,4BAA4B,CAAC,IAAwB,EAAW;IAC/E,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,6HAA6H,CAAC,IAAI,CACxI,IAAI,CACJ,CAAC;AAAA,CACF;AAED;;;;;;GAMG;AACH,MAAM,oBAAoB,GAAqE;IAC9F,EAAE,OAAO,EAAE,yCAAyC,EAAE,MAAM,EAAE,UAAU,EAAE;CAC1E,CAAC;AAEF,mFAAmF;AACnF,MAAM,UAAU,gBAAgB,CAAC,KAIhC,EAAsB;IACtB,KAAK,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,oBAAoB,EAAE,CAAC;QACxD,IAAI,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAAE,OAAO,MAAM,CAAC;QAC1C,IAAI,KAAK,CAAC,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;YAAE,OAAO,MAAM,CAAC;IAC3D,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED;;;;GAIG;AACH,MAAM,UAAU,wBAAwB,CAGvC,MAA8F,EAC9F,SAAuB,EACjB;IACN,MAAM,MAAM,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC;IACxC,IAAI,CAAC,MAAM;QAAE,OAAO,EAAE,CAAC;IACvB,MAAM,UAAU,GAAQ,EAAE,CAAC;IAC3B,KAAK,MAAM,KAAK,IAAI,SAAS,EAAE,CAAC;QAC/B,IAAI,KAAK,CAAC,QAAQ,KAAK,MAAM,CAAC,QAAQ,IAAI,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE;YAAE,SAAS;QAC3E,IAAI,gBAAgB,CAAC,KAAK,CAAC,KAAK,MAAM;YAAE,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,UAAU,CAAC;AAAA,CAClB;AAED;;;GAGG;AACH,MAAM,UAAU,+BAA+B,CAAC,IAAwB,EAAW;IAClF,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,0pBAA0pB,CAAC,IAAI,CACrqB,IAAI,CACJ,CAAC;AAAA,CACF;AAED,MAAM,UAAU,yBAAyB,CACxC,QAAgD,EACjB;IAC/B,MAAM,MAAM,GAAG,QAAQ,EAAE,kBAAkB,CAAC;IAC5C,IAAI,MAAM,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,MAAM,CAAC;IAC/C,OAAO,kCAAkC,CAAC;AAAA,CAC1C;AAED,0EAA0E;AAC1E,MAAM,UAAU,yBAAyB,CAAC,OAKzC,EAAW;IACX,IAAI,CAAC,OAAO,CAAC,wBAAwB,IAAI,OAAO,CAAC,WAAW;QAAE,OAAO,KAAK,CAAC;IAC3E,OAAO,CAAC,uBAAuB,CAAC,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;AAAA,CACnE;AAED,0EAA0E;AAC1E,MAAM,UAAU,4BAA4B,CAAC,OAG5C,EAAW;IACX,OAAO,OAAO,CAAC,uBAAuB,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC;AAAA,CAC/D;AAED,MAAM,UAAU,yBAAyB,CAAC,QAAgD,EAIxF;IACD,OAAO;QACN,uBAAuB,EAAE,QAAQ,EAAE,uBAAuB,IAAI,2BAA2B,CAAC,uBAAuB;QACjH,wBAAwB,EACvB,QAAQ,EAAE,wBAAwB,IAAI,2BAA2B,CAAC,wBAAwB;QAC3F,kBAAkB,EAAE,yBAAyB,CAAC,QAAQ,CAAC;KACvD,CAAC;AAAA,CACF;AAED;;;GAGG;AACH,MAAM,UAAU,qBAAqB,CACpC,UAAwC,EACxC,OAAyE,EACzE,SAAoD,EACpB;IAChC,KAAK,MAAM,CAAC,IAAI,UAAU,EAAE,CAAC;QAC5B,IAAI,OAAO,IAAI,CAAC,CAAC,QAAQ,KAAK,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC,EAAE,KAAK,OAAO,CAAC,EAAE;YAAE,SAAS;QAChF,IAAI,mBAAmB,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC;YAAE,SAAS;QACpD,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC;YAAE,SAAS;QAC5B,OAAO,CAAC,CAAC;IACV,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,MAAM,UAAU,wBAAwB,CAAC,OAAe,EAAE,QAAgB,EAAU;IACnF,OAAO,CACN,+BAA+B,QAAQ,IAAI,OAAO,iDAAiD;QACnG,mFAAmF;QACnF,0DAA0D,CAC1D,CAAC;AAAA,CACF","sourcesContent":["/**\n * Provider resilience — root-level recovery for sticky safety models and\n * sanitize-and-retry protocol failures (orphan tool_call_id, terminated streams).\n *\n * This is not a jailbreak layer. It keeps sessions alive when a provider\n * false-positives or returns a sticky transcript-shape error.\n */\n\nexport type FailoverCandidate = Readonly<{\n\treadonly provider: string;\n\treadonly id: string;\n}>;\n\n/** Models known to emit high false-positive content/safety stops on coding work. */\nexport const STICKY_SAFETY_MODEL_RE = /fable/i;\n\n/** Claude lines that must stay on-model when they refuse. No DeepSeek hop. */\nexport const NO_SAFETY_FAILOVER_MODEL_RE = /(?:claude-)?(?:fable|opus|sonnet)/i;\n\n/** Default failover chain when a sticky safety model refuses a turn. */\nexport const DEFAULT_SAFETY_FAILOVER_CANDIDATES: readonly FailoverCandidate[] = [\n\t{ provider: \"kimi-coding\", id: \"k3\" },\n\t{ provider: \"modelstudio-maas\", id: \"qwen3.8-max-preview\" },\n\t{ provider: \"xai\", id: \"grok-4.5\" },\n\t{ provider: \"deepseek\", id: \"deepseek-v4-pro\" },\n\t{ provider: \"deepseek\", id: \"deepseek-v4-flash\" },\n\t{ provider: \"modelstudio-maas\", id: \"deepseek-v4-pro\" },\n\t{ provider: \"kimi-coding\", id: \"kimi-for-coding\" },\n] as const;\n\nexport type ProviderResilienceSettings = Readonly<{\n\t/** When true (default), refuse selecting sticky safety models (e.g. claude-fable-5). */\n\treadonly blockStickySafetyModels?: boolean;\n\t/** When true (default), auto-switch model on content/safety stop before retry. */\n\treadonly autoFailoverOnSafetyStop?: boolean;\n\t/** Ordered failover targets. Falls back to DEFAULT_SAFETY_FAILOVER_CANDIDATES. */\n\treadonly failoverCandidates?: readonly FailoverCandidate[];\n}>;\n\nexport const DEFAULT_PROVIDER_RESILIENCE: Required<\n\tPick<ProviderResilienceSettings, \"blockStickySafetyModels\" | \"autoFailoverOnSafetyStop\">\n> & {\n\treadonly failoverCandidates: readonly FailoverCandidate[];\n} = {\n\tblockStickySafetyModels: true,\n\tautoFailoverOnSafetyStop: true,\n\tfailoverCandidates: DEFAULT_SAFETY_FAILOVER_CANDIDATES,\n};\n\nexport function isStickySafetyModel(modelId: string | undefined, provider?: string | undefined): boolean {\n\tconst id = (modelId ?? \"\").trim();\n\tif (!id) return false;\n\tif (STICKY_SAFETY_MODEL_RE.test(id)) return true;\n\tconst p = (provider ?? \"\").toLowerCase();\n\treturn p.includes(\"anthropic\") && STICKY_SAFETY_MODEL_RE.test(id);\n}\n\n/** Fable / Opus / Sonnet stay on the requested id after a safety stop. */\nexport function isNoSafetyFailoverModel(modelId: string | undefined, provider?: string | undefined): boolean {\n\tconst id = (modelId ?? \"\").trim();\n\tif (!id) return false;\n\tif (NO_SAFETY_FAILOVER_MODEL_RE.test(id)) return true;\n\tconst p = (provider ?? \"\").toLowerCase();\n\treturn p.includes(\"anthropic\") && NO_SAFETY_FAILOVER_MODEL_RE.test(id);\n}\n\nexport function isContentSafetyStopMessage(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /content\\/safety stop|stop_reason\\s*=\\s*(refusal|sensitive)|safety stop|provider\\.refusal|kind=provider_refusal/i.test(\n\t\ttext,\n\t);\n}\n\n/**\n * Billing/quota exhaustion (\"usage limit for this billing cycle\",\n * \"insufficient_quota\", zero balance, ...). Same-model retry is useless until\n * the cycle resets, but the turn can still be saved by failing over to another\n * candidate — so these classify as retryable-with-failover, not auth-fatal.\n */\nexport function isQuotaExhaustionMessage(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /usage limit|GoUsageLimitError|FreeUsageLimitError|available balance|insufficient_quota|out of budget|quota exceeded|billing/i.test(\n\t\ttext,\n\t);\n}\n\n/** Orphan tool results / Kimi-K3 protocol shape errors that heal after sanitize+retry. */\nexport function isOrphanToolCallIdError(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /tool_call_id\\s+is\\s+not\\s+found|tool_call_id\\s+not\\s+found|unknown\\s+tool_call_id/i.test(text);\n}\n\n/**\n * Transient upstream availability failures: gateway 5xx passthroughs\n * (\"503 Upstream request failed\", \"Endpoint is unavailable\") and dropped\n * streams that never carried a finish reason (\"Stream ended without\n * finish_reason\"). Transport problems, not transcript-shape problems — they\n * heal by backoff/retry or switching model, never by transcript sanitization.\n */\nexport function isUpstreamUnavailableMessage(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /upstream(?: request)? failed|upstream connect|endpoint (?:is )?unavailable|stream ended without finish_reason|\\b50[0234]\\b/i.test(\n\t\ttext,\n\t);\n}\n\n/**\n * Route families: the same underlying model served by several provider\n * routes. Matched on the model id OR display name so alias routes like\n * `openrouter/stealth/ox-alpha`, `opencode-go/ox-alpha-free`, and\n * `opencode/x-preview-f-free` (\"Ox Alpha Free (Unlimited)\") rotate as one\n * ox-alpha family.\n */\nconst MODEL_ROUTE_FAMILIES: readonly { readonly pattern: RegExp; readonly family: string }[] = [\n\t{ pattern: /(?:^|[-_/])ox[-_ ]?alpha(?:[-_/\\s(]|$)/i, family: \"ox-alpha\" },\n];\n\n/** Family key for a model route, or undefined when it has no known alias group. */\nexport function modelRouteFamily(model: {\n\treadonly provider?: string | undefined;\n\treadonly id: string;\n\treadonly name?: string | undefined;\n}): string | undefined {\n\tfor (const { pattern, family } of MODEL_ROUTE_FAMILIES) {\n\t\tif (pattern.test(model.id)) return family;\n\t\tif (model.name && pattern.test(model.name)) return family;\n\t}\n\treturn undefined;\n}\n\n/**\n * Alternative provider routes serving the same model family as\n * {@link failed}, in registry order, excluding the failed route itself.\n * Callers filter by authentication and per-turn failure bookkeeping.\n */\nexport function sameModelRouteCandidates<\n\tT extends { readonly provider: string; readonly id: string; readonly name?: string | undefined },\n>(\n\tfailed: { readonly provider: string; readonly id: string; readonly name?: string | undefined },\n\tavailable: readonly T[],\n): T[] {\n\tconst family = modelRouteFamily(failed);\n\tif (!family) return [];\n\tconst candidates: T[] = [];\n\tfor (const model of available) {\n\t\tif (model.provider === failed.provider && model.id === failed.id) continue;\n\t\tif (modelRouteFamily(model) === family) candidates.push(model);\n\t}\n\treturn candidates;\n}\n\n/**\n * Errors the agent loop may auto-retry (after optional failover / message sanitize).\n * Kept in one place so agent-session and tests share the same contract.\n */\nexport function isTransientProviderErrorMessage(text: string | undefined): boolean {\n\tif (!text) return false;\n\treturn /overloaded|provider.?returned.?error|rate.?limit|too many requests|429|500|502|503|504|service.?unavailable|server.?error|internal.?error|network.?error|connection.?error|connection.?refused|connection.?lost|websocket.?closed|websocket.?error|other side closed|fetch failed|upstream.?connect|reset before headers|socket hang up|ended without|stream ended before message_stop|http2 request did not get a response|timed? out|timeout|\\bterminated\\b|retry delay|content\\/safety stop|stop_reason\\s*=\\s*(refusal|sensitive)|safety stop|tool_call_id\\s+is\\s+not\\s+found|tool_call_id\\s+not\\s+found|invalid_request_error|json error injected into sse stream|injected into sse/i.test(\n\t\ttext,\n\t);\n}\n\nexport function resolveFailoverCandidates(\n\tsettings: ProviderResilienceSettings | undefined,\n): readonly FailoverCandidate[] {\n\tconst custom = settings?.failoverCandidates;\n\tif (custom && custom.length > 0) return custom;\n\treturn DEFAULT_SAFETY_FAILOVER_CANDIDATES;\n}\n\n/** Explicit `--model` / `--provider` pins stay on the requested model. */\nexport function shouldHonorSafetyFailover(options: {\n\treadonly autoFailoverOnSafetyStop: boolean;\n\treadonly modelPinned: boolean;\n\treadonly modelId?: string;\n\treadonly provider?: string;\n}): boolean {\n\tif (!options.autoFailoverOnSafetyStop || options.modelPinned) return false;\n\treturn !isNoSafetyFailoverModel(options.modelId, options.provider);\n}\n\n/** Sticky-model eject still applies unless the user pinned that model. */\nexport function shouldEjectStickySafetyModel(options: {\n\treadonly blockStickySafetyModels: boolean;\n\treadonly modelPinned: boolean;\n}): boolean {\n\treturn options.blockStickySafetyModels && !options.modelPinned;\n}\n\nexport function resolveProviderResilience(settings: ProviderResilienceSettings | undefined): {\n\treadonly blockStickySafetyModels: boolean;\n\treadonly autoFailoverOnSafetyStop: boolean;\n\treadonly failoverCandidates: readonly FailoverCandidate[];\n} {\n\treturn {\n\t\tblockStickySafetyModels: settings?.blockStickySafetyModels ?? DEFAULT_PROVIDER_RESILIENCE.blockStickySafetyModels,\n\t\tautoFailoverOnSafetyStop:\n\t\t\tsettings?.autoFailoverOnSafetyStop ?? DEFAULT_PROVIDER_RESILIENCE.autoFailoverOnSafetyStop,\n\t\tfailoverCandidates: resolveFailoverCandidates(settings),\n\t};\n}\n\n/**\n * Pick first failover candidate that is not the current model and passes `isAllowed`.\n * Pure — caller performs auth checks and setModel.\n */\nexport function pickFailoverCandidate(\n\tcandidates: readonly FailoverCandidate[],\n\tcurrent: { readonly provider?: string; readonly id?: string } | undefined,\n\tisAllowed: (candidate: FailoverCandidate) => boolean,\n): FailoverCandidate | undefined {\n\tfor (const c of candidates) {\n\t\tif (current && c.provider === current.provider && c.id === current.id) continue;\n\t\tif (isStickySafetyModel(c.id, c.provider)) continue;\n\t\tif (!isAllowed(c)) continue;\n\t\treturn c;\n\t}\n\treturn undefined;\n}\n\nexport function stickySafetyBlockMessage(modelId: string, provider: string): string {\n\treturn (\n\t\t`Blocked sticky safety model ${provider}/${modelId} (providerResilience.blockStickySafetyModels). ` +\n\t\t`Use kimi-coding/k3, modelstudio-maas/qwen3.8-max-preview, grok-4.5, or deepseek; ` +\n\t\t`or set providerResilience.blockStickySafetyModels=false.`\n\t);\n}\n"]}

package/dist/core/provider-retry.d.ts

@@ -7,7 +7,7 @@ import type { AssistantMessage } from "omk-ai";
 /**
  * Empty streamed completion — stop=stop but zero usable output (no text,
  * thinking, or toolCall). Live signature of relay first-token timeouts
- * (~181s silent kill on thinking-heavy ox-alpha turns, 2026-08-22 log evidence).
+ * killing long silent thinking turns.
  * These are transport failures wearing a success shape.
  */
 export declare function isEmptyStreamedCompletion(message: AssistantMessage): boolean;

package/dist/core/provider-retry.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"provider-retry.d.ts","sourceRoot":"","sources":["../../src/core/provider-retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,QAAQ,CAAC;AAQ/C;;;;GAIG;AACH;;;;;GAKG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAW5E;AAED,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,EAAE,MAAM,GAAG,OAAO,CAQnG;AAED;;;;GAIG;AACH,wBAAgB,4BAA4B,CAAC,OAAO,EAAE,gBAAgB,EAAE,oBAAoB,EAAE,MAAM,GAAG,MAAM,CAG5G;AAED,wBAAgB,gBAAgB,CAAC,KAAK,EAAE;IACvC,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;IAC1B,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC;IACnC,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;CAC5B,GAAG,MAAM,GAAG,SAAS,CAIrB;AAED,iFAAiF;AACjF,wBAAgB,mBAAmB,CAAC,WAAW,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,gBAAgB,EAAE,OAAO,GAAG,MAAM,CAE3G;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAEhF;AAED,iEAAiE;AACjE,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,GAAG,MAAM,CAErE","sourcesContent":["import type { AssistantMessage } from \"omk-ai\";\nimport { isContextOverflow } from \"omk-ai\";\nimport {\n\tisContentSafetyStopMessage,\n\tisQuotaExhaustionMessage,\n\tisTransientProviderErrorMessage,\n} from \"./provider-resilience.ts\";\n\n/**\n * Pure retry/failover decisions for agent-turn provider errors.\n * Context overflow is never retryable here — compaction owns that path.\n * Quota exhaustion is retryable so the failover chain can save the turn.\n */\n/**\n * Empty streamed completion — stop=stop but zero usable output (no text,\n * thinking, or toolCall). Live signature of relay first-token timeouts\n * (~181s silent kill on thinking-heavy ox-alpha turns, 2026-08-22 log evidence).\n * These are transport failures wearing a success shape.\n */\nexport function isEmptyStreamedCompletion(message: AssistantMessage): boolean {\n\tif (message.stopReason !== \"stop\") return false;\n\tconst content: unknown = message.content;\n\tif (!Array.isArray(content)) {\n\t\treturn typeof content === \"string\" ? content.trim().length === 0 : true;\n\t}\n\treturn content.every((block) => {\n\t\tconst t = (block as { type?: string } | null)?.type;\n\t\tif (t !== \"text\") return false; // thinking/toolCall/image = real output\n\t\treturn !String((block as { text?: string }).text ?? \"\").trim();\n\t});\n}\n\nexport function isRetryableAssistantError(message: AssistantMessage, contextWindow: number): boolean {\n\t// An empty streamed completion is a dead stream, not an answer — retry it\n\t// regardless of its success-shaped stopReason (bounded by maxRetries).\n\tif (isEmptyStreamedCompletion(message)) return true;\n\tif (message.stopReason !== \"error\" || !message.errorMessage) return false;\n\tif (isContextOverflow(message, contextWindow)) return false;\n\tif (isQuotaExhaustionMessage(message.errorMessage)) return true;\n\treturn isTransientProviderErrorMessage(message.errorMessage);\n}\n\n/**\n * Next attempt number within the retry budget, or undefined when retry is\n * disabled or the budget is exhausted. The caller's completed-attempt count\n * stays unchanged when undefined is returned.\n */\nexport function retryBudgetForAssistantError(message: AssistantMessage, configuredMaxRetries: number): number {\n\tif (configuredMaxRetries <= 0) return 0;\n\treturn isContentSafetyStopMessage(message.errorMessage) ? Math.min(1, configuredMaxRetries) : configuredMaxRetries;\n}\n\nexport function nextRetryAttempt(input: {\n\treadonly enabled: boolean;\n\treadonly completedAttempts: number;\n\treadonly maxRetries: number;\n}): number | undefined {\n\tif (!input.enabled) return undefined;\n\tconst attempt = input.completedAttempts + 1;\n\treturn attempt > input.maxRetries ? undefined : attempt;\n}\n\n/** Same-model retry backs off exponentially; a failed-over retry starts fast. */\nexport function computeRetryDelayMs(baseDelayMs: number, attempt: number, failoverOccurred: boolean): number {\n\treturn failoverOccurred ? Math.min(400, baseDelayMs) : baseDelayMs * 2 ** (attempt - 1);\n}\n\n/**\n * Errors that justify an immediate model switch instead of a same-model retry:\n * safety-stop false positives and billing/quota exhaustion. Both mean the\n * current model cannot finish this turn.\n */\nexport function isFailoverTriggerError(errorMessage: string | undefined): boolean {\n\treturn isContentSafetyStopMessage(errorMessage) || isQuotaExhaustionMessage(errorMessage);\n}\n\n/** Bookkeeping key for the per-turn refused/failed model set. */\nexport function failoverModelKey(provider: string, id: string): string {\n\treturn `${provider}/${id}`;\n}\n"]}
+{"version":3,"file":"provider-retry.d.ts","sourceRoot":"","sources":["../../src/core/provider-retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,QAAQ,CAAC;AAQ/C;;;;GAIG;AACH;;;;;GAKG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAW5E;AAED,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,EAAE,MAAM,GAAG,OAAO,CAQnG;AAED;;;;GAIG;AACH,wBAAgB,4BAA4B,CAAC,OAAO,EAAE,gBAAgB,EAAE,oBAAoB,EAAE,MAAM,GAAG,MAAM,CAG5G;AAED,wBAAgB,gBAAgB,CAAC,KAAK,EAAE;IACvC,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;IAC1B,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC;IACnC,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;CAC5B,GAAG,MAAM,GAAG,SAAS,CAIrB;AAED,iFAAiF;AACjF,wBAAgB,mBAAmB,CAAC,WAAW,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,gBAAgB,EAAE,OAAO,GAAG,MAAM,CAE3G;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAEhF;AAED,iEAAiE;AACjE,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,GAAG,MAAM,CAErE","sourcesContent":["import type { AssistantMessage } from \"omk-ai\";\nimport { isContextOverflow } from \"omk-ai\";\nimport {\n\tisContentSafetyStopMessage,\n\tisQuotaExhaustionMessage,\n\tisTransientProviderErrorMessage,\n} from \"./provider-resilience.ts\";\n\n/**\n * Pure retry/failover decisions for agent-turn provider errors.\n * Context overflow is never retryable here — compaction owns that path.\n * Quota exhaustion is retryable so the failover chain can save the turn.\n */\n/**\n * Empty streamed completion — stop=stop but zero usable output (no text,\n * thinking, or toolCall). Live signature of relay first-token timeouts\n * killing long silent thinking turns.\n * These are transport failures wearing a success shape.\n */\nexport function isEmptyStreamedCompletion(message: AssistantMessage): boolean {\n\tif (message.stopReason !== \"stop\") return false;\n\tconst content: unknown = message.content;\n\tif (!Array.isArray(content)) {\n\t\treturn typeof content === \"string\" ? content.trim().length === 0 : true;\n\t}\n\treturn content.every((block) => {\n\t\tconst t = (block as { type?: string } | null)?.type;\n\t\tif (t !== \"text\") return false; // thinking/toolCall/image = real output\n\t\treturn !String((block as { text?: string }).text ?? \"\").trim();\n\t});\n}\n\nexport function isRetryableAssistantError(message: AssistantMessage, contextWindow: number): boolean {\n\t// An empty streamed completion is a dead stream, not an answer — retry it\n\t// regardless of its success-shaped stopReason (bounded by maxRetries).\n\tif (isEmptyStreamedCompletion(message)) return true;\n\tif (message.stopReason !== \"error\" || !message.errorMessage) return false;\n\tif (isContextOverflow(message, contextWindow)) return false;\n\tif (isQuotaExhaustionMessage(message.errorMessage)) return true;\n\treturn isTransientProviderErrorMessage(message.errorMessage);\n}\n\n/**\n * Next attempt number within the retry budget, or undefined when retry is\n * disabled or the budget is exhausted. The caller's completed-attempt count\n * stays unchanged when undefined is returned.\n */\nexport function retryBudgetForAssistantError(message: AssistantMessage, configuredMaxRetries: number): number {\n\tif (configuredMaxRetries <= 0) return 0;\n\treturn isContentSafetyStopMessage(message.errorMessage) ? Math.min(1, configuredMaxRetries) : configuredMaxRetries;\n}\n\nexport function nextRetryAttempt(input: {\n\treadonly enabled: boolean;\n\treadonly completedAttempts: number;\n\treadonly maxRetries: number;\n}): number | undefined {\n\tif (!input.enabled) return undefined;\n\tconst attempt = input.completedAttempts + 1;\n\treturn attempt > input.maxRetries ? undefined : attempt;\n}\n\n/** Same-model retry backs off exponentially; a failed-over retry starts fast. */\nexport function computeRetryDelayMs(baseDelayMs: number, attempt: number, failoverOccurred: boolean): number {\n\treturn failoverOccurred ? Math.min(400, baseDelayMs) : baseDelayMs * 2 ** (attempt - 1);\n}\n\n/**\n * Errors that justify an immediate model switch instead of a same-model retry:\n * safety-stop false positives and billing/quota exhaustion. Both mean the\n * current model cannot finish this turn.\n */\nexport function isFailoverTriggerError(errorMessage: string | undefined): boolean {\n\treturn isContentSafetyStopMessage(errorMessage) || isQuotaExhaustionMessage(errorMessage);\n}\n\n/** Bookkeeping key for the per-turn refused/failed model set. */\nexport function failoverModelKey(provider: string, id: string): string {\n\treturn `${provider}/${id}`;\n}\n"]}

package/dist/core/provider-retry.js

@@ -8,7 +8,7 @@ import { isContentSafetyStopMessage, isQuotaExhaustionMessage, isTransientProvid
 /**
  * Empty streamed completion — stop=stop but zero usable output (no text,
  * thinking, or toolCall). Live signature of relay first-token timeouts
- * (~181s silent kill on thinking-heavy ox-alpha turns, 2026-08-22 log evidence).
+ * killing long silent thinking turns.
  * These are transport failures wearing a success shape.
  */
 export function isEmptyStreamedCompletion(message) {

package/dist/core/provider-retry.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider-retry.js","sourceRoot":"","sources":["../../src/core/provider-retry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iBAAiB,EAAE,MAAM,QAAQ,CAAC;AAC3C,OAAO,EACN,0BAA0B,EAC1B,wBAAwB,EACxB,+BAA+B,GAC/B,MAAM,0BAA0B,CAAC;AAElC;;;;GAIG;AACH;;;;;GAKG;AACH,MAAM,UAAU,yBAAyB,CAAC,OAAyB,EAAW;IAC7E,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM;QAAE,OAAO,KAAK,CAAC;IAChD,MAAM,OAAO,GAAY,OAAO,CAAC,OAAO,CAAC;IACzC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QAC7B,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACzE,CAAC;IACD,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;QAC/B,MAAM,CAAC,GAAI,KAAkC,EAAE,IAAI,CAAC;QACpD,IAAI,CAAC,KAAK,MAAM;YAAE,OAAO,KAAK,CAAC,CAAC,wCAAwC;QACxE,OAAO,CAAC,MAAM,CAAE,KAA2B,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAAA,CAC/D,CAAC,CAAC;AAAA,CACH;AAED,MAAM,UAAU,yBAAyB,CAAC,OAAyB,EAAE,aAAqB,EAAW;IACpG,4EAA0E;IAC1E,uEAAuE;IACvE,IAAI,yBAAyB,CAAC,OAAO,CAAC;QAAE,OAAO,IAAI,CAAC;IACpD,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,YAAY;QAAE,OAAO,KAAK,CAAC;IAC1E,IAAI,iBAAiB,CAAC,OAAO,EAAE,aAAa,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5D,IAAI,wBAAwB,CAAC,OAAO,CAAC,YAAY,CAAC;QAAE,OAAO,IAAI,CAAC;IAChE,OAAO,+BAA+B,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC;AAAA,CAC7D;AAED;;;;GAIG;AACH,MAAM,UAAU,4BAA4B,CAAC,OAAyB,EAAE,oBAA4B,EAAU;IAC7G,IAAI,oBAAoB,IAAI,CAAC;QAAE,OAAO,CAAC,CAAC;IACxC,OAAO,0BAA0B,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,oBAAoB,CAAC,CAAC,CAAC,CAAC,oBAAoB,CAAC;AAAA,CACnH;AAED,MAAM,UAAU,gBAAgB,CAAC,KAIhC,EAAsB;IACtB,IAAI,CAAC,KAAK,CAAC,OAAO;QAAE,OAAO,SAAS,CAAC;IACrC,MAAM,OAAO,GAAG,KAAK,CAAC,iBAAiB,GAAG,CAAC,CAAC;IAC5C,OAAO,OAAO,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;AAAA,CACxD;AAED,iFAAiF;AACjF,MAAM,UAAU,mBAAmB,CAAC,WAAmB,EAAE,OAAe,EAAE,gBAAyB,EAAU;IAC5G,OAAO,gBAAgB,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,WAAW,CAAC,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;AAAA,CACxF;AAED;;;;GAIG;AACH,MAAM,UAAU,sBAAsB,CAAC,YAAgC,EAAW;IACjF,OAAO,0BAA0B,CAAC,YAAY,CAAC,IAAI,wBAAwB,CAAC,YAAY,CAAC,CAAC;AAAA,CAC1F;AAED,iEAAiE;AACjE,MAAM,UAAU,gBAAgB,CAAC,QAAgB,EAAE,EAAU,EAAU;IACtE,OAAO,GAAG,QAAQ,IAAI,EAAE,EAAE,CAAC;AAAA,CAC3B","sourcesContent":["import type { AssistantMessage } from \"omk-ai\";\nimport { isContextOverflow } from \"omk-ai\";\nimport {\n\tisContentSafetyStopMessage,\n\tisQuotaExhaustionMessage,\n\tisTransientProviderErrorMessage,\n} from \"./provider-resilience.ts\";\n\n/**\n * Pure retry/failover decisions for agent-turn provider errors.\n * Context overflow is never retryable here — compaction owns that path.\n * Quota exhaustion is retryable so the failover chain can save the turn.\n */\n/**\n * Empty streamed completion — stop=stop but zero usable output (no text,\n * thinking, or toolCall). Live signature of relay first-token timeouts\n * (~181s silent kill on thinking-heavy ox-alpha turns, 2026-08-22 log evidence).\n * These are transport failures wearing a success shape.\n */\nexport function isEmptyStreamedCompletion(message: AssistantMessage): boolean {\n\tif (message.stopReason !== \"stop\") return false;\n\tconst content: unknown = message.content;\n\tif (!Array.isArray(content)) {\n\t\treturn typeof content === \"string\" ? content.trim().length === 0 : true;\n\t}\n\treturn content.every((block) => {\n\t\tconst t = (block as { type?: string } | null)?.type;\n\t\tif (t !== \"text\") return false; // thinking/toolCall/image = real output\n\t\treturn !String((block as { text?: string }).text ?? \"\").trim();\n\t});\n}\n\nexport function isRetryableAssistantError(message: AssistantMessage, contextWindow: number): boolean {\n\t// An empty streamed completion is a dead stream, not an answer — retry it\n\t// regardless of its success-shaped stopReason (bounded by maxRetries).\n\tif (isEmptyStreamedCompletion(message)) return true;\n\tif (message.stopReason !== \"error\" || !message.errorMessage) return false;\n\tif (isContextOverflow(message, contextWindow)) return false;\n\tif (isQuotaExhaustionMessage(message.errorMessage)) return true;\n\treturn isTransientProviderErrorMessage(message.errorMessage);\n}\n\n/**\n * Next attempt number within the retry budget, or undefined when retry is\n * disabled or the budget is exhausted. The caller's completed-attempt count\n * stays unchanged when undefined is returned.\n */\nexport function retryBudgetForAssistantError(message: AssistantMessage, configuredMaxRetries: number): number {\n\tif (configuredMaxRetries <= 0) return 0;\n\treturn isContentSafetyStopMessage(message.errorMessage) ? Math.min(1, configuredMaxRetries) : configuredMaxRetries;\n}\n\nexport function nextRetryAttempt(input: {\n\treadonly enabled: boolean;\n\treadonly completedAttempts: number;\n\treadonly maxRetries: number;\n}): number | undefined {\n\tif (!input.enabled) return undefined;\n\tconst attempt = input.completedAttempts + 1;\n\treturn attempt > input.maxRetries ? undefined : attempt;\n}\n\n/** Same-model retry backs off exponentially; a failed-over retry starts fast. */\nexport function computeRetryDelayMs(baseDelayMs: number, attempt: number, failoverOccurred: boolean): number {\n\treturn failoverOccurred ? Math.min(400, baseDelayMs) : baseDelayMs * 2 ** (attempt - 1);\n}\n\n/**\n * Errors that justify an immediate model switch instead of a same-model retry:\n * safety-stop false positives and billing/quota exhaustion. Both mean the\n * current model cannot finish this turn.\n */\nexport function isFailoverTriggerError(errorMessage: string | undefined): boolean {\n\treturn isContentSafetyStopMessage(errorMessage) || isQuotaExhaustionMessage(errorMessage);\n}\n\n/** Bookkeeping key for the per-turn refused/failed model set. */\nexport function failoverModelKey(provider: string, id: string): string {\n\treturn `${provider}/${id}`;\n}\n"]}
+{"version":3,"file":"provider-retry.js","sourceRoot":"","sources":["../../src/core/provider-retry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iBAAiB,EAAE,MAAM,QAAQ,CAAC;AAC3C,OAAO,EACN,0BAA0B,EAC1B,wBAAwB,EACxB,+BAA+B,GAC/B,MAAM,0BAA0B,CAAC;AAElC;;;;GAIG;AACH;;;;;GAKG;AACH,MAAM,UAAU,yBAAyB,CAAC,OAAyB,EAAW;IAC7E,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM;QAAE,OAAO,KAAK,CAAC;IAChD,MAAM,OAAO,GAAY,OAAO,CAAC,OAAO,CAAC;IACzC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QAC7B,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACzE,CAAC;IACD,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;QAC/B,MAAM,CAAC,GAAI,KAAkC,EAAE,IAAI,CAAC;QACpD,IAAI,CAAC,KAAK,MAAM;YAAE,OAAO,KAAK,CAAC,CAAC,wCAAwC;QACxE,OAAO,CAAC,MAAM,CAAE,KAA2B,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAAA,CAC/D,CAAC,CAAC;AAAA,CACH;AAED,MAAM,UAAU,yBAAyB,CAAC,OAAyB,EAAE,aAAqB,EAAW;IACpG,4EAA0E;IAC1E,uEAAuE;IACvE,IAAI,yBAAyB,CAAC,OAAO,CAAC;QAAE,OAAO,IAAI,CAAC;IACpD,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,YAAY;QAAE,OAAO,KAAK,CAAC;IAC1E,IAAI,iBAAiB,CAAC,OAAO,EAAE,aAAa,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5D,IAAI,wBAAwB,CAAC,OAAO,CAAC,YAAY,CAAC;QAAE,OAAO,IAAI,CAAC;IAChE,OAAO,+BAA+B,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC;AAAA,CAC7D;AAED;;;;GAIG;AACH,MAAM,UAAU,4BAA4B,CAAC,OAAyB,EAAE,oBAA4B,EAAU;IAC7G,IAAI,oBAAoB,IAAI,CAAC;QAAE,OAAO,CAAC,CAAC;IACxC,OAAO,0BAA0B,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,oBAAoB,CAAC,CAAC,CAAC,CAAC,oBAAoB,CAAC;AAAA,CACnH;AAED,MAAM,UAAU,gBAAgB,CAAC,KAIhC,EAAsB;IACtB,IAAI,CAAC,KAAK,CAAC,OAAO;QAAE,OAAO,SAAS,CAAC;IACrC,MAAM,OAAO,GAAG,KAAK,CAAC,iBAAiB,GAAG,CAAC,CAAC;IAC5C,OAAO,OAAO,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;AAAA,CACxD;AAED,iFAAiF;AACjF,MAAM,UAAU,mBAAmB,CAAC,WAAmB,EAAE,OAAe,EAAE,gBAAyB,EAAU;IAC5G,OAAO,gBAAgB,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,WAAW,CAAC,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;AAAA,CACxF;AAED;;;;GAIG;AACH,MAAM,UAAU,sBAAsB,CAAC,YAAgC,EAAW;IACjF,OAAO,0BAA0B,CAAC,YAAY,CAAC,IAAI,wBAAwB,CAAC,YAAY,CAAC,CAAC;AAAA,CAC1F;AAED,iEAAiE;AACjE,MAAM,UAAU,gBAAgB,CAAC,QAAgB,EAAE,EAAU,EAAU;IACtE,OAAO,GAAG,QAAQ,IAAI,EAAE,EAAE,CAAC;AAAA,CAC3B","sourcesContent":["import type { AssistantMessage } from \"omk-ai\";\nimport { isContextOverflow } from \"omk-ai\";\nimport {\n\tisContentSafetyStopMessage,\n\tisQuotaExhaustionMessage,\n\tisTransientProviderErrorMessage,\n} from \"./provider-resilience.ts\";\n\n/**\n * Pure retry/failover decisions for agent-turn provider errors.\n * Context overflow is never retryable here — compaction owns that path.\n * Quota exhaustion is retryable so the failover chain can save the turn.\n */\n/**\n * Empty streamed completion — stop=stop but zero usable output (no text,\n * thinking, or toolCall). Live signature of relay first-token timeouts\n * killing long silent thinking turns.\n * These are transport failures wearing a success shape.\n */\nexport function isEmptyStreamedCompletion(message: AssistantMessage): boolean {\n\tif (message.stopReason !== \"stop\") return false;\n\tconst content: unknown = message.content;\n\tif (!Array.isArray(content)) {\n\t\treturn typeof content === \"string\" ? content.trim().length === 0 : true;\n\t}\n\treturn content.every((block) => {\n\t\tconst t = (block as { type?: string } | null)?.type;\n\t\tif (t !== \"text\") return false; // thinking/toolCall/image = real output\n\t\treturn !String((block as { text?: string }).text ?? \"\").trim();\n\t});\n}\n\nexport function isRetryableAssistantError(message: AssistantMessage, contextWindow: number): boolean {\n\t// An empty streamed completion is a dead stream, not an answer — retry it\n\t// regardless of its success-shaped stopReason (bounded by maxRetries).\n\tif (isEmptyStreamedCompletion(message)) return true;\n\tif (message.stopReason !== \"error\" || !message.errorMessage) return false;\n\tif (isContextOverflow(message, contextWindow)) return false;\n\tif (isQuotaExhaustionMessage(message.errorMessage)) return true;\n\treturn isTransientProviderErrorMessage(message.errorMessage);\n}\n\n/**\n * Next attempt number within the retry budget, or undefined when retry is\n * disabled or the budget is exhausted. The caller's completed-attempt count\n * stays unchanged when undefined is returned.\n */\nexport function retryBudgetForAssistantError(message: AssistantMessage, configuredMaxRetries: number): number {\n\tif (configuredMaxRetries <= 0) return 0;\n\treturn isContentSafetyStopMessage(message.errorMessage) ? Math.min(1, configuredMaxRetries) : configuredMaxRetries;\n}\n\nexport function nextRetryAttempt(input: {\n\treadonly enabled: boolean;\n\treadonly completedAttempts: number;\n\treadonly maxRetries: number;\n}): number | undefined {\n\tif (!input.enabled) return undefined;\n\tconst attempt = input.completedAttempts + 1;\n\treturn attempt > input.maxRetries ? undefined : attempt;\n}\n\n/** Same-model retry backs off exponentially; a failed-over retry starts fast. */\nexport function computeRetryDelayMs(baseDelayMs: number, attempt: number, failoverOccurred: boolean): number {\n\treturn failoverOccurred ? Math.min(400, baseDelayMs) : baseDelayMs * 2 ** (attempt - 1);\n}\n\n/**\n * Errors that justify an immediate model switch instead of a same-model retry:\n * safety-stop false positives and billing/quota exhaustion. Both mean the\n * current model cannot finish this turn.\n */\nexport function isFailoverTriggerError(errorMessage: string | undefined): boolean {\n\treturn isContentSafetyStopMessage(errorMessage) || isQuotaExhaustionMessage(errorMessage);\n}\n\n/** Bookkeeping key for the per-turn refused/failed model set. */\nexport function failoverModelKey(provider: string, id: string): string {\n\treturn `${provider}/${id}`;\n}\n"]}

package/dist/core/provider-usage.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"provider-usage.d.ts","sourceRoot":"","sources":["../../src/core/provider-usage.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,yBAAyB,EAA2B,MAAM,QAAQ,CAAC;AACjF,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AA+CvD,MAAM,MAAM,uBAAuB,GAAG;IACrC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;CAC3B,CAAC;AAEF,MAAM,MAAM,yBAAyB,GAAG;IACvC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,SAAS,uBAAuB,EAAE,CAAC;IACrD,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;CAC1B,CAAC;AAEF,KAAK,gBAAgB,GAAG;IAAE,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC;AACrF,MAAM,MAAM,kBAAkB,GAAG;IAChC,QAAQ,CAAC,QAAQ,CAAC,EAAE,gBAAgB,CAAC;IACrC,QAAQ,CAAC,QAAQ,CAAC,EAAE,gBAAgB,CAAC;CACrC,CAAC;AAKF,KAAK,SAAS,GAAG,OAAO,GAAG,QAAQ,GAAG,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,iBAAiB,GAAG,aAAa,CAAC;AAClG,KAAK,mBAAmB,GAAG;IAAE,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAA;CAAE,CAAC;AAEtF,MAAM,MAAM,uBAAuB,GAAG;IACrC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC;IACzB,QAAQ,CAAC,WAAW,EAAE,SAAS,mBAAmB,EAAE,CAAC;IACrD,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,kBAAkB,CAAC,EAAE,MAAM,CAAC;CACrC,CAAC;AAEF,KAAK,YAAY,GAAG,IAAI,CAAC,YAAY,EAAE,OAAO,GAAG,eAAe,CAAC,CAAC;AAClE,KAAK,SAAS,GAAG,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;AA8D1F,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,GAAG,uBAAuB,GAAG,SAAS,CAW5G;AASD,wBAAgB,4BAA4B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAErE;AAOD,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,YAAY,EAAE,QAAQ,SAAgC,GAAG,OAAO,CAGlH;AAED,8EAA8E;AAC9E,wBAAgB,uCAAuC,CAAC,OAAO,EAAE,YAAY,GAAG,SAAS,MAAM,EAAE,CAUhG;AAED,wBAAsB,qBAAqB,CAC1C,OAAO,EAAE,YAAY,EACrB,SAAS,GAAE,SAAiB,EAC5B,QAAQ,SAAgC,EACxC,aAAa,GAAE,aAA0B,GACvC,OAAO,CAAC,yBAAyB,GAAG,SAAS,CAAC,CAsChD;AAED,wBAAgB,uBAAuB,CACtC,KAAK,EAAE,OAAO,EACd,UAAU,SAAoB,GAC5B,kBAAkB,GAAG,SAAS,CAchC;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAcvG;AAED,wBAAgB,sBAAsB,CACrC,KAAK,EAAE,OAAO,EACd,UAAU,SAAoB,GAC5B,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAuBhD;AAED,sFAAsF;AACtF,MAAM,MAAM,aAAa,GACtB;IAAE,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAA;CAAE,GAC5B;IAAE,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC;IAAC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CAAC;AAChF,MAAM,MAAM,aAAa,GAAG,CAAC,IAAI,EAAE,SAAS,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,KAAK,OAAO,CAAC,aAAa,CAAC,CAAC;AAEnG,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,MAAM,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,CAOpF;AAED,qFAAmF;AACnF,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CActG;AAED,wBAAsB,uBAAuB,CAC5C,KAAK,EAAE,MAAM,EACb,MAAM,GAAE,aAA0B,GAChC,OAAO,CAAC,yBAAyB,CAAC,CAiBpC;AA8BD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAiBrG;AAED,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAapG;AAED,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,yBAAyB,EAAE,KAAK,SAAa,GAAG,IAAI,CAyBrH;AAED,wBAAgB,wBAAwB,CACvC,MAAM,EAAE,MAAM,EACd,QAAQ,EAAE,yBAAyB,EACnC,KAAK,SAAa,GAChB,IAAI,CAgCN","sourcesContent":["import { execFile } from \"node:child_process\";\nimport { createHash } from \"node:crypto\";\nimport type { ProviderRateLimitSnapshot, ProviderRateLimitWindow } from \"omk-ai\";\nimport type { AgentSession } from \"./agent-session.ts\";\n\nconst FIVE_HOUR_SECONDS = 5 * 60 * 60;\nconst SEVEN_DAY_SECONDS = 7 * 24 * 60 * 60;\nconst WINDOW_TOLERANCE_SECONDS = 120;\nconst CODEX_USAGE_URL = \"https://chatgpt.com/backend-api/wham/usage\";\nconst CLAUDE_USAGE_URL = \"https://api.anthropic.com/api/oauth/usage\";\nconst CLAUDE_MESSAGES_URL = \"https://api.anthropic.com/v1/messages\";\nconst KIMI_USAGE_URL = \"https://api.kimi.com/coding/v1/usages\";\nconst GROK_BILLING_URL = \"https://cli-chat-proxy.grok.com/v1/billing?format=credits\";\nconst OPENAI_AUTH_CLAIM = \"https://api.openai.com/auth\";\nconst MAX_USAGE_RESPONSE_BYTES = 1024 * 1024;\nconst MAX_USAGE_LIMITS = 64;\nconst MAX_PASSIVE_ACCOUNTS = 64;\nconst PASSIVE_USAGE_TTL_MS = 6 * 60 * 60 * 1000;\nconst CLAUDE_QUOTA_PROBE_COOLDOWN_MS = 60 * 60 * 1000;\nconst CLAUDE_QUOTA_PROBE_TIMEOUT_MS = 10_000;\nconst QWEN_CLI_TIMEOUT_MS = 15_000;\nconst QWEN_CLI_ARGS = [\"usage\", \"summary\", \"--format\", \"json\"] as const;\nconst QWEN_CONNECT_HINT = \"connect: npm i -g @qwencloud/qwencloud-cli && qwencloud auth login\";\nconst QWEN_CLI_ENV_KEYS = [\n\t\"PATH\",\n\t\"Path\",\n\t\"HOME\",\n\t\"USERPROFILE\",\n\t\"HOMEDRIVE\",\n\t\"HOMEPATH\",\n\t\"APPDATA\",\n\t\"LOCALAPPDATA\",\n\t\"XDG_CONFIG_HOME\",\n\t\"XDG_DATA_HOME\",\n\t\"XDG_CACHE_HOME\",\n\t\"TMPDIR\",\n\t\"TEMP\",\n\t\"TMP\",\n\t\"SystemRoot\",\n\t\"WINDIR\",\n\t\"ComSpec\",\n\t\"PATHEXT\",\n\t\"LANG\",\n\t\"LC_ALL\",\n\t\"LC_CTYPE\",\n\t\"TERM\",\n\t\"SSL_CERT_FILE\",\n\t\"SSL_CERT_DIR\",\n] as const;\n\nexport type SubscriptionUsageWindow = {\n\treadonly label: string;\n\treadonly usedPercent: number;\n\treadonly resetsAt?: number;\n};\n\nexport type SubscriptionUsageSnapshot = {\n\treadonly label: string;\n\treadonly windows: readonly SubscriptionUsageWindow[];\n\treadonly message?: string;\n};\n\ntype CodexUsageWindow = { readonly usedPercent: number; readonly resetsAt?: number };\nexport type CodexUsageSnapshot = {\n\treadonly fiveHour?: CodexUsageWindow;\n\treadonly sevenDay?: CodexUsageWindow;\n};\n\ntype ParsedCodexWindow = CodexUsageWindow & { readonly windowSeconds?: number };\ntype ObservedCodexWindow = { readonly window: ParsedCodexWindow; readonly observedAt: number };\ntype PassiveUsageEntry = { readonly primary?: ObservedCodexWindow; readonly secondary?: ObservedCodexWindow };\ntype UsageKind = \"codex\" | \"claude\" | \"kimi\" | \"zai\" | \"grok\" | \"qwen-token-plan\" | \"unavailable\";\ntype CredentialCandidate = { readonly provider: string; readonly oauthOnly: boolean };\n\nexport type SubscriptionUsageSource = {\n\treadonly label: string;\n\treadonly kind: UsageKind;\n\treadonly credentials: readonly CredentialCandidate[];\n\treadonly ttlMs: number;\n\treadonly unavailableMessage?: string;\n};\n\ntype UsageSession = Pick<AgentSession, \"state\" | \"modelRegistry\">;\ntype FetchLike = (input: string | URL | Request, init?: RequestInit) => Promise<Response>;\ntype FetchJsonResult = { readonly status: number; readonly payload?: unknown };\n\nconst passiveCodexUsage = new Map<string, PassiveUsageEntry>();\nconst passiveClaudeUsage = new Map<string, PassiveUsageEntry>();\nconst claudeQuotaProbeAttempts = new Map<string, number>();\nconst claudeQuotaProbesInFlight = new Map<string, Promise<void>>();\nconst subscriptionUsageRevisions = new Map<string, number>();\nconst MINUTE_TTL_MS = 60_000;\nconst VISIBLE_USAGE_PROVIDERS = [\n\t\"openai-codex\",\n\t\"anthropic\",\n\t\"kimi-coding\",\n\t\"zai\",\n\t\"modelstudio-maas\",\n\t\"qwen-oauth\",\n\t\"xai\",\n] as const;\nconst SOURCES: Readonly<Record<string, SubscriptionUsageSource>> = {\n\t\"openai-codex\": source(\"CODEX\", \"codex\", [{ provider: \"openai-codex\", oauthOnly: true }]),\n\tanthropic: source(\"CLAUDE\", \"claude\", [{ provider: \"anthropic\", oauthOnly: true }], 5 * MINUTE_TTL_MS),\n\t\"qwen-oauth\": source(\"QWEN\", \"unavailable\", [{ provider: \"qwen-oauth\", oauthOnly: true }]),\n\t// Model Studio exposes no API-key quota endpoint (console gateway is\n\t// session-only), so the token-plan quota rides the official QwenCloud\n\t// management CLI, which holds its own OAuth credential.\n\t\"modelstudio-maas\": source(\n\t\t\"QWEN TOKEN PLAN\",\n\t\t\"qwen-token-plan\",\n\t\t[{ provider: \"modelstudio-maas\", oauthOnly: false }],\n\t\t5 * MINUTE_TTL_MS,\n\t),\n\t\"kimi-code\": source(\"KIMI\", \"kimi\", [\n\t\t{ provider: \"kimi-code\", oauthOnly: true },\n\t\t{ provider: \"kimi-coding\", oauthOnly: false },\n\t]),\n\t\"kimi-coding\": source(\"KIMI\", \"kimi\", [\n\t\t{ provider: \"kimi-code\", oauthOnly: true },\n\t\t{ provider: \"kimi-coding\", oauthOnly: false },\n\t]),\n\t\"zhipu-coding-plan\": source(\"GLM\", \"zai\", [{ provider: \"zhipu-coding-plan\", oauthOnly: true }]),\n\tzai: source(\"GLM\", \"zai\", [\n\t\t{ provider: \"zai\", oauthOnly: false },\n\t\t{ provider: \"zai-coding-cn\", oauthOnly: false },\n\t\t{ provider: \"zhipu-coding-plan\", oauthOnly: true },\n\t]),\n\t\"zai-coding-cn\": source(\"GLM\", \"zai\", [\n\t\t{ provider: \"zai-coding-cn\", oauthOnly: false },\n\t\t{ provider: \"zhipu-coding-plan\", oauthOnly: true },\n\t]),\n\txai: source(\"GROK\", \"grok\", [{ provider: \"xai\", oauthOnly: true }]),\n};\n\nfunction source(\n\tlabel: string,\n\tkind: UsageKind,\n\tcredentials: readonly CredentialCandidate[],\n\tttlMs = MINUTE_TTL_MS,\n\tunavailableMessage?: string,\n): SubscriptionUsageSource {\n\treturn { label, kind, credentials, ttlMs, unavailableMessage };\n}\n\nexport function getSubscriptionUsageSource(provider: string | undefined): SubscriptionUsageSource | undefined {\n\tif (!provider) return undefined;\n\tconst configured = SOURCES[provider];\n\tif (configured) return configured;\n\tif (/^openai-codex-\\d+$/.test(provider)) {\n\t\treturn source(\"CODEX\", \"codex\", [\n\t\t\t{ provider, oauthOnly: true },\n\t\t\t{ provider: \"openai-codex\", oauthOnly: true },\n\t\t]);\n\t}\n\treturn undefined;\n}\n\nfunction usageRevisionKey(provider: string): string {\n\tconst kind = getSubscriptionUsageSource(provider)?.kind;\n\tif (kind === \"codex\") return \"codex\";\n\tif (kind === \"claude\") return \"claude\";\n\treturn provider;\n}\n\nexport function getSubscriptionUsageRevision(provider: string): number {\n\treturn subscriptionUsageRevisions.get(usageRevisionKey(provider)) ?? 0;\n}\n\nfunction bumpSubscriptionUsageRevision(provider: string): void {\n\tconst key = usageRevisionKey(provider);\n\tsubscriptionUsageRevisions.set(key, (subscriptionUsageRevisions.get(key) ?? 0) + 1);\n}\n\nexport function supportsSubscriptionUsage(session: UsageSession, provider = session.state.model?.provider): boolean {\n\tconst usageSource = getSubscriptionUsageSource(provider);\n\treturn usageSource?.credentials.some((candidate) => credentialConfigured(session, candidate)) ?? false;\n}\n\n/** Configured quota groups shown in the rail, with the active group first. */\nexport function getConfiguredSubscriptionUsageProviders(session: UsageSession): readonly string[] {\n\tconst activeProvider = session.state.model?.provider;\n\tconst activeSource = getSubscriptionUsageSource(activeProvider);\n\tconst providers = VISIBLE_USAGE_PROVIDERS.filter((provider) => supportsSubscriptionUsage(session, provider));\n\tif (!activeProvider || !activeSource || !supportsSubscriptionUsage(session, activeProvider)) return providers;\n\tconst activeCanonical = providers.find(\n\t\t(provider) => getSubscriptionUsageSource(provider)?.label === activeSource.label,\n\t);\n\tif (!activeCanonical) return [activeProvider, ...providers];\n\treturn [activeCanonical, ...providers.filter((provider) => provider !== activeCanonical)];\n}\n\nexport async function loadSubscriptionUsage(\n\tsession: UsageSession,\n\tfetchImpl: FetchLike = fetch,\n\tprovider = session.state.model?.provider,\n\tqwenCliRunner: QwenCliRunner = runQwenCli,\n): Promise<SubscriptionUsageSnapshot | undefined> {\n\tconst model = session.state.model;\n\tconst usageSource = getSubscriptionUsageSource(provider);\n\tif (!model || !usageSource || !supportsSubscriptionUsage(session, provider)) return undefined;\n\tif (usageSource.kind === \"unavailable\") {\n\t\treturn {\n\t\t\tlabel: usageSource.label,\n\t\t\twindows: [],\n\t\t\tmessage: usageSource.unavailableMessage ?? \"quota API unavailable\",\n\t\t};\n\t}\n\tif (offline()) return { label: usageSource.label, windows: [], message: \"offline\" };\n\tif (usageSource.kind === \"qwen-token-plan\") {\n\t\t// The QwenCloud CLI authenticates itself; no OMK-held credential is used.\n\t\treturn await fetchQwenTokenPlanUsage(usageSource.label, qwenCliRunner);\n\t}\n\n\tconst credential = await resolveCredential(session, usageSource);\n\tif (!credential) return { label: usageSource.label, windows: [], message: \"usage unavailable\" };\n\n\ttry {\n\t\tswitch (usageSource.kind) {\n\t\t\tcase \"codex\":\n\t\t\t\treturn await fetchCodexUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tcase \"claude\":\n\t\t\t\treturn await fetchClaudeUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tcase \"kimi\":\n\t\t\t\treturn await fetchKimiUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tcase \"zai\":\n\t\t\t\treturn await fetchZaiUsage(usageSource.label, provider ?? model.provider, credential, fetchImpl);\n\t\t\tcase \"grok\":\n\t\t\t\treturn await fetchGrokUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tdefault:\n\t\t\t\treturn { label: usageSource.label, windows: [], message: \"usage unavailable\" };\n\t\t}\n\t} catch {\n\t\treturn { label: usageSource.label, windows: [], message: \"usage unavailable\" };\n\t}\n}\n\nexport function parseCodexUsageSnapshot(\n\tvalue: unknown,\n\tnowSeconds = Date.now() / 1000,\n): CodexUsageSnapshot | undefined {\n\tconst rateLimit = record(record(value)?.rate_limit);\n\tconst primary = codexWindow(rateLimit?.primary_window, nowSeconds);\n\tconst secondary = codexWindow(rateLimit?.secondary_window, nowSeconds);\n\tconst windows = [primary, secondary].filter((window): window is ParsedCodexWindow => window !== undefined);\n\tconst fiveHour = windows.find((window) => near(window.windowSeconds, FIVE_HOUR_SECONDS));\n\tconst sevenDay = windows.find((window) => near(window.windowSeconds, SEVEN_DAY_SECONDS));\n\tconst fallbackFiveHour = fiveHour ?? (primary?.windowSeconds === undefined ? primary : undefined);\n\tconst fallbackSevenDay = sevenDay ?? (secondary?.windowSeconds === undefined ? secondary : undefined);\n\tif (!fallbackFiveHour && !fallbackSevenDay) return undefined;\n\treturn {\n\t\t...(fallbackFiveHour ? { fiveHour: publicCodexWindow(fallbackFiveHour) } : {}),\n\t\t...(fallbackSevenDay ? { sevenDay: publicCodexWindow(fallbackSevenDay) } : {}),\n\t};\n}\n\nexport function parseClaudeUsageSnapshot(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst payload = record(value);\n\tif (!payload) return undefined;\n\tconst entries = Array.isArray(payload.limits)\n\t\t? payload.limits.slice(0, MAX_USAGE_LIMITS).map(record).filter(isDefined)\n\t\t: [];\n\tconst fiveHour = claudeWindow(payload.five_hour) ?? claudeWindow(entries.find((entry) => entry.kind === \"session\"));\n\tconst sevenDay =\n\t\tclaudeWindow(payload.seven_day) ?? claudeWindow(entries.find((entry) => entry.kind === \"weekly_all\"));\n\tconst windows = [\n\t\tfiveHour ? withLabel(\"5H\", fiveHour) : undefined,\n\t\tsevenDay ? withLabel(\"7D\", sevenDay) : undefined,\n\t].filter(isDefined);\n\treturn windows.length > 0 ? windows : undefined;\n}\n\nexport function parseKimiUsageSnapshot(\n\tvalue: unknown,\n\tnowSeconds = Date.now() / 1000,\n): readonly SubscriptionUsageWindow[] | undefined {\n\tconst payload = record(value);\n\tif (!payload) return undefined;\n\tconst windows: SubscriptionUsageWindow[] = [];\n\tconst total = usageRatio(record(payload.usage), nowSeconds);\n\tif (total) windows.push(withLabel(\"TOTAL\", total));\n\tif (Array.isArray(payload.limits)) {\n\t\tfor (const rawLimit of payload.limits.slice(0, MAX_USAGE_LIMITS)) {\n\t\t\tconst limit = record(rawLimit);\n\t\t\tif (!limit) continue;\n\t\t\tconst detail = record(limit.detail) ?? limit;\n\t\t\tconst windowData = record(limit.window);\n\t\t\tconst parsed = usageRatio(detail, nowSeconds);\n\t\t\tif (!parsed) continue;\n\t\t\tconst resetsAt = parseReset(windowData, nowSeconds) ?? parsed.resetsAt;\n\t\t\twindows.push({\n\t\t\t\tlabel: durationLabel(windowData) ?? shortLabel(limit.name ?? limit.title ?? limit.scope, windows.length),\n\t\t\t\tusedPercent: parsed.usedPercent,\n\t\t\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t\t\t});\n\t\t}\n\t}\n\treturn windows.length > 0 ? windows.slice(0, 4) : undefined;\n}\n\n/** Result of one `qwencloud` invocation; `missing` means the CLI is not installed. */\nexport type QwenCliResult =\n\t| { readonly kind: \"missing\" }\n\t| { readonly kind: \"ran\"; readonly exitCode: number; readonly stdout: string };\nexport type QwenCliRunner = (args: readonly string[], timeoutMs: number) => Promise<QwenCliResult>;\n\nexport function buildQwenCliEnvironment(source: NodeJS.ProcessEnv): NodeJS.ProcessEnv {\n\tconst childEnvironment: NodeJS.ProcessEnv = {};\n\tfor (const key of QWEN_CLI_ENV_KEYS) {\n\t\tconst value = source[key];\n\t\tif (value !== undefined) childEnvironment[key] = value;\n\t}\n\treturn childEnvironment;\n}\n\n/** Parse `qwencloud usage summary --format json` → the token-plan 7-day window. */\nexport function parseQwenTokenPlanUsage(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst tokenPlan = record(record(value)?.token_plan);\n\tif (!tokenPlan || tokenPlan.subscribed === false) return undefined;\n\tconst resetsAt = epochSeconds(tokenPlan.resetDate);\n\tlet usedPercent = finiteNumber(tokenPlan.usedPct) ?? finiteNumber(tokenPlan.used_pct);\n\tif (usedPercent === undefined) {\n\t\tconst total = finiteNumber(tokenPlan.totalCredits);\n\t\tconst remaining = finiteNumber(tokenPlan.remainingCredits);\n\t\tif (total !== undefined && remaining !== undefined && total > 0) {\n\t\t\tusedPercent = ((total - remaining) / total) * 100;\n\t\t}\n\t}\n\tif (usedPercent === undefined) return undefined;\n\treturn [{ label: \"7D\", usedPercent: clampPercent(usedPercent), ...(resetsAt === undefined ? {} : { resetsAt }) }];\n}\n\nexport async function fetchQwenTokenPlanUsage(\n\tlabel: string,\n\trunner: QwenCliRunner = runQwenCli,\n): Promise<SubscriptionUsageSnapshot> {\n\tconst unavailable = (message: string): SubscriptionUsageSnapshot => ({ label, windows: [], message });\n\ttry {\n\t\tconst result = await runner(QWEN_CLI_ARGS, QWEN_CLI_TIMEOUT_MS);\n\t\tif (result.kind === \"missing\") return unavailable(QWEN_CONNECT_HINT);\n\t\tif (result.exitCode === 2) return unavailable(\"run: qwencloud auth login\");\n\t\tif (result.exitCode !== 0 || result.stdout.length > MAX_USAGE_RESPONSE_BYTES) {\n\t\t\treturn unavailable(\"usage unavailable\");\n\t\t}\n\t\tconst payload: unknown = JSON.parse(result.stdout);\n\t\tconst tokenPlan = record(record(payload)?.token_plan);\n\t\tif (tokenPlan?.subscribed === false) return unavailable(\"no active token plan\");\n\t\tconst windows = parseQwenTokenPlanUsage(payload);\n\t\treturn windows ? { label, windows } : unavailable(\"usage unavailable\");\n\t} catch {\n\t\treturn unavailable(\"usage unavailable\");\n\t}\n}\n\nfunction runQwenCli(args: readonly string[], timeoutMs: number): Promise<QwenCliResult> {\n\treturn new Promise((resolve) => {\n\t\tconst binary = process.env.QWENCLOUD_CLI ?? \"qwencloud\";\n\t\texecFile(\n\t\t\tbinary,\n\t\t\t[...args],\n\t\t\t{\n\t\t\t\ttimeout: timeoutMs,\n\t\t\t\tmaxBuffer: MAX_USAGE_RESPONSE_BYTES,\n\t\t\t\twindowsHide: true,\n\t\t\t\tenv: buildQwenCliEnvironment(process.env),\n\t\t\t},\n\t\t\t(error, stdout) => {\n\t\t\t\tif (!error) {\n\t\t\t\t\tresolve({ kind: \"ran\", exitCode: 0, stdout });\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst errno = (error as NodeJS.ErrnoException).code;\n\t\t\t\tif (errno === \"ENOENT\") {\n\t\t\t\t\tresolve({ kind: \"missing\" });\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tresolve({ kind: \"ran\", exitCode: typeof errno === \"number\" ? errno : 1, stdout: stdout ?? \"\" });\n\t\t\t},\n\t\t);\n\t});\n}\n\nexport function parseGrokUsageSnapshot(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst config = record(record(value)?.config);\n\tif (!config) return undefined;\n\tconst period = record(config.currentPeriod);\n\tconst resetsAt = epochSeconds(period?.end) ?? epochSeconds(config.billingPeriodEnd);\n\tconst percent = finiteNumber(config.creditUsagePercent);\n\tconst onDemandUsed = finiteNumber(record(config.onDemandUsed)?.val);\n\tconst onDemandCap = finiteNumber(record(config.onDemandCap)?.val);\n\tlet usedPercent = percent;\n\tif (usedPercent === undefined && onDemandUsed !== undefined && onDemandCap !== undefined && onDemandCap > 0) {\n\t\tusedPercent = (onDemandUsed / onDemandCap) * 100;\n\t}\n\tif (usedPercent === undefined && (period || resetsAt !== undefined)) usedPercent = 0;\n\tif (usedPercent === undefined) return undefined;\n\tconst periodType = typeof period?.type === \"string\" ? period.type : \"\";\n\tconst label = periodType.includes(\"MONTHLY\") ? \"30D\" : \"7D\";\n\treturn [{ label, usedPercent: clampPercent(usedPercent), ...(resetsAt === undefined ? {} : { resetsAt }) }];\n}\n\nexport function parseZaiUsageSnapshot(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst payload = record(value);\n\tconst limits = record(payload?.data)?.limits;\n\tif (payload?.success !== true || !Array.isArray(limits)) return undefined;\n\tconst parsed = limits.slice(0, MAX_USAGE_LIMITS).map(zaiWindow).filter(isDefined);\n\tconst requestWindows = parsed.filter((window) => window.type === \"TIME_LIMIT\" && !window.featureOnly);\n\tconst selected =\n\t\trequestWindows.length > 0 ? requestWindows : parsed.filter((window) => window.type === \"TOKENS_LIMIT\");\n\tselected.sort((left, right) => left.durationSeconds - right.durationSeconds);\n\tconst windows = selected\n\t\t.slice(0, 4)\n\t\t.map(({ type: _type, durationSeconds: _duration, featureOnly: _feature, ...window }) => window);\n\treturn windows.length > 0 ? windows : undefined;\n}\n\nexport function recordCodexPassiveUsage(apiKey: string, snapshot: ProviderRateLimitSnapshot, nowMs = Date.now()): void {\n\tconst accountId = codexAccountId(apiKey);\n\tconst limitId = snapshot.limitId?.trim().toLowerCase().replace(/-/g, \"_\");\n\tif (!accountId || !Number.isFinite(nowMs) || nowMs < 0 || (limitId && limitId !== \"codex\")) return;\n\tprunePassiveCodexUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(accountId);\n\tconst primary = normalizePassiveCodexWindow(snapshot.primary, nowMs);\n\tconst secondary = normalizePassiveCodexWindow(snapshot.secondary, nowMs);\n\tif (!primary && !secondary) return;\n\n\tconst previous = passiveCodexUsage.get(cacheKey);\n\tconst next: PassiveUsageEntry = {\n\t\tprimary: primary ? { window: primary, observedAt: nowMs } : freshObservedCodexWindow(previous?.primary, nowMs),\n\t\tsecondary: secondary\n\t\t\t? { window: secondary, observedAt: nowMs }\n\t\t\t: freshObservedCodexWindow(previous?.secondary, nowMs),\n\t};\n\tpassiveCodexUsage.delete(cacheKey);\n\tpassiveCodexUsage.set(cacheKey, next);\n\twhile (passiveCodexUsage.size > MAX_PASSIVE_ACCOUNTS) {\n\t\tconst oldestCacheKey = passiveCodexUsage.keys().next().value;\n\t\tif (oldestCacheKey === undefined) break;\n\t\tpassiveCodexUsage.delete(oldestCacheKey);\n\t}\n\tbumpSubscriptionUsageRevision(\"openai-codex\");\n}\n\nexport function recordClaudePassiveUsage(\n\tapiKey: string,\n\tsnapshot: ProviderRateLimitSnapshot,\n\tnowMs = Date.now(),\n): void {\n\tconst limitId = snapshot.limitId?.trim().toLowerCase().replace(/-/g, \"_\");\n\tif (\n\t\t!apiKey ||\n\t\tapiKey.length > 32_768 ||\n\t\t!Number.isFinite(nowMs) ||\n\t\tnowMs < 0 ||\n\t\t(limitId && limitId !== \"anthropic_unified\")\n\t) {\n\t\treturn;\n\t}\n\tprunePassiveClaudeUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(apiKey);\n\tconst primary = normalizePassiveCodexWindow(snapshot.primary, nowMs);\n\tconst secondary = normalizePassiveCodexWindow(snapshot.secondary, nowMs);\n\tif (!primary && !secondary) return;\n\n\tconst previous = passiveClaudeUsage.get(cacheKey);\n\tconst next: PassiveUsageEntry = {\n\t\tprimary: primary ? { window: primary, observedAt: nowMs } : freshObservedCodexWindow(previous?.primary, nowMs),\n\t\tsecondary: secondary\n\t\t\t? { window: secondary, observedAt: nowMs }\n\t\t\t: freshObservedCodexWindow(previous?.secondary, nowMs),\n\t};\n\tpassiveClaudeUsage.delete(cacheKey);\n\tpassiveClaudeUsage.set(cacheKey, next);\n\twhile (passiveClaudeUsage.size > MAX_PASSIVE_ACCOUNTS) {\n\t\tconst oldestCacheKey = passiveClaudeUsage.keys().next().value;\n\t\tif (oldestCacheKey === undefined) break;\n\t\tpassiveClaudeUsage.delete(oldestCacheKey);\n\t}\n\tbumpSubscriptionUsageRevision(\"anthropic\");\n}\n\nasync function fetchCodexUsage(\n\tlabel: string,\n\tapiKey: string,\n\tfetchImpl: FetchLike,\n): Promise<SubscriptionUsageSnapshot> {\n\tconst accountId = codexAccountId(apiKey);\n\tif (!accountId) return { label, windows: [], message: \"usage unavailable\" };\n\tconst passive = passiveCodexSnapshot(apiKey);\n\tlet polled: CodexUsageSnapshot | undefined;\n\ttry {\n\t\tconst payload = await fetchJson(fetchImpl, CODEX_USAGE_URL, {\n\t\t\theaders: {\n\t\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\t\tAccept: \"application/json\",\n\t\t\t\t\"chatgpt-account-id\": accountId,\n\t\t\t\t\"User-Agent\": \"omk\",\n\t\t\t},\n\t\t});\n\t\tpolled = parseCodexUsageSnapshot(payload);\n\t} catch (error) {\n\t\tif (!passive) throw error;\n\t}\n\tconst merged: CodexUsageSnapshot = {\n\t\tfiveHour: polled?.fiveHour ?? passive?.fiveHour,\n\t\tsevenDay: polled?.sevenDay ?? passive?.sevenDay,\n\t};\n\tif (!merged.fiveHour && !merged.sevenDay) return { label, windows: [], message: \"usage unavailable\" };\n\treturn {\n\t\tlabel,\n\t\twindows: [\n\t\t\tmerged.fiveHour ? withLabel(\"5H\", merged.fiveHour) : undefined,\n\t\t\tmerged.sevenDay ? withLabel(\"7D\", merged.sevenDay) : undefined,\n\t\t].filter(isDefined),\n\t};\n}\n\nasync function fetchClaudeUsage(\n\tlabel: string,\n\tapiKey: string,\n\tfetchImpl: FetchLike,\n): Promise<SubscriptionUsageSnapshot> {\n\tlet passive = passiveClaudeSnapshot(apiKey);\n\tconst response = await fetchJsonResult(fetchImpl, CLAUDE_USAGE_URL, {\n\t\theaders: {\n\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"anthropic-beta\": \"claude-code-20250219,oauth-2025-04-20\",\n\t\t\t\"User-Agent\": \"claude-cli/2.1.177 (external, cli)\",\n\t\t},\n\t});\n\tconst windows = [...(parseClaudeUsageSnapshot(response.payload) ?? [])];\n\tif (response.status === 429 && (!passive?.fiveHour || !passive?.sevenDay)) {\n\t\tawait probeClaudeQuota(apiKey, fetchImpl);\n\t\tpassive = passiveClaudeSnapshot(apiKey);\n\t}\n\tfor (const passiveWindow of [\n\t\tpassive?.fiveHour ? withLabel(\"5H\", passive.fiveHour) : undefined,\n\t\tpassive?.sevenDay ? withLabel(\"7D\", passive.sevenDay) : undefined,\n\t].filter(isDefined)) {\n\t\tif (!windows.some((window) => window.label === passiveWindow.label)) windows.push(passiveWindow);\n\t}\n\treturn windows.length > 0\n\t\t? { label, windows }\n\t\t: { label, windows: [], message: response.status === 429 ? \"rate limited · retry later\" : \"usage unavailable\" };\n}\n\n/**\n * Claude Code 2.1.177 performs the same `quota`/`max_tokens: 1` request during startup.\n * Keep it account-scoped and hourly so a pinned sidebar cannot turn the fallback into polling.\n */\nasync function probeClaudeQuota(apiKey: string, fetchImpl: FetchLike, nowMs = Date.now()): Promise<void> {\n\tif (!apiKey || apiKey.length > 32_768 || !Number.isFinite(nowMs) || nowMs < 0) return;\n\tconst cacheKey = passiveCredentialCacheKey(apiKey);\n\tconst activeProbe = claudeQuotaProbesInFlight.get(cacheKey);\n\tif (activeProbe) return activeProbe;\n\tconst lastAttempt = claudeQuotaProbeAttempts.get(cacheKey);\n\tif (lastAttempt !== undefined && nowMs - lastAttempt < CLAUDE_QUOTA_PROBE_COOLDOWN_MS) return;\n\tif (claudeQuotaProbesInFlight.size >= MAX_PASSIVE_ACCOUNTS) return;\n\n\tclaudeQuotaProbeAttempts.delete(cacheKey);\n\tclaudeQuotaProbeAttempts.set(cacheKey, nowMs);\n\twhile (claudeQuotaProbeAttempts.size > MAX_PASSIVE_ACCOUNTS) {\n\t\tconst oldestCacheKey = claudeQuotaProbeAttempts.keys().next().value;\n\t\tif (oldestCacheKey === undefined) break;\n\t\tclaudeQuotaProbeAttempts.delete(oldestCacheKey);\n\t}\n\n\tconst probe = (async () => {\n\t\ttry {\n\t\t\tconst response = await fetchImpl(CLAUDE_MESSAGES_URL, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: {\n\t\t\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\t\t\tAccept: \"application/json\",\n\t\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t\t\t\"anthropic-version\": \"2023-06-01\",\n\t\t\t\t\t\"anthropic-beta\": \"claude-code-20250219,oauth-2025-04-20\",\n\t\t\t\t\t\"anthropic-dangerous-direct-browser-access\": \"true\",\n\t\t\t\t\t\"User-Agent\": \"claude-cli/2.1.177 (external, cli)\",\n\t\t\t\t\t\"x-app\": \"cli\",\n\t\t\t\t},\n\t\t\t\tbody: JSON.stringify({\n\t\t\t\t\tmodel: \"claude-haiku-4-5\",\n\t\t\t\t\tmax_tokens: 1,\n\t\t\t\t\tmessages: [{ role: \"user\", content: \"quota\" }],\n\t\t\t\t}),\n\t\t\t\tsignal: AbortSignal.timeout(CLAUDE_QUOTA_PROBE_TIMEOUT_MS),\n\t\t\t});\n\t\t\tconst snapshot = parseClaudeQuotaProbeHeaders(response.headers, nowMs);\n\t\t\tif (snapshot) recordClaudePassiveUsage(apiKey, snapshot, nowMs);\n\t\t\ttry {\n\t\t\t\tawait response.body?.cancel();\n\t\t\t} catch {\n\t\t\t\t// The quota snapshot lives entirely in response headers.\n\t\t\t}\n\t\t} catch {\n\t\t\t// This mirrors Claude Code's best-effort startup quota check.\n\t\t}\n\t})().finally(() => {\n\t\tclaudeQuotaProbesInFlight.delete(cacheKey);\n\t});\n\tclaudeQuotaProbesInFlight.set(cacheKey, probe);\n\treturn probe;\n}\n\nfunction parseClaudeQuotaProbeHeaders(headers: Headers, nowMs: number): ProviderRateLimitSnapshot | undefined {\n\tconst parseWindow = (prefix: \"5h\" | \"7d\", windowSeconds: number): ProviderRateLimitWindow | undefined => {\n\t\tconst rawUtilization = headers.get(`anthropic-ratelimit-unified-${prefix}-utilization`);\n\t\tconst rawReset = headers.get(`anthropic-ratelimit-unified-${prefix}-reset`);\n\t\tif (rawUtilization === null || rawReset === null) return undefined;\n\t\tconst utilization = Number(rawUtilization);\n\t\tconst resetsAt = Number(rawReset);\n\t\tconst maxResetAt = nowMs / 1000 + 10 * 365 * 24 * 60 * 60;\n\t\tif (\n\t\t\t!Number.isFinite(utilization) ||\n\t\t\tutilization < 0 ||\n\t\t\tutilization > 1 ||\n\t\t\t!Number.isSafeInteger(resetsAt) ||\n\t\t\tresetsAt <= 0 ||\n\t\t\tresetsAt > maxResetAt\n\t\t) {\n\t\t\treturn undefined;\n\t\t}\n\t\treturn { usedPercent: Math.round(utilization * 10_000) / 100, windowSeconds, resetsAt };\n\t};\n\tconst primary = parseWindow(\"5h\", FIVE_HOUR_SECONDS);\n\tconst secondary = parseWindow(\"7d\", SEVEN_DAY_SECONDS);\n\treturn primary || secondary ? { limitId: \"anthropic-unified\", primary, secondary } : undefined;\n}\n\nasync function fetchGrokUsage(label: string, apiKey: string, fetchImpl: FetchLike): Promise<SubscriptionUsageSnapshot> {\n\tconst payload = await fetchJson(fetchImpl, GROK_BILLING_URL, {\n\t\theaders: {\n\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"x-xai-token-auth\": \"xai-grok-cli\",\n\t\t\t\"User-Agent\": \"omk\",\n\t\t},\n\t});\n\tconst windows = parseGrokUsageSnapshot(payload);\n\treturn windows ? { label, windows } : { label, windows: [], message: \"usage unavailable\" };\n}\n\nasync function fetchKimiUsage(label: string, apiKey: string, fetchImpl: FetchLike): Promise<SubscriptionUsageSnapshot> {\n\tconst payload = await fetchJson(fetchImpl, KIMI_USAGE_URL, {\n\t\theaders: {\n\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"User-Agent\": \"KimiCLI/1.5\",\n\t\t\t\"X-Msh-Platform\": \"kimi_cli\",\n\t\t\t\"X-Msh-Version\": \"1.5\",\n\t\t},\n\t});\n\tconst windows = parseKimiUsageSnapshot(payload);\n\treturn windows ? { label, windows } : { label, windows: [], message: \"usage unavailable\" };\n}\n\nasync function fetchZaiUsage(\n\tlabel: string,\n\tmodelProvider: string,\n\tcredential: { readonly provider: string; readonly apiKey: string },\n\tfetchImpl: FetchLike,\n): Promise<SubscriptionUsageSnapshot> {\n\tconst origin =\n\t\tmodelProvider === \"zai-coding-cn\" ||\n\t\tcredential.provider === \"zai-coding-cn\" ||\n\t\tcredential.provider === \"zhipu-coding-plan\"\n\t\t\t? \"https://open.bigmodel.cn\"\n\t\t\t: \"https://api.z.ai\";\n\tconst payload = await fetchJson(fetchImpl, `${origin}/api/monitor/usage/quota/limit`, {\n\t\theaders: {\n\t\t\tAuthorization: credential.apiKey,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"User-Agent\": \"omk\",\n\t\t},\n\t});\n\tconst windows = parseZaiUsageSnapshot(payload);\n\treturn windows ? { label, windows } : { label, windows: [], message: \"usage unavailable\" };\n}\n\nasync function fetchJson(fetchImpl: FetchLike, url: string, init: RequestInit): Promise<unknown> {\n\treturn (await fetchJsonResult(fetchImpl, url, init)).payload;\n}\n\nasync function fetchJsonResult(fetchImpl: FetchLike, url: string, init: RequestInit): Promise<FetchJsonResult> {\n\tconst response = await fetchImpl(url, { ...init, signal: AbortSignal.timeout(10_000) });\n\tif (!response.ok) return { status: response.status };\n\tconst declaredBytes = Number(response.headers?.get(\"content-length\"));\n\tif (Number.isFinite(declaredBytes) && declaredBytes > MAX_USAGE_RESPONSE_BYTES) {\n\t\treturn { status: response.status };\n\t}\n\tif (!response.body || typeof response.body.getReader !== \"function\") {\n\t\treturn { status: response.status, payload: await response.json() };\n\t}\n\n\tconst reader = response.body.getReader();\n\tconst chunks: Uint8Array[] = [];\n\tlet totalBytes = 0;\n\twhile (true) {\n\t\tconst { done, value } = await reader.read();\n\t\tif (done) break;\n\t\ttotalBytes += value.byteLength;\n\t\tif (totalBytes > MAX_USAGE_RESPONSE_BYTES) {\n\t\t\tawait reader.cancel();\n\t\t\treturn { status: response.status };\n\t\t}\n\t\tchunks.push(value);\n\t}\n\tconst body = new Uint8Array(totalBytes);\n\tlet offset = 0;\n\tfor (const chunk of chunks) {\n\t\tbody.set(chunk, offset);\n\t\toffset += chunk.byteLength;\n\t}\n\ttry {\n\t\treturn { status: response.status, payload: JSON.parse(new TextDecoder().decode(body)) as unknown };\n\t} catch {\n\t\treturn { status: response.status };\n\t}\n}\n\nfunction credentialConfigured(session: UsageSession, candidate: CredentialCandidate): boolean {\n\tif (candidate.oauthOnly) return session.modelRegistry.isUsingOAuthProvider(candidate.provider);\n\tconst status = session.modelRegistry.getProviderAuthStatus(candidate.provider);\n\treturn status.configured || status.source !== undefined;\n}\n\nasync function resolveCredential(\n\tsession: UsageSession,\n\tusageSource: SubscriptionUsageSource,\n): Promise<{ readonly provider: string; readonly apiKey: string } | undefined> {\n\tfor (const candidate of usageSource.credentials) {\n\t\tif (!credentialConfigured(session, candidate)) continue;\n\t\tconst apiKey = await session.modelRegistry.getApiKeyForProvider(candidate.provider);\n\t\tif (apiKey) return { provider: candidate.provider, apiKey };\n\t}\n\treturn undefined;\n}\n\nfunction codexWindow(value: unknown, nowSeconds: number): ParsedCodexWindow | undefined {\n\tconst window = record(value);\n\tconst usedPercent = finiteNumber(window?.used_percent);\n\tif (usedPercent === undefined) return undefined;\n\tconst windowSeconds = finiteNumber(window?.limit_window_seconds);\n\tconst resetAt = finiteNumber(window?.reset_at);\n\tconst resetAfter = finiteNumber(window?.reset_after_seconds);\n\tconst resetsAt = resetAt ?? (resetAfter === undefined ? undefined : nowSeconds + resetAfter);\n\treturn {\n\t\tusedPercent: clampPercent(usedPercent),\n\t\t...(windowSeconds === undefined ? {} : { windowSeconds }),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction claudeWindow(value: unknown): Omit<SubscriptionUsageWindow, \"label\"> | undefined {\n\tconst bucket = record(value);\n\tif (!bucket || bucket.is_active === false) return undefined;\n\tconst usedPercent = finiteNumber(bucket.utilization ?? bucket.percent);\n\tif (usedPercent === undefined) return undefined;\n\tconst resetsAt = epochSeconds(bucket.resets_at);\n\treturn { usedPercent: clampPercent(usedPercent), ...(resetsAt === undefined ? {} : { resetsAt }) };\n}\n\nfunction usageRatio(\n\tdata: Record<string, unknown> | undefined,\n\tnowSeconds: number,\n): Omit<SubscriptionUsageWindow, \"label\"> | undefined {\n\tif (!data) return undefined;\n\tconst limit = finiteNumber(data.limit);\n\tlet used = finiteNumber(data.used);\n\tconst remaining = finiteNumber(data.remaining);\n\tif (used === undefined && limit !== undefined && remaining !== undefined) used = limit - remaining;\n\tif (used === undefined || limit === undefined || limit <= 0) return undefined;\n\tconst resetsAt = parseReset(data, nowSeconds);\n\treturn {\n\t\tusedPercent: clampPercent((used / limit) * 100),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction zaiWindow(value: unknown):\n\t| (SubscriptionUsageWindow & {\n\t\t\treadonly type: string;\n\t\t\treadonly durationSeconds: number;\n\t\t\treadonly featureOnly: boolean;\n\t  })\n\t| undefined {\n\tconst item = record(value);\n\tif (!item || (item.type !== \"TIME_LIMIT\" && item.type !== \"TOKENS_LIMIT\")) return undefined;\n\tconst percentage = finiteNumber(item.percentage);\n\tconst current = finiteNumber(item.currentValue);\n\tconst limit = finiteNumber(item.usage);\n\tconst usedPercent =\n\t\tpercentage ?? (current !== undefined && limit !== undefined && limit > 0 ? (current / limit) * 100 : undefined);\n\tif (usedPercent === undefined) return undefined;\n\tconst duration = zaiDuration(finiteNumber(item.unit), finiteNumber(item.number));\n\tconst resetsAt = epochSeconds(item.nextResetTime);\n\tconst detailCodes = Array.isArray(item.usageDetails)\n\t\t? item.usageDetails\n\t\t\t\t.map((detail) => record(detail)?.modelCode)\n\t\t\t\t.filter((code): code is string => typeof code === \"string\")\n\t\t: [];\n\treturn {\n\t\ttype: item.type,\n\t\tlabel: duration.label,\n\t\tdurationSeconds: duration.seconds,\n\t\tfeatureOnly: [\"search-prime\", \"web-reader\", \"zread\"].every((code) => detailCodes.includes(code)),\n\t\tusedPercent: clampPercent(usedPercent),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction zaiDuration(unit: number | undefined, countValue: number | undefined): { label: string; seconds: number } {\n\tconst count = countValue && countValue > 0 ? countValue : 1;\n\tif (unit === 3) return { label: `${count}H`, seconds: count * 60 * 60 };\n\tif (unit === 4) return { label: `${count}D`, seconds: count * 24 * 60 * 60 };\n\tif (unit === 5) return { label: count === 1 ? \"30D\" : `${count}MO`, seconds: count * 30 * 24 * 60 * 60 };\n\tif (unit === 6) return { label: \"7D\", seconds: 7 * 24 * 60 * 60 };\n\treturn { label: \"QUOTA\", seconds: Number.POSITIVE_INFINITY };\n}\n\nfunction durationLabel(window: Record<string, unknown> | undefined): string | undefined {\n\tif (!window) return undefined;\n\tconst duration = finiteNumber(window.duration);\n\tconst unit = typeof window.timeUnit === \"string\" ? window.timeUnit.toUpperCase() : \"\";\n\tif (duration === undefined) return undefined;\n\tif (unit.includes(\"MINUTE\")) return duration % 60 === 0 ? `${duration / 60}H` : `${duration}M`;\n\tif (unit.includes(\"HOUR\")) return `${duration}H`;\n\tif (unit.includes(\"DAY\")) return `${duration}D`;\n\tif (unit.includes(\"SECOND\")) return `${duration}S`;\n\treturn undefined;\n}\n\nfunction parseReset(data: Record<string, unknown> | undefined, nowSeconds: number): number | undefined {\n\tif (!data) return undefined;\n\tfor (const key of [\"reset_at\", \"resetAt\", \"reset_time\", \"resetTime\"] as const) {\n\t\tconst parsed = epochSeconds(data[key]);\n\t\tif (parsed !== undefined) return parsed;\n\t}\n\tfor (const key of [\"reset_in\", \"resetIn\", \"ttl\"] as const) {\n\t\tconst seconds = finiteNumber(data[key]);\n\t\tif (seconds !== undefined) return nowSeconds + seconds;\n\t}\n\treturn undefined;\n}\n\nfunction epochSeconds(value: unknown): number | undefined {\n\tconst numeric = finiteNumber(value);\n\tif (numeric !== undefined) return numeric > 1_000_000_000_000 ? numeric / 1000 : numeric;\n\tif (typeof value !== \"string\" || !value.trim()) return undefined;\n\tconst parsed = Date.parse(value);\n\treturn Number.isFinite(parsed) ? parsed / 1000 : undefined;\n}\n\nfunction normalizePassiveCodexWindow(\n\twindow: ProviderRateLimitWindow | undefined,\n\tnowMs: number,\n): ParsedCodexWindow | undefined {\n\tif (!window || !Number.isFinite(window.usedPercent)) return undefined;\n\tconst windowSeconds =\n\t\twindow.windowSeconds !== undefined &&\n\t\tNumber.isFinite(window.windowSeconds) &&\n\t\twindow.windowSeconds > 0 &&\n\t\twindow.windowSeconds <= Number.MAX_SAFE_INTEGER\n\t\t\t? Math.round(window.windowSeconds)\n\t\t\t: undefined;\n\tconst maxResetAt = nowMs / 1000 + 10 * 365 * 24 * 60 * 60;\n\tconst resetsAtValue = window.resetsAt === undefined ? undefined : Math.floor(window.resetsAt);\n\tconst resetsAt =\n\t\tresetsAtValue !== undefined &&\n\t\tNumber.isSafeInteger(resetsAtValue) &&\n\t\tresetsAtValue > 0 &&\n\t\tresetsAtValue <= maxResetAt\n\t\t\t? resetsAtValue\n\t\t\t: undefined;\n\treturn {\n\t\tusedPercent: clampPercent(window.usedPercent),\n\t\t...(windowSeconds === undefined ? {} : { windowSeconds }),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction passiveCredentialCacheKey(identifier: string): string {\n\treturn createHash(\"sha256\").update(identifier).digest(\"base64url\");\n}\n\nfunction freshObservedCodexWindow(\n\tobserved: ObservedCodexWindow | undefined,\n\tnowMs: number,\n): ObservedCodexWindow | undefined {\n\tif (!observed || nowMs - observed.observedAt > PASSIVE_USAGE_TTL_MS) return undefined;\n\tif (observed.window.resetsAt !== undefined && observed.window.resetsAt <= nowMs / 1000) return undefined;\n\treturn observed;\n}\n\nfunction prunePassiveCodexUsage(nowMs: number): void {\n\tfor (const [cacheKey, entry] of passiveCodexUsage) {\n\t\tif (!freshObservedCodexWindow(entry.primary, nowMs) && !freshObservedCodexWindow(entry.secondary, nowMs)) {\n\t\t\tpassiveCodexUsage.delete(cacheKey);\n\t\t}\n\t}\n}\n\nfunction prunePassiveClaudeUsage(nowMs: number): void {\n\tfor (const [cacheKey, entry] of passiveClaudeUsage) {\n\t\tif (!freshObservedCodexWindow(entry.primary, nowMs) && !freshObservedCodexWindow(entry.secondary, nowMs)) {\n\t\t\tpassiveClaudeUsage.delete(cacheKey);\n\t\t}\n\t}\n}\n\nfunction passiveCodexSnapshot(apiKey: string, nowMs = Date.now()): CodexUsageSnapshot | undefined {\n\tconst accountId = codexAccountId(apiKey);\n\tif (!accountId || !Number.isFinite(nowMs) || nowMs < 0) return undefined;\n\tprunePassiveCodexUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(accountId);\n\tconst entry = passiveCodexUsage.get(cacheKey);\n\tif (!entry) return undefined;\n\tconst primary = freshObservedCodexWindow(entry.primary, nowMs);\n\tconst secondary = freshObservedCodexWindow(entry.secondary, nowMs);\n\tif (!primary && !secondary) {\n\t\tpassiveCodexUsage.delete(cacheKey);\n\t\treturn undefined;\n\t}\n\tconst freshEntry = { primary, secondary };\n\tpassiveCodexUsage.delete(cacheKey);\n\tpassiveCodexUsage.set(cacheKey, freshEntry);\n\n\tconst windows = [primary?.window, secondary?.window].filter(isDefined);\n\tconst fiveHour = windows.find((window) => near(window.windowSeconds, FIVE_HOUR_SECONDS));\n\tconst sevenDay = windows.find((window) => near(window.windowSeconds, SEVEN_DAY_SECONDS));\n\tif (!fiveHour && !sevenDay) return undefined;\n\treturn {\n\t\t...(fiveHour ? { fiveHour: publicCodexWindow(fiveHour) } : {}),\n\t\t...(sevenDay ? { sevenDay: publicCodexWindow(sevenDay) } : {}),\n\t};\n}\n\nfunction passiveClaudeSnapshot(apiKey: string, nowMs = Date.now()): CodexUsageSnapshot | undefined {\n\tif (!apiKey || apiKey.length > 32_768 || !Number.isFinite(nowMs) || nowMs < 0) return undefined;\n\tprunePassiveClaudeUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(apiKey);\n\tconst entry = passiveClaudeUsage.get(cacheKey);\n\tif (!entry) return undefined;\n\tconst primary = freshObservedCodexWindow(entry.primary, nowMs);\n\tconst secondary = freshObservedCodexWindow(entry.secondary, nowMs);\n\tif (!primary && !secondary) {\n\t\tpassiveClaudeUsage.delete(cacheKey);\n\t\treturn undefined;\n\t}\n\tpassiveClaudeUsage.delete(cacheKey);\n\tpassiveClaudeUsage.set(cacheKey, { primary, secondary });\n\n\tconst windows = [primary?.window, secondary?.window].filter(isDefined);\n\tconst fiveHour = windows.find((window) => near(window.windowSeconds, FIVE_HOUR_SECONDS));\n\tconst sevenDay = windows.find((window) => near(window.windowSeconds, SEVEN_DAY_SECONDS));\n\tif (!fiveHour && !sevenDay) return undefined;\n\treturn {\n\t\t...(fiveHour ? { fiveHour: publicCodexWindow(fiveHour) } : {}),\n\t\t...(sevenDay ? { sevenDay: publicCodexWindow(sevenDay) } : {}),\n\t};\n}\n\nfunction codexAccountId(token: string): string | undefined {\n\ttry {\n\t\tconst payload = token.split(\".\")[1];\n\t\tif (!payload) return undefined;\n\t\tconst claims = record(JSON.parse(Buffer.from(payload, \"base64url\").toString(\"utf8\")));\n\t\tconst auth = record(claims?.[OPENAI_AUTH_CLAIM]);\n\t\tconst accountId = typeof auth?.chatgpt_account_id === \"string\" ? auth.chatgpt_account_id.trim() : \"\";\n\t\treturn accountId && accountId.length <= 256 ? accountId : undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nfunction publicCodexWindow(window: ParsedCodexWindow): CodexUsageWindow {\n\treturn { usedPercent: window.usedPercent, ...(window.resetsAt === undefined ? {} : { resetsAt: window.resetsAt }) };\n}\n\nfunction withLabel(label: string, window: Omit<SubscriptionUsageWindow, \"label\">): SubscriptionUsageWindow {\n\treturn { label, ...window };\n}\n\nfunction shortLabel(value: unknown, index: number): string {\n\tif (typeof value !== \"string\") return `LIMIT${index + 1}`;\n\tconst safe = value\n\t\t.replace(/[\\u0000-\\u001f\\u007f-\\u009f\\u200b-\\u200f\\u202a-\\u202e\\u2060-\\u206f]/g, \"\")\n\t\t.replace(/\\s+/g, \" \")\n\t\t.trim();\n\treturn safe ? safe.toUpperCase().slice(0, 8) : `LIMIT${index + 1}`;\n}\n\nfunction finiteNumber(value: unknown): number | undefined {\n\tconst number =\n\t\ttypeof value === \"number\" ? value : typeof value === \"string\" && value.trim() ? Number(value) : Number.NaN;\n\treturn Number.isFinite(number) ? number : undefined;\n}\n\nfunction clampPercent(value: number): number {\n\treturn Number(Math.max(0, Math.min(100, value)).toFixed(2));\n}\n\nfunction near(actual: number | undefined, expected: number): boolean {\n\treturn actual !== undefined && Math.abs(actual - expected) <= WINDOW_TOLERANCE_SECONDS;\n}\n\nfunction record(value: unknown): Record<string, unknown> | undefined {\n\treturn typeof value === \"object\" && value !== null && !Array.isArray(value)\n\t\t? (value as Record<string, unknown>)\n\t\t: undefined;\n}\n\nfunction isDefined<T>(value: T | undefined): value is T {\n\treturn value !== undefined;\n}\n\nfunction offline(): boolean {\n\tconst value = process.env.OMK_OFFLINE?.toLowerCase();\n\treturn value === \"1\" || value === \"true\" || value === \"yes\";\n}\n"]}
+{"version":3,"file":"provider-usage.d.ts","sourceRoot":"","sources":["../../src/core/provider-usage.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,yBAAyB,EAA2B,MAAM,QAAQ,CAAC;AACjF,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAiDvD,MAAM,MAAM,uBAAuB,GAAG;IACrC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;CAC3B,CAAC;AAEF,MAAM,MAAM,yBAAyB,GAAG;IACvC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,SAAS,uBAAuB,EAAE,CAAC;IACrD,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;CAC1B,CAAC;AAEF,KAAK,gBAAgB,GAAG;IAAE,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC;AACrF,MAAM,MAAM,kBAAkB,GAAG;IAChC,QAAQ,CAAC,QAAQ,CAAC,EAAE,gBAAgB,CAAC;IACrC,QAAQ,CAAC,QAAQ,CAAC,EAAE,gBAAgB,CAAC;CACrC,CAAC;AAKF,KAAK,SAAS,GAAG,OAAO,GAAG,QAAQ,GAAG,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,iBAAiB,GAAG,aAAa,CAAC;AAClG,KAAK,mBAAmB,GAAG;IAAE,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAA;CAAE,CAAC;AAEtF,MAAM,MAAM,uBAAuB,GAAG;IACrC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC;IACzB,QAAQ,CAAC,WAAW,EAAE,SAAS,mBAAmB,EAAE,CAAC;IACrD,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,kBAAkB,CAAC,EAAE,MAAM,CAAC;CACrC,CAAC;AAEF,KAAK,YAAY,GAAG,IAAI,CAAC,YAAY,EAAE,OAAO,GAAG,eAAe,CAAC,CAAC;AAClE,KAAK,SAAS,GAAG,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;AA8D1F,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,GAAG,uBAAuB,GAAG,SAAS,CAW5G;AASD,wBAAgB,4BAA4B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAErE;AAOD,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,YAAY,EAAE,QAAQ,SAAgC,GAAG,OAAO,CAGlH;AAED,8EAA8E;AAC9E,wBAAgB,uCAAuC,CAAC,OAAO,EAAE,YAAY,GAAG,SAAS,MAAM,EAAE,CAUhG;AAED,wBAAsB,qBAAqB,CAC1C,OAAO,EAAE,YAAY,EACrB,SAAS,GAAE,SAAiB,EAC5B,QAAQ,SAAgC,EACxC,aAAa,GAAE,aAA0B,GACvC,OAAO,CAAC,yBAAyB,GAAG,SAAS,CAAC,CAsChD;AAED,wBAAgB,uBAAuB,CACtC,KAAK,EAAE,OAAO,EACd,UAAU,SAAoB,GAC5B,kBAAkB,GAAG,SAAS,CAchC;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAcvG;AAED,wBAAgB,sBAAsB,CACrC,KAAK,EAAE,OAAO,EACd,UAAU,SAAoB,GAC5B,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAuBhD;AAED,sFAAsF;AACtF,MAAM,MAAM,aAAa,GACtB;IAAE,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAA;CAAE,GAC5B;IAAE,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC;IAAC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CAAC;AAChF,MAAM,MAAM,aAAa,GAAG,CAAC,IAAI,EAAE,SAAS,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,KAAK,OAAO,CAAC,aAAa,CAAC,CAAC;AAEnG,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,MAAM,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,CAOpF;AAED,qFAAmF;AACnF,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CActG;AAED,wBAAsB,uBAAuB,CAC5C,KAAK,EAAE,MAAM,EACb,MAAM,GAAE,aAA0B,GAChC,OAAO,CAAC,yBAAyB,CAAC,CAqBpC;AAyFD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAiBrG;AAED,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,uBAAuB,EAAE,GAAG,SAAS,CAapG;AAED,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,yBAAyB,EAAE,KAAK,SAAa,GAAG,IAAI,CAyBrH;AAED,wBAAgB,wBAAwB,CACvC,MAAM,EAAE,MAAM,EACd,QAAQ,EAAE,yBAAyB,EACnC,KAAK,SAAa,GAChB,IAAI,CAgCN","sourcesContent":["import { execFile } from \"node:child_process\";\nimport { createHash } from \"node:crypto\";\nimport type { ProviderRateLimitSnapshot, ProviderRateLimitWindow } from \"omk-ai\";\nimport type { AgentSession } from \"./agent-session.ts\";\n\nconst FIVE_HOUR_SECONDS = 5 * 60 * 60;\nconst SEVEN_DAY_SECONDS = 7 * 24 * 60 * 60;\nconst WINDOW_TOLERANCE_SECONDS = 120;\nconst CODEX_USAGE_URL = \"https://chatgpt.com/backend-api/wham/usage\";\nconst CLAUDE_USAGE_URL = \"https://api.anthropic.com/api/oauth/usage\";\nconst CLAUDE_MESSAGES_URL = \"https://api.anthropic.com/v1/messages\";\nconst KIMI_USAGE_URL = \"https://api.kimi.com/coding/v1/usages\";\nconst GROK_BILLING_URL = \"https://cli-chat-proxy.grok.com/v1/billing?format=credits\";\nconst OPENAI_AUTH_CLAIM = \"https://api.openai.com/auth\";\nconst MAX_USAGE_RESPONSE_BYTES = 1024 * 1024;\nconst MAX_USAGE_LIMITS = 64;\nconst MAX_PASSIVE_ACCOUNTS = 64;\nconst PASSIVE_USAGE_TTL_MS = 6 * 60 * 60 * 1000;\nconst CLAUDE_QUOTA_PROBE_COOLDOWN_MS = 60 * 60 * 1000;\nconst CLAUDE_QUOTA_PROBE_TIMEOUT_MS = 10_000;\nconst QWEN_CLI_TIMEOUT_MS = 15_000;\nconst QWEN_CLI_ARGS = [\"usage\", \"summary\", \"--format\", \"json\"] as const;\nconst qwenBillingArgs = (month: string): readonly string[] =>\n\t[\"billing\", \"breakdown\", \"--from\", month, \"--to\", month, \"--group-by\", \"model\", \"--format\", \"json\"] as const;\nconst QWEN_CONNECT_HINT = \"connect: npm i -g @qwencloud/qwencloud-cli && qwencloud auth login\";\nconst QWEN_CLI_ENV_KEYS = [\n\t\"PATH\",\n\t\"Path\",\n\t\"HOME\",\n\t\"USERPROFILE\",\n\t\"HOMEDRIVE\",\n\t\"HOMEPATH\",\n\t\"APPDATA\",\n\t\"LOCALAPPDATA\",\n\t\"XDG_CONFIG_HOME\",\n\t\"XDG_DATA_HOME\",\n\t\"XDG_CACHE_HOME\",\n\t\"TMPDIR\",\n\t\"TEMP\",\n\t\"TMP\",\n\t\"SystemRoot\",\n\t\"WINDIR\",\n\t\"ComSpec\",\n\t\"PATHEXT\",\n\t\"LANG\",\n\t\"LC_ALL\",\n\t\"LC_CTYPE\",\n\t\"TERM\",\n\t\"SSL_CERT_FILE\",\n\t\"SSL_CERT_DIR\",\n] as const;\n\nexport type SubscriptionUsageWindow = {\n\treadonly label: string;\n\treadonly usedPercent: number;\n\treadonly resetsAt?: number;\n};\n\nexport type SubscriptionUsageSnapshot = {\n\treadonly label: string;\n\treadonly windows: readonly SubscriptionUsageWindow[];\n\treadonly message?: string;\n};\n\ntype CodexUsageWindow = { readonly usedPercent: number; readonly resetsAt?: number };\nexport type CodexUsageSnapshot = {\n\treadonly fiveHour?: CodexUsageWindow;\n\treadonly sevenDay?: CodexUsageWindow;\n};\n\ntype ParsedCodexWindow = CodexUsageWindow & { readonly windowSeconds?: number };\ntype ObservedCodexWindow = { readonly window: ParsedCodexWindow; readonly observedAt: number };\ntype PassiveUsageEntry = { readonly primary?: ObservedCodexWindow; readonly secondary?: ObservedCodexWindow };\ntype UsageKind = \"codex\" | \"claude\" | \"kimi\" | \"zai\" | \"grok\" | \"qwen-token-plan\" | \"unavailable\";\ntype CredentialCandidate = { readonly provider: string; readonly oauthOnly: boolean };\n\nexport type SubscriptionUsageSource = {\n\treadonly label: string;\n\treadonly kind: UsageKind;\n\treadonly credentials: readonly CredentialCandidate[];\n\treadonly ttlMs: number;\n\treadonly unavailableMessage?: string;\n};\n\ntype UsageSession = Pick<AgentSession, \"state\" | \"modelRegistry\">;\ntype FetchLike = (input: string | URL | Request, init?: RequestInit) => Promise<Response>;\ntype FetchJsonResult = { readonly status: number; readonly payload?: unknown };\n\nconst passiveCodexUsage = new Map<string, PassiveUsageEntry>();\nconst passiveClaudeUsage = new Map<string, PassiveUsageEntry>();\nconst claudeQuotaProbeAttempts = new Map<string, number>();\nconst claudeQuotaProbesInFlight = new Map<string, Promise<void>>();\nconst subscriptionUsageRevisions = new Map<string, number>();\nconst MINUTE_TTL_MS = 60_000;\nconst VISIBLE_USAGE_PROVIDERS = [\n\t\"openai-codex\",\n\t\"anthropic\",\n\t\"kimi-coding\",\n\t\"zai\",\n\t\"modelstudio-maas\",\n\t\"qwen-oauth\",\n\t\"xai\",\n] as const;\nconst SOURCES: Readonly<Record<string, SubscriptionUsageSource>> = {\n\t\"openai-codex\": source(\"CODEX\", \"codex\", [{ provider: \"openai-codex\", oauthOnly: true }]),\n\tanthropic: source(\"CLAUDE\", \"claude\", [{ provider: \"anthropic\", oauthOnly: true }], 5 * MINUTE_TTL_MS),\n\t\"qwen-oauth\": source(\"QWEN\", \"unavailable\", [{ provider: \"qwen-oauth\", oauthOnly: true }]),\n\t// Model Studio exposes no API-key quota endpoint (console gateway is\n\t// session-only), so the token-plan quota rides the official QwenCloud\n\t// management CLI, which holds its own OAuth credential.\n\t\"modelstudio-maas\": source(\n\t\t\"QWEN TOKEN PLAN\",\n\t\t\"qwen-token-plan\",\n\t\t[{ provider: \"modelstudio-maas\", oauthOnly: false }],\n\t\t5 * MINUTE_TTL_MS,\n\t),\n\t\"kimi-code\": source(\"KIMI\", \"kimi\", [\n\t\t{ provider: \"kimi-code\", oauthOnly: true },\n\t\t{ provider: \"kimi-coding\", oauthOnly: false },\n\t]),\n\t\"kimi-coding\": source(\"KIMI\", \"kimi\", [\n\t\t{ provider: \"kimi-code\", oauthOnly: true },\n\t\t{ provider: \"kimi-coding\", oauthOnly: false },\n\t]),\n\t\"zhipu-coding-plan\": source(\"GLM\", \"zai\", [{ provider: \"zhipu-coding-plan\", oauthOnly: true }]),\n\tzai: source(\"GLM\", \"zai\", [\n\t\t{ provider: \"zai\", oauthOnly: false },\n\t\t{ provider: \"zai-coding-cn\", oauthOnly: false },\n\t\t{ provider: \"zhipu-coding-plan\", oauthOnly: true },\n\t]),\n\t\"zai-coding-cn\": source(\"GLM\", \"zai\", [\n\t\t{ provider: \"zai-coding-cn\", oauthOnly: false },\n\t\t{ provider: \"zhipu-coding-plan\", oauthOnly: true },\n\t]),\n\txai: source(\"GROK\", \"grok\", [{ provider: \"xai\", oauthOnly: true }]),\n};\n\nfunction source(\n\tlabel: string,\n\tkind: UsageKind,\n\tcredentials: readonly CredentialCandidate[],\n\tttlMs = MINUTE_TTL_MS,\n\tunavailableMessage?: string,\n): SubscriptionUsageSource {\n\treturn { label, kind, credentials, ttlMs, unavailableMessage };\n}\n\nexport function getSubscriptionUsageSource(provider: string | undefined): SubscriptionUsageSource | undefined {\n\tif (!provider) return undefined;\n\tconst configured = SOURCES[provider];\n\tif (configured) return configured;\n\tif (/^openai-codex-\\d+$/.test(provider)) {\n\t\treturn source(\"CODEX\", \"codex\", [\n\t\t\t{ provider, oauthOnly: true },\n\t\t\t{ provider: \"openai-codex\", oauthOnly: true },\n\t\t]);\n\t}\n\treturn undefined;\n}\n\nfunction usageRevisionKey(provider: string): string {\n\tconst kind = getSubscriptionUsageSource(provider)?.kind;\n\tif (kind === \"codex\") return \"codex\";\n\tif (kind === \"claude\") return \"claude\";\n\treturn provider;\n}\n\nexport function getSubscriptionUsageRevision(provider: string): number {\n\treturn subscriptionUsageRevisions.get(usageRevisionKey(provider)) ?? 0;\n}\n\nfunction bumpSubscriptionUsageRevision(provider: string): void {\n\tconst key = usageRevisionKey(provider);\n\tsubscriptionUsageRevisions.set(key, (subscriptionUsageRevisions.get(key) ?? 0) + 1);\n}\n\nexport function supportsSubscriptionUsage(session: UsageSession, provider = session.state.model?.provider): boolean {\n\tconst usageSource = getSubscriptionUsageSource(provider);\n\treturn usageSource?.credentials.some((candidate) => credentialConfigured(session, candidate)) ?? false;\n}\n\n/** Configured quota groups shown in the rail, with the active group first. */\nexport function getConfiguredSubscriptionUsageProviders(session: UsageSession): readonly string[] {\n\tconst activeProvider = session.state.model?.provider;\n\tconst activeSource = getSubscriptionUsageSource(activeProvider);\n\tconst providers = VISIBLE_USAGE_PROVIDERS.filter((provider) => supportsSubscriptionUsage(session, provider));\n\tif (!activeProvider || !activeSource || !supportsSubscriptionUsage(session, activeProvider)) return providers;\n\tconst activeCanonical = providers.find(\n\t\t(provider) => getSubscriptionUsageSource(provider)?.label === activeSource.label,\n\t);\n\tif (!activeCanonical) return [activeProvider, ...providers];\n\treturn [activeCanonical, ...providers.filter((provider) => provider !== activeCanonical)];\n}\n\nexport async function loadSubscriptionUsage(\n\tsession: UsageSession,\n\tfetchImpl: FetchLike = fetch,\n\tprovider = session.state.model?.provider,\n\tqwenCliRunner: QwenCliRunner = runQwenCli,\n): Promise<SubscriptionUsageSnapshot | undefined> {\n\tconst model = session.state.model;\n\tconst usageSource = getSubscriptionUsageSource(provider);\n\tif (!model || !usageSource || !supportsSubscriptionUsage(session, provider)) return undefined;\n\tif (usageSource.kind === \"unavailable\") {\n\t\treturn {\n\t\t\tlabel: usageSource.label,\n\t\t\twindows: [],\n\t\t\tmessage: usageSource.unavailableMessage ?? \"quota API unavailable\",\n\t\t};\n\t}\n\tif (offline()) return { label: usageSource.label, windows: [], message: \"offline\" };\n\tif (usageSource.kind === \"qwen-token-plan\") {\n\t\t// The QwenCloud CLI authenticates itself; no OMK-held credential is used.\n\t\treturn await fetchQwenTokenPlanUsage(usageSource.label, qwenCliRunner);\n\t}\n\n\tconst credential = await resolveCredential(session, usageSource);\n\tif (!credential) return { label: usageSource.label, windows: [], message: \"usage unavailable\" };\n\n\ttry {\n\t\tswitch (usageSource.kind) {\n\t\t\tcase \"codex\":\n\t\t\t\treturn await fetchCodexUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tcase \"claude\":\n\t\t\t\treturn await fetchClaudeUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tcase \"kimi\":\n\t\t\t\treturn await fetchKimiUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tcase \"zai\":\n\t\t\t\treturn await fetchZaiUsage(usageSource.label, provider ?? model.provider, credential, fetchImpl);\n\t\t\tcase \"grok\":\n\t\t\t\treturn await fetchGrokUsage(usageSource.label, credential.apiKey, fetchImpl);\n\t\t\tdefault:\n\t\t\t\treturn { label: usageSource.label, windows: [], message: \"usage unavailable\" };\n\t\t}\n\t} catch {\n\t\treturn { label: usageSource.label, windows: [], message: \"usage unavailable\" };\n\t}\n}\n\nexport function parseCodexUsageSnapshot(\n\tvalue: unknown,\n\tnowSeconds = Date.now() / 1000,\n): CodexUsageSnapshot | undefined {\n\tconst rateLimit = record(record(value)?.rate_limit);\n\tconst primary = codexWindow(rateLimit?.primary_window, nowSeconds);\n\tconst secondary = codexWindow(rateLimit?.secondary_window, nowSeconds);\n\tconst windows = [primary, secondary].filter((window): window is ParsedCodexWindow => window !== undefined);\n\tconst fiveHour = windows.find((window) => near(window.windowSeconds, FIVE_HOUR_SECONDS));\n\tconst sevenDay = windows.find((window) => near(window.windowSeconds, SEVEN_DAY_SECONDS));\n\tconst fallbackFiveHour = fiveHour ?? (primary?.windowSeconds === undefined ? primary : undefined);\n\tconst fallbackSevenDay = sevenDay ?? (secondary?.windowSeconds === undefined ? secondary : undefined);\n\tif (!fallbackFiveHour && !fallbackSevenDay) return undefined;\n\treturn {\n\t\t...(fallbackFiveHour ? { fiveHour: publicCodexWindow(fallbackFiveHour) } : {}),\n\t\t...(fallbackSevenDay ? { sevenDay: publicCodexWindow(fallbackSevenDay) } : {}),\n\t};\n}\n\nexport function parseClaudeUsageSnapshot(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst payload = record(value);\n\tif (!payload) return undefined;\n\tconst entries = Array.isArray(payload.limits)\n\t\t? payload.limits.slice(0, MAX_USAGE_LIMITS).map(record).filter(isDefined)\n\t\t: [];\n\tconst fiveHour = claudeWindow(payload.five_hour) ?? claudeWindow(entries.find((entry) => entry.kind === \"session\"));\n\tconst sevenDay =\n\t\tclaudeWindow(payload.seven_day) ?? claudeWindow(entries.find((entry) => entry.kind === \"weekly_all\"));\n\tconst windows = [\n\t\tfiveHour ? withLabel(\"5H\", fiveHour) : undefined,\n\t\tsevenDay ? withLabel(\"7D\", sevenDay) : undefined,\n\t].filter(isDefined);\n\treturn windows.length > 0 ? windows : undefined;\n}\n\nexport function parseKimiUsageSnapshot(\n\tvalue: unknown,\n\tnowSeconds = Date.now() / 1000,\n): readonly SubscriptionUsageWindow[] | undefined {\n\tconst payload = record(value);\n\tif (!payload) return undefined;\n\tconst windows: SubscriptionUsageWindow[] = [];\n\tconst total = usageRatio(record(payload.usage), nowSeconds);\n\tif (total) windows.push(withLabel(\"TOTAL\", total));\n\tif (Array.isArray(payload.limits)) {\n\t\tfor (const rawLimit of payload.limits.slice(0, MAX_USAGE_LIMITS)) {\n\t\t\tconst limit = record(rawLimit);\n\t\t\tif (!limit) continue;\n\t\t\tconst detail = record(limit.detail) ?? limit;\n\t\t\tconst windowData = record(limit.window);\n\t\t\tconst parsed = usageRatio(detail, nowSeconds);\n\t\t\tif (!parsed) continue;\n\t\t\tconst resetsAt = parseReset(windowData, nowSeconds) ?? parsed.resetsAt;\n\t\t\twindows.push({\n\t\t\t\tlabel: durationLabel(windowData) ?? shortLabel(limit.name ?? limit.title ?? limit.scope, windows.length),\n\t\t\t\tusedPercent: parsed.usedPercent,\n\t\t\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t\t\t});\n\t\t}\n\t}\n\treturn windows.length > 0 ? windows.slice(0, 4) : undefined;\n}\n\n/** Result of one `qwencloud` invocation; `missing` means the CLI is not installed. */\nexport type QwenCliResult =\n\t| { readonly kind: \"missing\" }\n\t| { readonly kind: \"ran\"; readonly exitCode: number; readonly stdout: string };\nexport type QwenCliRunner = (args: readonly string[], timeoutMs: number) => Promise<QwenCliResult>;\n\nexport function buildQwenCliEnvironment(source: NodeJS.ProcessEnv): NodeJS.ProcessEnv {\n\tconst childEnvironment: NodeJS.ProcessEnv = {};\n\tfor (const key of QWEN_CLI_ENV_KEYS) {\n\t\tconst value = source[key];\n\t\tif (value !== undefined) childEnvironment[key] = value;\n\t}\n\treturn childEnvironment;\n}\n\n/** Parse `qwencloud usage summary --format json` → the token-plan 7-day window. */\nexport function parseQwenTokenPlanUsage(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst tokenPlan = record(record(value)?.token_plan);\n\tif (!tokenPlan || tokenPlan.subscribed === false) return undefined;\n\tconst resetsAt = epochSeconds(tokenPlan.resetDate);\n\tlet usedPercent = finiteNumber(tokenPlan.usedPct) ?? finiteNumber(tokenPlan.used_pct);\n\tif (usedPercent === undefined) {\n\t\tconst total = finiteNumber(tokenPlan.totalCredits);\n\t\tconst remaining = finiteNumber(tokenPlan.remainingCredits);\n\t\tif (total !== undefined && remaining !== undefined && total > 0) {\n\t\t\tusedPercent = ((total - remaining) / total) * 100;\n\t\t}\n\t}\n\tif (usedPercent === undefined) return undefined;\n\treturn [{ label: \"7D\", usedPercent: clampPercent(usedPercent), ...(resetsAt === undefined ? {} : { resetsAt }) }];\n}\n\nexport async function fetchQwenTokenPlanUsage(\n\tlabel: string,\n\trunner: QwenCliRunner = runQwenCli,\n): Promise<SubscriptionUsageSnapshot> {\n\tconst unavailable = (message: string): SubscriptionUsageSnapshot => ({ label, windows: [], message });\n\ttry {\n\t\tconst result = await runner(QWEN_CLI_ARGS, QWEN_CLI_TIMEOUT_MS);\n\t\tif (result.kind === \"missing\") return unavailable(QWEN_CONNECT_HINT);\n\t\tif (result.exitCode === 2) return unavailable(\"run: qwencloud auth login\");\n\t\tif (result.exitCode !== 0 || result.stdout.length > MAX_USAGE_RESPONSE_BYTES) {\n\t\t\treturn unavailable(\"usage unavailable\");\n\t\t}\n\t\tconst payload: unknown = JSON.parse(result.stdout);\n\t\tconst tokenPlan = record(record(payload)?.token_plan);\n\t\tif (tokenPlan?.subscribed === false) {\n\t\t\tconst periodTo = record(record(payload)?.period)?.to;\n\t\t\tconst month = typeof periodTo === \"string\" ? periodTo.slice(0, 7) : new Date().toISOString().slice(0, 7);\n\t\t\treturn unavailable(await billingFallbackMessage(runner, month, payload));\n\t\t}\n\t\tconst windows = parseQwenTokenPlanUsage(payload);\n\t\treturn windows ? { label, windows } : unavailable(\"usage unavailable\");\n\t} catch {\n\t\treturn unavailable(\"usage unavailable\");\n\t}\n}\n\n/** No active token plan: fall back to settled billing — subscription charge + PAYG spend — when the CLI reports it. */\nasync function billingFallbackMessage(runner: QwenCliRunner, month: string, summaryPayload: unknown): Promise<string> {\n\ttry {\n\t\tconst result = await runner(qwenBillingArgs(month), QWEN_CLI_TIMEOUT_MS);\n\t\tif (result.kind !== \"ran\" || result.exitCode !== 0 || result.stdout.length > MAX_USAGE_RESPONSE_BYTES) {\n\t\t\treturn paygFallbackMessage(summaryPayload);\n\t\t}\n\t\tconst billing = parseBillingBreakdown(JSON.parse(result.stdout));\n\t\tconst cur = billing.currency === undefined || billing.currency === \"USD\" ? \"$\" : `${billing.currency} `;\n\t\tconst parts: string[] = [];\n\t\tif (billing.subscription !== undefined && billing.subscription > 0) {\n\t\t\tparts.push(`subscription ${cur}${billing.subscription.toFixed(2)}`);\n\t\t}\n\t\tconst payg = billing.payg ?? paygTotalCost(summaryPayload);\n\t\tif (payg !== undefined && payg > 0) parts.push(`PAYG ${cur}${payg.toFixed(2)}`);\n\t\tif (parts.length === 0) return \"no active token plan\";\n\t\treturn `${parts.join(\" · \")} (${month})`;\n\t} catch {\n\t\treturn paygFallbackMessage(summaryPayload);\n\t}\n}\n\n/** Split settled billing rows into the fixed subscription charge vs per-model PAYG spend. */\nfunction parseBillingBreakdown(payload: unknown): { subscription?: number; payg?: number; currency?: string } {\n\tconst data = record(payload);\n\tconst rows = Array.isArray(data?.rows) ? data.rows : [];\n\tlet subscription: number | undefined;\n\tlet payg = 0;\n\tlet paygSeen = false;\n\tfor (const raw of rows) {\n\t\tconst row = record(raw);\n\t\tconst amount = finiteNumber(row?.amount);\n\t\tif (amount === undefined) continue;\n\t\tconst key = typeof row?.groupKey === \"string\" ? row.groupKey : \"\";\n\t\tif (key === \"__tax__\") continue;\n\t\tif (key === \"DIMENSION_FILTER_NULL_VALUE\") subscription = (subscription ?? 0) + amount;\n\t\telse {\n\t\t\tpayg += amount;\n\t\t\tpaygSeen = true;\n\t\t}\n\t}\n\tconst currency = typeof data?.currency === \"string\" ? data.currency : undefined;\n\treturn { subscription, payg: paygSeen ? payg : undefined, currency };\n}\n\nfunction paygTotalCost(payload: unknown): number | undefined {\n\treturn finiteNumber(record(record(record(payload)?.pay_as_you_go)?.total)?.cost);\n}\n\n/** No active token plan: fall back to the account's pay-as-you-go spend when the CLI reports it. */\nfunction paygFallbackMessage(payload: unknown): string {\n\tconst cost = paygTotalCost(payload);\n\tif (cost === undefined) return \"no active token plan\";\n\tconst currency = \"$\";\n\tconst period = record(record(payload)?.period);\n\tconst from = typeof period?.from === \"string\" ? period.from.slice(0, 7) : undefined;\n\treturn `no token plan · PAYG ${currency}${cost.toFixed(2)}${from ? ` (${from})` : \"\"}`;\n}\n\nfunction runQwenCli(args: readonly string[], timeoutMs: number): Promise<QwenCliResult> {\n\treturn new Promise((resolve) => {\n\t\tconst binary = process.env.QWENCLOUD_CLI ?? \"qwencloud\";\n\t\texecFile(\n\t\t\tbinary,\n\t\t\t[...args],\n\t\t\t{\n\t\t\t\ttimeout: timeoutMs,\n\t\t\t\tmaxBuffer: MAX_USAGE_RESPONSE_BYTES,\n\t\t\t\twindowsHide: true,\n\t\t\t\tenv: buildQwenCliEnvironment(process.env),\n\t\t\t},\n\t\t\t(error, stdout) => {\n\t\t\t\tif (!error) {\n\t\t\t\t\tresolve({ kind: \"ran\", exitCode: 0, stdout });\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst errno = (error as NodeJS.ErrnoException).code;\n\t\t\t\tif (errno === \"ENOENT\") {\n\t\t\t\t\tresolve({ kind: \"missing\" });\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tresolve({ kind: \"ran\", exitCode: typeof errno === \"number\" ? errno : 1, stdout: stdout ?? \"\" });\n\t\t\t},\n\t\t);\n\t});\n}\n\nexport function parseGrokUsageSnapshot(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst config = record(record(value)?.config);\n\tif (!config) return undefined;\n\tconst period = record(config.currentPeriod);\n\tconst resetsAt = epochSeconds(period?.end) ?? epochSeconds(config.billingPeriodEnd);\n\tconst percent = finiteNumber(config.creditUsagePercent);\n\tconst onDemandUsed = finiteNumber(record(config.onDemandUsed)?.val);\n\tconst onDemandCap = finiteNumber(record(config.onDemandCap)?.val);\n\tlet usedPercent = percent;\n\tif (usedPercent === undefined && onDemandUsed !== undefined && onDemandCap !== undefined && onDemandCap > 0) {\n\t\tusedPercent = (onDemandUsed / onDemandCap) * 100;\n\t}\n\tif (usedPercent === undefined && (period || resetsAt !== undefined)) usedPercent = 0;\n\tif (usedPercent === undefined) return undefined;\n\tconst periodType = typeof period?.type === \"string\" ? period.type : \"\";\n\tconst label = periodType.includes(\"MONTHLY\") ? \"30D\" : \"7D\";\n\treturn [{ label, usedPercent: clampPercent(usedPercent), ...(resetsAt === undefined ? {} : { resetsAt }) }];\n}\n\nexport function parseZaiUsageSnapshot(value: unknown): readonly SubscriptionUsageWindow[] | undefined {\n\tconst payload = record(value);\n\tconst limits = record(payload?.data)?.limits;\n\tif (payload?.success !== true || !Array.isArray(limits)) return undefined;\n\tconst parsed = limits.slice(0, MAX_USAGE_LIMITS).map(zaiWindow).filter(isDefined);\n\tconst requestWindows = parsed.filter((window) => window.type === \"TIME_LIMIT\" && !window.featureOnly);\n\tconst selected =\n\t\trequestWindows.length > 0 ? requestWindows : parsed.filter((window) => window.type === \"TOKENS_LIMIT\");\n\tselected.sort((left, right) => left.durationSeconds - right.durationSeconds);\n\tconst windows = selected\n\t\t.slice(0, 4)\n\t\t.map(({ type: _type, durationSeconds: _duration, featureOnly: _feature, ...window }) => window);\n\treturn windows.length > 0 ? windows : undefined;\n}\n\nexport function recordCodexPassiveUsage(apiKey: string, snapshot: ProviderRateLimitSnapshot, nowMs = Date.now()): void {\n\tconst accountId = codexAccountId(apiKey);\n\tconst limitId = snapshot.limitId?.trim().toLowerCase().replace(/-/g, \"_\");\n\tif (!accountId || !Number.isFinite(nowMs) || nowMs < 0 || (limitId && limitId !== \"codex\")) return;\n\tprunePassiveCodexUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(accountId);\n\tconst primary = normalizePassiveCodexWindow(snapshot.primary, nowMs);\n\tconst secondary = normalizePassiveCodexWindow(snapshot.secondary, nowMs);\n\tif (!primary && !secondary) return;\n\n\tconst previous = passiveCodexUsage.get(cacheKey);\n\tconst next: PassiveUsageEntry = {\n\t\tprimary: primary ? { window: primary, observedAt: nowMs } : freshObservedCodexWindow(previous?.primary, nowMs),\n\t\tsecondary: secondary\n\t\t\t? { window: secondary, observedAt: nowMs }\n\t\t\t: freshObservedCodexWindow(previous?.secondary, nowMs),\n\t};\n\tpassiveCodexUsage.delete(cacheKey);\n\tpassiveCodexUsage.set(cacheKey, next);\n\twhile (passiveCodexUsage.size > MAX_PASSIVE_ACCOUNTS) {\n\t\tconst oldestCacheKey = passiveCodexUsage.keys().next().value;\n\t\tif (oldestCacheKey === undefined) break;\n\t\tpassiveCodexUsage.delete(oldestCacheKey);\n\t}\n\tbumpSubscriptionUsageRevision(\"openai-codex\");\n}\n\nexport function recordClaudePassiveUsage(\n\tapiKey: string,\n\tsnapshot: ProviderRateLimitSnapshot,\n\tnowMs = Date.now(),\n): void {\n\tconst limitId = snapshot.limitId?.trim().toLowerCase().replace(/-/g, \"_\");\n\tif (\n\t\t!apiKey ||\n\t\tapiKey.length > 32_768 ||\n\t\t!Number.isFinite(nowMs) ||\n\t\tnowMs < 0 ||\n\t\t(limitId && limitId !== \"anthropic_unified\")\n\t) {\n\t\treturn;\n\t}\n\tprunePassiveClaudeUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(apiKey);\n\tconst primary = normalizePassiveCodexWindow(snapshot.primary, nowMs);\n\tconst secondary = normalizePassiveCodexWindow(snapshot.secondary, nowMs);\n\tif (!primary && !secondary) return;\n\n\tconst previous = passiveClaudeUsage.get(cacheKey);\n\tconst next: PassiveUsageEntry = {\n\t\tprimary: primary ? { window: primary, observedAt: nowMs } : freshObservedCodexWindow(previous?.primary, nowMs),\n\t\tsecondary: secondary\n\t\t\t? { window: secondary, observedAt: nowMs }\n\t\t\t: freshObservedCodexWindow(previous?.secondary, nowMs),\n\t};\n\tpassiveClaudeUsage.delete(cacheKey);\n\tpassiveClaudeUsage.set(cacheKey, next);\n\twhile (passiveClaudeUsage.size > MAX_PASSIVE_ACCOUNTS) {\n\t\tconst oldestCacheKey = passiveClaudeUsage.keys().next().value;\n\t\tif (oldestCacheKey === undefined) break;\n\t\tpassiveClaudeUsage.delete(oldestCacheKey);\n\t}\n\tbumpSubscriptionUsageRevision(\"anthropic\");\n}\n\nasync function fetchCodexUsage(\n\tlabel: string,\n\tapiKey: string,\n\tfetchImpl: FetchLike,\n): Promise<SubscriptionUsageSnapshot> {\n\tconst accountId = codexAccountId(apiKey);\n\tif (!accountId) return { label, windows: [], message: \"usage unavailable\" };\n\tconst passive = passiveCodexSnapshot(apiKey);\n\tlet polled: CodexUsageSnapshot | undefined;\n\ttry {\n\t\tconst payload = await fetchJson(fetchImpl, CODEX_USAGE_URL, {\n\t\t\theaders: {\n\t\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\t\tAccept: \"application/json\",\n\t\t\t\t\"chatgpt-account-id\": accountId,\n\t\t\t\t\"User-Agent\": \"omk\",\n\t\t\t},\n\t\t});\n\t\tpolled = parseCodexUsageSnapshot(payload);\n\t} catch (error) {\n\t\tif (!passive) throw error;\n\t}\n\tconst merged: CodexUsageSnapshot = {\n\t\tfiveHour: polled?.fiveHour ?? passive?.fiveHour,\n\t\tsevenDay: polled?.sevenDay ?? passive?.sevenDay,\n\t};\n\tif (!merged.fiveHour && !merged.sevenDay) return { label, windows: [], message: \"usage unavailable\" };\n\treturn {\n\t\tlabel,\n\t\twindows: [\n\t\t\tmerged.fiveHour ? withLabel(\"5H\", merged.fiveHour) : undefined,\n\t\t\tmerged.sevenDay ? withLabel(\"7D\", merged.sevenDay) : undefined,\n\t\t].filter(isDefined),\n\t};\n}\n\nasync function fetchClaudeUsage(\n\tlabel: string,\n\tapiKey: string,\n\tfetchImpl: FetchLike,\n): Promise<SubscriptionUsageSnapshot> {\n\tlet passive = passiveClaudeSnapshot(apiKey);\n\tconst response = await fetchJsonResult(fetchImpl, CLAUDE_USAGE_URL, {\n\t\theaders: {\n\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"anthropic-beta\": \"claude-code-20250219,oauth-2025-04-20\",\n\t\t\t\"User-Agent\": \"claude-cli/2.1.177 (external, cli)\",\n\t\t},\n\t});\n\tconst windows = [...(parseClaudeUsageSnapshot(response.payload) ?? [])];\n\tif (response.status === 429 && (!passive?.fiveHour || !passive?.sevenDay)) {\n\t\tawait probeClaudeQuota(apiKey, fetchImpl);\n\t\tpassive = passiveClaudeSnapshot(apiKey);\n\t}\n\tfor (const passiveWindow of [\n\t\tpassive?.fiveHour ? withLabel(\"5H\", passive.fiveHour) : undefined,\n\t\tpassive?.sevenDay ? withLabel(\"7D\", passive.sevenDay) : undefined,\n\t].filter(isDefined)) {\n\t\tif (!windows.some((window) => window.label === passiveWindow.label)) windows.push(passiveWindow);\n\t}\n\treturn windows.length > 0\n\t\t? { label, windows }\n\t\t: { label, windows: [], message: response.status === 429 ? \"rate limited · retry later\" : \"usage unavailable\" };\n}\n\n/**\n * Claude Code 2.1.177 performs the same `quota`/`max_tokens: 1` request during startup.\n * Keep it account-scoped and hourly so a pinned sidebar cannot turn the fallback into polling.\n */\nasync function probeClaudeQuota(apiKey: string, fetchImpl: FetchLike, nowMs = Date.now()): Promise<void> {\n\tif (!apiKey || apiKey.length > 32_768 || !Number.isFinite(nowMs) || nowMs < 0) return;\n\tconst cacheKey = passiveCredentialCacheKey(apiKey);\n\tconst activeProbe = claudeQuotaProbesInFlight.get(cacheKey);\n\tif (activeProbe) return activeProbe;\n\tconst lastAttempt = claudeQuotaProbeAttempts.get(cacheKey);\n\tif (lastAttempt !== undefined && nowMs - lastAttempt < CLAUDE_QUOTA_PROBE_COOLDOWN_MS) return;\n\tif (claudeQuotaProbesInFlight.size >= MAX_PASSIVE_ACCOUNTS) return;\n\n\tclaudeQuotaProbeAttempts.delete(cacheKey);\n\tclaudeQuotaProbeAttempts.set(cacheKey, nowMs);\n\twhile (claudeQuotaProbeAttempts.size > MAX_PASSIVE_ACCOUNTS) {\n\t\tconst oldestCacheKey = claudeQuotaProbeAttempts.keys().next().value;\n\t\tif (oldestCacheKey === undefined) break;\n\t\tclaudeQuotaProbeAttempts.delete(oldestCacheKey);\n\t}\n\n\tconst probe = (async () => {\n\t\ttry {\n\t\t\tconst response = await fetchImpl(CLAUDE_MESSAGES_URL, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: {\n\t\t\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\t\t\tAccept: \"application/json\",\n\t\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t\t\t\"anthropic-version\": \"2023-06-01\",\n\t\t\t\t\t\"anthropic-beta\": \"claude-code-20250219,oauth-2025-04-20\",\n\t\t\t\t\t\"anthropic-dangerous-direct-browser-access\": \"true\",\n\t\t\t\t\t\"User-Agent\": \"claude-cli/2.1.177 (external, cli)\",\n\t\t\t\t\t\"x-app\": \"cli\",\n\t\t\t\t},\n\t\t\t\tbody: JSON.stringify({\n\t\t\t\t\tmodel: \"claude-haiku-4-5\",\n\t\t\t\t\tmax_tokens: 1,\n\t\t\t\t\tmessages: [{ role: \"user\", content: \"quota\" }],\n\t\t\t\t}),\n\t\t\t\tsignal: AbortSignal.timeout(CLAUDE_QUOTA_PROBE_TIMEOUT_MS),\n\t\t\t});\n\t\t\tconst snapshot = parseClaudeQuotaProbeHeaders(response.headers, nowMs);\n\t\t\tif (snapshot) recordClaudePassiveUsage(apiKey, snapshot, nowMs);\n\t\t\ttry {\n\t\t\t\tawait response.body?.cancel();\n\t\t\t} catch {\n\t\t\t\t// The quota snapshot lives entirely in response headers.\n\t\t\t}\n\t\t} catch {\n\t\t\t// This mirrors Claude Code's best-effort startup quota check.\n\t\t}\n\t})().finally(() => {\n\t\tclaudeQuotaProbesInFlight.delete(cacheKey);\n\t});\n\tclaudeQuotaProbesInFlight.set(cacheKey, probe);\n\treturn probe;\n}\n\nfunction parseClaudeQuotaProbeHeaders(headers: Headers, nowMs: number): ProviderRateLimitSnapshot | undefined {\n\tconst parseWindow = (prefix: \"5h\" | \"7d\", windowSeconds: number): ProviderRateLimitWindow | undefined => {\n\t\tconst rawUtilization = headers.get(`anthropic-ratelimit-unified-${prefix}-utilization`);\n\t\tconst rawReset = headers.get(`anthropic-ratelimit-unified-${prefix}-reset`);\n\t\tif (rawUtilization === null || rawReset === null) return undefined;\n\t\tconst utilization = Number(rawUtilization);\n\t\tconst resetsAt = Number(rawReset);\n\t\tconst maxResetAt = nowMs / 1000 + 10 * 365 * 24 * 60 * 60;\n\t\tif (\n\t\t\t!Number.isFinite(utilization) ||\n\t\t\tutilization < 0 ||\n\t\t\tutilization > 1 ||\n\t\t\t!Number.isSafeInteger(resetsAt) ||\n\t\t\tresetsAt <= 0 ||\n\t\t\tresetsAt > maxResetAt\n\t\t) {\n\t\t\treturn undefined;\n\t\t}\n\t\treturn { usedPercent: Math.round(utilization * 10_000) / 100, windowSeconds, resetsAt };\n\t};\n\tconst primary = parseWindow(\"5h\", FIVE_HOUR_SECONDS);\n\tconst secondary = parseWindow(\"7d\", SEVEN_DAY_SECONDS);\n\treturn primary || secondary ? { limitId: \"anthropic-unified\", primary, secondary } : undefined;\n}\n\nasync function fetchGrokUsage(label: string, apiKey: string, fetchImpl: FetchLike): Promise<SubscriptionUsageSnapshot> {\n\tconst payload = await fetchJson(fetchImpl, GROK_BILLING_URL, {\n\t\theaders: {\n\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"x-xai-token-auth\": \"xai-grok-cli\",\n\t\t\t\"User-Agent\": \"omk\",\n\t\t},\n\t});\n\tconst windows = parseGrokUsageSnapshot(payload);\n\treturn windows ? { label, windows } : { label, windows: [], message: \"usage unavailable\" };\n}\n\nasync function fetchKimiUsage(label: string, apiKey: string, fetchImpl: FetchLike): Promise<SubscriptionUsageSnapshot> {\n\tconst payload = await fetchJson(fetchImpl, KIMI_USAGE_URL, {\n\t\theaders: {\n\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"User-Agent\": \"KimiCLI/1.5\",\n\t\t\t\"X-Msh-Platform\": \"kimi_cli\",\n\t\t\t\"X-Msh-Version\": \"1.5\",\n\t\t},\n\t});\n\tconst windows = parseKimiUsageSnapshot(payload);\n\treturn windows ? { label, windows } : { label, windows: [], message: \"usage unavailable\" };\n}\n\nasync function fetchZaiUsage(\n\tlabel: string,\n\tmodelProvider: string,\n\tcredential: { readonly provider: string; readonly apiKey: string },\n\tfetchImpl: FetchLike,\n): Promise<SubscriptionUsageSnapshot> {\n\tconst origin =\n\t\tmodelProvider === \"zai-coding-cn\" ||\n\t\tcredential.provider === \"zai-coding-cn\" ||\n\t\tcredential.provider === \"zhipu-coding-plan\"\n\t\t\t? \"https://open.bigmodel.cn\"\n\t\t\t: \"https://api.z.ai\";\n\tconst payload = await fetchJson(fetchImpl, `${origin}/api/monitor/usage/quota/limit`, {\n\t\theaders: {\n\t\t\tAuthorization: credential.apiKey,\n\t\t\tAccept: \"application/json\",\n\t\t\t\"User-Agent\": \"omk\",\n\t\t},\n\t});\n\tconst windows = parseZaiUsageSnapshot(payload);\n\treturn windows ? { label, windows } : { label, windows: [], message: \"usage unavailable\" };\n}\n\nasync function fetchJson(fetchImpl: FetchLike, url: string, init: RequestInit): Promise<unknown> {\n\treturn (await fetchJsonResult(fetchImpl, url, init)).payload;\n}\n\nasync function fetchJsonResult(fetchImpl: FetchLike, url: string, init: RequestInit): Promise<FetchJsonResult> {\n\tconst response = await fetchImpl(url, { ...init, signal: AbortSignal.timeout(10_000) });\n\tif (!response.ok) return { status: response.status };\n\tconst declaredBytes = Number(response.headers?.get(\"content-length\"));\n\tif (Number.isFinite(declaredBytes) && declaredBytes > MAX_USAGE_RESPONSE_BYTES) {\n\t\treturn { status: response.status };\n\t}\n\tif (!response.body || typeof response.body.getReader !== \"function\") {\n\t\treturn { status: response.status, payload: await response.json() };\n\t}\n\n\tconst reader = response.body.getReader();\n\tconst chunks: Uint8Array[] = [];\n\tlet totalBytes = 0;\n\twhile (true) {\n\t\tconst { done, value } = await reader.read();\n\t\tif (done) break;\n\t\ttotalBytes += value.byteLength;\n\t\tif (totalBytes > MAX_USAGE_RESPONSE_BYTES) {\n\t\t\tawait reader.cancel();\n\t\t\treturn { status: response.status };\n\t\t}\n\t\tchunks.push(value);\n\t}\n\tconst body = new Uint8Array(totalBytes);\n\tlet offset = 0;\n\tfor (const chunk of chunks) {\n\t\tbody.set(chunk, offset);\n\t\toffset += chunk.byteLength;\n\t}\n\ttry {\n\t\treturn { status: response.status, payload: JSON.parse(new TextDecoder().decode(body)) as unknown };\n\t} catch {\n\t\treturn { status: response.status };\n\t}\n}\n\nfunction credentialConfigured(session: UsageSession, candidate: CredentialCandidate): boolean {\n\tif (candidate.oauthOnly) return session.modelRegistry.isUsingOAuthProvider(candidate.provider);\n\tconst status = session.modelRegistry.getProviderAuthStatus(candidate.provider);\n\treturn status.configured || status.source !== undefined;\n}\n\nasync function resolveCredential(\n\tsession: UsageSession,\n\tusageSource: SubscriptionUsageSource,\n): Promise<{ readonly provider: string; readonly apiKey: string } | undefined> {\n\tfor (const candidate of usageSource.credentials) {\n\t\tif (!credentialConfigured(session, candidate)) continue;\n\t\tconst apiKey = await session.modelRegistry.getApiKeyForProvider(candidate.provider);\n\t\tif (apiKey) return { provider: candidate.provider, apiKey };\n\t}\n\treturn undefined;\n}\n\nfunction codexWindow(value: unknown, nowSeconds: number): ParsedCodexWindow | undefined {\n\tconst window = record(value);\n\tconst usedPercent = finiteNumber(window?.used_percent);\n\tif (usedPercent === undefined) return undefined;\n\tconst windowSeconds = finiteNumber(window?.limit_window_seconds);\n\tconst resetAt = finiteNumber(window?.reset_at);\n\tconst resetAfter = finiteNumber(window?.reset_after_seconds);\n\tconst resetsAt = resetAt ?? (resetAfter === undefined ? undefined : nowSeconds + resetAfter);\n\treturn {\n\t\tusedPercent: clampPercent(usedPercent),\n\t\t...(windowSeconds === undefined ? {} : { windowSeconds }),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction claudeWindow(value: unknown): Omit<SubscriptionUsageWindow, \"label\"> | undefined {\n\tconst bucket = record(value);\n\tif (!bucket || bucket.is_active === false) return undefined;\n\tconst usedPercent = finiteNumber(bucket.utilization ?? bucket.percent);\n\tif (usedPercent === undefined) return undefined;\n\tconst resetsAt = epochSeconds(bucket.resets_at);\n\treturn { usedPercent: clampPercent(usedPercent), ...(resetsAt === undefined ? {} : { resetsAt }) };\n}\n\nfunction usageRatio(\n\tdata: Record<string, unknown> | undefined,\n\tnowSeconds: number,\n): Omit<SubscriptionUsageWindow, \"label\"> | undefined {\n\tif (!data) return undefined;\n\tconst limit = finiteNumber(data.limit);\n\tlet used = finiteNumber(data.used);\n\tconst remaining = finiteNumber(data.remaining);\n\tif (used === undefined && limit !== undefined && remaining !== undefined) used = limit - remaining;\n\tif (used === undefined || limit === undefined || limit <= 0) return undefined;\n\tconst resetsAt = parseReset(data, nowSeconds);\n\treturn {\n\t\tusedPercent: clampPercent((used / limit) * 100),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction zaiWindow(value: unknown):\n\t| (SubscriptionUsageWindow & {\n\t\t\treadonly type: string;\n\t\t\treadonly durationSeconds: number;\n\t\t\treadonly featureOnly: boolean;\n\t  })\n\t| undefined {\n\tconst item = record(value);\n\tif (!item || (item.type !== \"TIME_LIMIT\" && item.type !== \"TOKENS_LIMIT\")) return undefined;\n\tconst percentage = finiteNumber(item.percentage);\n\tconst current = finiteNumber(item.currentValue);\n\tconst limit = finiteNumber(item.usage);\n\tconst usedPercent =\n\t\tpercentage ?? (current !== undefined && limit !== undefined && limit > 0 ? (current / limit) * 100 : undefined);\n\tif (usedPercent === undefined) return undefined;\n\tconst duration = zaiDuration(finiteNumber(item.unit), finiteNumber(item.number));\n\tconst resetsAt = epochSeconds(item.nextResetTime);\n\tconst detailCodes = Array.isArray(item.usageDetails)\n\t\t? item.usageDetails\n\t\t\t\t.map((detail) => record(detail)?.modelCode)\n\t\t\t\t.filter((code): code is string => typeof code === \"string\")\n\t\t: [];\n\treturn {\n\t\ttype: item.type,\n\t\tlabel: duration.label,\n\t\tdurationSeconds: duration.seconds,\n\t\tfeatureOnly: [\"search-prime\", \"web-reader\", \"zread\"].every((code) => detailCodes.includes(code)),\n\t\tusedPercent: clampPercent(usedPercent),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction zaiDuration(unit: number | undefined, countValue: number | undefined): { label: string; seconds: number } {\n\tconst count = countValue && countValue > 0 ? countValue : 1;\n\tif (unit === 3) return { label: `${count}H`, seconds: count * 60 * 60 };\n\tif (unit === 4) return { label: `${count}D`, seconds: count * 24 * 60 * 60 };\n\tif (unit === 5) return { label: count === 1 ? \"30D\" : `${count}MO`, seconds: count * 30 * 24 * 60 * 60 };\n\tif (unit === 6) return { label: \"7D\", seconds: 7 * 24 * 60 * 60 };\n\treturn { label: \"QUOTA\", seconds: Number.POSITIVE_INFINITY };\n}\n\nfunction durationLabel(window: Record<string, unknown> | undefined): string | undefined {\n\tif (!window) return undefined;\n\tconst duration = finiteNumber(window.duration);\n\tconst unit = typeof window.timeUnit === \"string\" ? window.timeUnit.toUpperCase() : \"\";\n\tif (duration === undefined) return undefined;\n\tif (unit.includes(\"MINUTE\")) return duration % 60 === 0 ? `${duration / 60}H` : `${duration}M`;\n\tif (unit.includes(\"HOUR\")) return `${duration}H`;\n\tif (unit.includes(\"DAY\")) return `${duration}D`;\n\tif (unit.includes(\"SECOND\")) return `${duration}S`;\n\treturn undefined;\n}\n\nfunction parseReset(data: Record<string, unknown> | undefined, nowSeconds: number): number | undefined {\n\tif (!data) return undefined;\n\tfor (const key of [\"reset_at\", \"resetAt\", \"reset_time\", \"resetTime\"] as const) {\n\t\tconst parsed = epochSeconds(data[key]);\n\t\tif (parsed !== undefined) return parsed;\n\t}\n\tfor (const key of [\"reset_in\", \"resetIn\", \"ttl\"] as const) {\n\t\tconst seconds = finiteNumber(data[key]);\n\t\tif (seconds !== undefined) return nowSeconds + seconds;\n\t}\n\treturn undefined;\n}\n\nfunction epochSeconds(value: unknown): number | undefined {\n\tconst numeric = finiteNumber(value);\n\tif (numeric !== undefined) return numeric > 1_000_000_000_000 ? numeric / 1000 : numeric;\n\tif (typeof value !== \"string\" || !value.trim()) return undefined;\n\tconst parsed = Date.parse(value);\n\treturn Number.isFinite(parsed) ? parsed / 1000 : undefined;\n}\n\nfunction normalizePassiveCodexWindow(\n\twindow: ProviderRateLimitWindow | undefined,\n\tnowMs: number,\n): ParsedCodexWindow | undefined {\n\tif (!window || !Number.isFinite(window.usedPercent)) return undefined;\n\tconst windowSeconds =\n\t\twindow.windowSeconds !== undefined &&\n\t\tNumber.isFinite(window.windowSeconds) &&\n\t\twindow.windowSeconds > 0 &&\n\t\twindow.windowSeconds <= Number.MAX_SAFE_INTEGER\n\t\t\t? Math.round(window.windowSeconds)\n\t\t\t: undefined;\n\tconst maxResetAt = nowMs / 1000 + 10 * 365 * 24 * 60 * 60;\n\tconst resetsAtValue = window.resetsAt === undefined ? undefined : Math.floor(window.resetsAt);\n\tconst resetsAt =\n\t\tresetsAtValue !== undefined &&\n\t\tNumber.isSafeInteger(resetsAtValue) &&\n\t\tresetsAtValue > 0 &&\n\t\tresetsAtValue <= maxResetAt\n\t\t\t? resetsAtValue\n\t\t\t: undefined;\n\treturn {\n\t\tusedPercent: clampPercent(window.usedPercent),\n\t\t...(windowSeconds === undefined ? {} : { windowSeconds }),\n\t\t...(resetsAt === undefined ? {} : { resetsAt }),\n\t};\n}\n\nfunction passiveCredentialCacheKey(identifier: string): string {\n\treturn createHash(\"sha256\").update(identifier).digest(\"base64url\");\n}\n\nfunction freshObservedCodexWindow(\n\tobserved: ObservedCodexWindow | undefined,\n\tnowMs: number,\n): ObservedCodexWindow | undefined {\n\tif (!observed || nowMs - observed.observedAt > PASSIVE_USAGE_TTL_MS) return undefined;\n\tif (observed.window.resetsAt !== undefined && observed.window.resetsAt <= nowMs / 1000) return undefined;\n\treturn observed;\n}\n\nfunction prunePassiveCodexUsage(nowMs: number): void {\n\tfor (const [cacheKey, entry] of passiveCodexUsage) {\n\t\tif (!freshObservedCodexWindow(entry.primary, nowMs) && !freshObservedCodexWindow(entry.secondary, nowMs)) {\n\t\t\tpassiveCodexUsage.delete(cacheKey);\n\t\t}\n\t}\n}\n\nfunction prunePassiveClaudeUsage(nowMs: number): void {\n\tfor (const [cacheKey, entry] of passiveClaudeUsage) {\n\t\tif (!freshObservedCodexWindow(entry.primary, nowMs) && !freshObservedCodexWindow(entry.secondary, nowMs)) {\n\t\t\tpassiveClaudeUsage.delete(cacheKey);\n\t\t}\n\t}\n}\n\nfunction passiveCodexSnapshot(apiKey: string, nowMs = Date.now()): CodexUsageSnapshot | undefined {\n\tconst accountId = codexAccountId(apiKey);\n\tif (!accountId || !Number.isFinite(nowMs) || nowMs < 0) return undefined;\n\tprunePassiveCodexUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(accountId);\n\tconst entry = passiveCodexUsage.get(cacheKey);\n\tif (!entry) return undefined;\n\tconst primary = freshObservedCodexWindow(entry.primary, nowMs);\n\tconst secondary = freshObservedCodexWindow(entry.secondary, nowMs);\n\tif (!primary && !secondary) {\n\t\tpassiveCodexUsage.delete(cacheKey);\n\t\treturn undefined;\n\t}\n\tconst freshEntry = { primary, secondary };\n\tpassiveCodexUsage.delete(cacheKey);\n\tpassiveCodexUsage.set(cacheKey, freshEntry);\n\n\tconst windows = [primary?.window, secondary?.window].filter(isDefined);\n\tconst fiveHour = windows.find((window) => near(window.windowSeconds, FIVE_HOUR_SECONDS));\n\tconst sevenDay = windows.find((window) => near(window.windowSeconds, SEVEN_DAY_SECONDS));\n\tif (!fiveHour && !sevenDay) return undefined;\n\treturn {\n\t\t...(fiveHour ? { fiveHour: publicCodexWindow(fiveHour) } : {}),\n\t\t...(sevenDay ? { sevenDay: publicCodexWindow(sevenDay) } : {}),\n\t};\n}\n\nfunction passiveClaudeSnapshot(apiKey: string, nowMs = Date.now()): CodexUsageSnapshot | undefined {\n\tif (!apiKey || apiKey.length > 32_768 || !Number.isFinite(nowMs) || nowMs < 0) return undefined;\n\tprunePassiveClaudeUsage(nowMs);\n\tconst cacheKey = passiveCredentialCacheKey(apiKey);\n\tconst entry = passiveClaudeUsage.get(cacheKey);\n\tif (!entry) return undefined;\n\tconst primary = freshObservedCodexWindow(entry.primary, nowMs);\n\tconst secondary = freshObservedCodexWindow(entry.secondary, nowMs);\n\tif (!primary && !secondary) {\n\t\tpassiveClaudeUsage.delete(cacheKey);\n\t\treturn undefined;\n\t}\n\tpassiveClaudeUsage.delete(cacheKey);\n\tpassiveClaudeUsage.set(cacheKey, { primary, secondary });\n\n\tconst windows = [primary?.window, secondary?.window].filter(isDefined);\n\tconst fiveHour = windows.find((window) => near(window.windowSeconds, FIVE_HOUR_SECONDS));\n\tconst sevenDay = windows.find((window) => near(window.windowSeconds, SEVEN_DAY_SECONDS));\n\tif (!fiveHour && !sevenDay) return undefined;\n\treturn {\n\t\t...(fiveHour ? { fiveHour: publicCodexWindow(fiveHour) } : {}),\n\t\t...(sevenDay ? { sevenDay: publicCodexWindow(sevenDay) } : {}),\n\t};\n}\n\nfunction codexAccountId(token: string): string | undefined {\n\ttry {\n\t\tconst payload = token.split(\".\")[1];\n\t\tif (!payload) return undefined;\n\t\tconst claims = record(JSON.parse(Buffer.from(payload, \"base64url\").toString(\"utf8\")));\n\t\tconst auth = record(claims?.[OPENAI_AUTH_CLAIM]);\n\t\tconst accountId = typeof auth?.chatgpt_account_id === \"string\" ? auth.chatgpt_account_id.trim() : \"\";\n\t\treturn accountId && accountId.length <= 256 ? accountId : undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nfunction publicCodexWindow(window: ParsedCodexWindow): CodexUsageWindow {\n\treturn { usedPercent: window.usedPercent, ...(window.resetsAt === undefined ? {} : { resetsAt: window.resetsAt }) };\n}\n\nfunction withLabel(label: string, window: Omit<SubscriptionUsageWindow, \"label\">): SubscriptionUsageWindow {\n\treturn { label, ...window };\n}\n\nfunction shortLabel(value: unknown, index: number): string {\n\tif (typeof value !== \"string\") return `LIMIT${index + 1}`;\n\tconst safe = value\n\t\t.replace(/[\\u0000-\\u001f\\u007f-\\u009f\\u200b-\\u200f\\u202a-\\u202e\\u2060-\\u206f]/g, \"\")\n\t\t.replace(/\\s+/g, \" \")\n\t\t.trim();\n\treturn safe ? safe.toUpperCase().slice(0, 8) : `LIMIT${index + 1}`;\n}\n\nfunction finiteNumber(value: unknown): number | undefined {\n\tconst number =\n\t\ttypeof value === \"number\" ? value : typeof value === \"string\" && value.trim() ? Number(value) : Number.NaN;\n\treturn Number.isFinite(number) ? number : undefined;\n}\n\nfunction clampPercent(value: number): number {\n\treturn Number(Math.max(0, Math.min(100, value)).toFixed(2));\n}\n\nfunction near(actual: number | undefined, expected: number): boolean {\n\treturn actual !== undefined && Math.abs(actual - expected) <= WINDOW_TOLERANCE_SECONDS;\n}\n\nfunction record(value: unknown): Record<string, unknown> | undefined {\n\treturn typeof value === \"object\" && value !== null && !Array.isArray(value)\n\t\t? (value as Record<string, unknown>)\n\t\t: undefined;\n}\n\nfunction isDefined<T>(value: T | undefined): value is T {\n\treturn value !== undefined;\n}\n\nfunction offline(): boolean {\n\tconst value = process.env.OMK_OFFLINE?.toLowerCase();\n\treturn value === \"1\" || value === \"true\" || value === \"yes\";\n}\n"]}

package/dist/core/provider-usage.js

@@ -17,6 +17,7 @@ const CLAUDE_QUOTA_PROBE_COOLDOWN_MS = 60 * 60 * 1000;
 const CLAUDE_QUOTA_PROBE_TIMEOUT_MS = 10_000;
 const QWEN_CLI_TIMEOUT_MS = 15_000;
 const QWEN_CLI_ARGS = ["usage", "summary", "--format", "json"];
+const qwenBillingArgs = (month) => ["billing", "breakdown", "--from", month, "--to", month, "--group-by", "model", "--format", "json"];
 const QWEN_CONNECT_HINT = "connect: npm i -g @qwencloud/qwencloud-cli && qwencloud auth login";
 const QWEN_CLI_ENV_KEYS = [
     "PATH",
@@ -275,8 +276,11 @@ export async function fetchQwenTokenPlanUsage(label, runner = runQwenCli) {
         }
         const payload = JSON.parse(result.stdout);
         const tokenPlan = record(record(payload)?.token_plan);
-        if (tokenPlan?.subscribed === false)
-            return unavailable("no active token plan");
+        if (tokenPlan?.subscribed === false) {
+            const periodTo = record(record(payload)?.period)?.to;
+            const month = typeof periodTo === "string" ? periodTo.slice(0, 7) : new Date().toISOString().slice(0, 7);
+            return unavailable(await billingFallbackMessage(runner, month, payload));
+        }
         const windows = parseQwenTokenPlanUsage(payload);
         return windows ? { label, windows } : unavailable("usage unavailable");
     }
@@ -284,6 +288,68 @@ export async function fetchQwenTokenPlanUsage(label, runner = runQwenCli) {
         return unavailable("usage unavailable");
     }
 }
+/** No active token plan: fall back to settled billing — subscription charge + PAYG spend — when the CLI reports it. */
+async function billingFallbackMessage(runner, month, summaryPayload) {
+    try {
+        const result = await runner(qwenBillingArgs(month), QWEN_CLI_TIMEOUT_MS);
+        if (result.kind !== "ran" || result.exitCode !== 0 || result.stdout.length > MAX_USAGE_RESPONSE_BYTES) {
+            return paygFallbackMessage(summaryPayload);
+        }
+        const billing = parseBillingBreakdown(JSON.parse(result.stdout));
+        const cur = billing.currency === undefined || billing.currency === "USD" ? "$" : `${billing.currency} `;
+        const parts = [];
+        if (billing.subscription !== undefined && billing.subscription > 0) {
+            parts.push(`subscription ${cur}${billing.subscription.toFixed(2)}`);
+        }
+        const payg = billing.payg ?? paygTotalCost(summaryPayload);
+        if (payg !== undefined && payg > 0)
+            parts.push(`PAYG ${cur}${payg.toFixed(2)}`);
+        if (parts.length === 0)
+            return "no active token plan";
+        return `${parts.join(" · ")} (${month})`;
+    }
+    catch {
+        return paygFallbackMessage(summaryPayload);
+    }
+}
+/** Split settled billing rows into the fixed subscription charge vs per-model PAYG spend. */
+function parseBillingBreakdown(payload) {
+    const data = record(payload);
+    const rows = Array.isArray(data?.rows) ? data.rows : [];
+    let subscription;
+    let payg = 0;
+    let paygSeen = false;
+    for (const raw of rows) {
+        const row = record(raw);
+        const amount = finiteNumber(row?.amount);
+        if (amount === undefined)
+            continue;
+        const key = typeof row?.groupKey === "string" ? row.groupKey : "";
+        if (key === "__tax__")
+            continue;
+        if (key === "DIMENSION_FILTER_NULL_VALUE")
+            subscription = (subscription ?? 0) + amount;
+        else {
+            payg += amount;
+            paygSeen = true;
+        }
+    }
+    const currency = typeof data?.currency === "string" ? data.currency : undefined;
+    return { subscription, payg: paygSeen ? payg : undefined, currency };
+}
+function paygTotalCost(payload) {
+    return finiteNumber(record(record(record(payload)?.pay_as_you_go)?.total)?.cost);
+}
+/** No active token plan: fall back to the account's pay-as-you-go spend when the CLI reports it. */
+function paygFallbackMessage(payload) {
+    const cost = paygTotalCost(payload);
+    if (cost === undefined)
+        return "no active token plan";
+    const currency = "$";
+    const period = record(record(payload)?.period);
+    const from = typeof period?.from === "string" ? period.from.slice(0, 7) : undefined;
+    return `no token plan · PAYG ${currency}${cost.toFixed(2)}${from ? ` (${from})` : ""}`;
+}
 function runQwenCli(args, timeoutMs) {
     return new Promise((resolve) => {
         const binary = process.env.QWENCLOUD_CLI ?? "qwencloud";