sidekick-shared 0.24.0 → 0.24.2

package/README.md

@@ -69,17 +69,17 @@ npm install sidekick-shared
 
 `sidekick-shared` ships three public entry points plus a few convenience subpaths. Pick the one that matches your runtime.
 
-| Path                           | Runtime                    | What it exposes                                                                                                                                     |
-| ------------------------------ | -------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `sidekick-shared`              | Node (CLI, extension host) | Full public API (readers, providers, parsers, pricing, …).                                                                                          |
-| `sidekick-shared/browser`      | **Browser / webview**      | Pure helpers: usage normalization/pricing, token estimation, transcript and assistant-turn projection, model/context math.                          |
-| `sidekick-shared/node`         | Node only                  | LiteLLM pricing hydration, recent-session/transcript reads, and resilient observed-session collection (`fs` + providers).                           |
-| `sidekick-shared/phrases`      | Any runtime                | Phrase arrays + `getRandomPhrase()`.                                                                                                                |
-| `sidekick-shared/modelContext` | Any runtime                | Direct access to the context-window module.                                                                                                         |
-| `sidekick-shared/modelInfo`    | Any runtime                | Direct access to model parsing and cost math.                                                                                                       |
-| `sidekick-shared/formatting`   | Any runtime                | Direct access to pure token and duration display helpers.                                                                                           |
-| `sidekick-shared/schemas`      | Any runtime                | Pure Zod boundary schemas (session events, assistant turns, observed-session V1, quota, account status, quota history) — fs-free, no Node builtins. |
-| `sidekick-shared/statusline`   | Node only                  | Cache-only statusline surface: account status + quota snapshot readers, `formatStatusline()`, and burn-rate estimates.                              |
+| Path                           | Runtime                    | What it exposes                                                                                                                                                                   |
+| ------------------------------ | -------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `sidekick-shared`              | Node (CLI, extension host) | Full public API (readers, providers, parsers, pricing, …).                                                                                                                        |
+| `sidekick-shared/browser`      | **Browser / webview**      | Pure helpers: usage normalization/pricing, token estimation, transcript and assistant-turn projection, model/context math.                                                        |
+| `sidekick-shared/node`         | Node only                  | LiteLLM pricing and context-window hydration, observed context-window persistence, recent-session/transcript reads, and resilient observed-session collection (`fs` + providers). |
+| `sidekick-shared/phrases`      | Any runtime                | Phrase arrays + `getRandomPhrase()`.                                                                                                                                              |
+| `sidekick-shared/modelContext` | Any runtime                | Direct access to the context-window module.                                                                                                                                       |
+| `sidekick-shared/modelInfo`    | Any runtime                | Direct access to model parsing and cost math.                                                                                                                                     |
+| `sidekick-shared/formatting`   | Any runtime                | Direct access to pure token and duration display helpers.                                                                                                                         |
+| `sidekick-shared/schemas`      | Any runtime                | Pure Zod boundary schemas (session events, assistant turns, observed-session V1, quota, account status, quota history) — fs-free, no Node builtins.                               |
+| `sidekick-shared/statusline`   | Node only                  | Cache-only statusline surface: account status + quota snapshot readers, `formatStatusline()`, and burn-rate estimates.                                                            |
 
 ### Browser / webview runtimes
 
@@ -103,14 +103,23 @@ import {
 
 ### Node / CLI / extension host
 
-Hydrate the pricing catalog from the `node` subpath:
+Hydrate the model catalog from the `node` subpath. One fetch covers both prices and context window sizes:
 
 ```typescript
-import { hydratePricingCatalog } from 'sidekick-shared/node';
+import { hydratePricingCatalog, loadObservedContextWindows } from 'sidekick-shared/node';
 
-await hydratePricingCatalog({ cacheDir: '~/.config/sidekick' });
+const { entries, contextWindowEntries } = await hydratePricingCatalog({
+  cacheDir: '~/.config/sidekick',
+});
+
+// Context windows a provider reported on an earlier run. These outrank the
+// catalog, which only knows each model's published maximum — Codex reports the
+// window your account tier actually gets. Local read, offline-safe.
+await loadObservedContextWindows();
 ```
 
+Both calls always resolve; a network failure, missing cache, or malformed store leaves lookups on the static tables. To record a window a provider reports at runtime, call `recordObservedContextWindow(modelId, window)` — it persists only when the value changes.
+
 ### Normalize and price provider usage
 
 ```typescript
@@ -165,6 +174,18 @@ const transcript = readSessionTranscript(provider, recent.sessionPath, {
 
 The same API works with `ClaudeCodeProvider` and `OpenCodeProvider`. It covers provider sessions discoverable by those readers. Remote Claude SDK-only history and Codex history entries whose rollout file has disappeared are not synthesized.
 
+Canonical transcript messages retain provider-origin facts on `message.source`, including the original role, entrypoint, meta/sidechain flags, and source identifiers. Transcript-level `cwd` and `gitBranch` use the latest non-empty value reported by the provider. Consumers can therefore distinguish a human Claude Code CLI prompt without reparsing JSONL:
+
+```typescript
+const humanCliPrompts = transcript.messages.filter(
+  (message) =>
+    message.source.originalRole === 'user' &&
+    message.source.entrypoint === 'cli' &&
+    message.source.isMeta !== true &&
+    message.source.isSidechain !== true,
+);
+```
+
 For an already-open incremental Codex rollout, keep one `CodexRolloutParser` per stream and pass each parsed `SessionEvent[]` batch to `projectSessionTranscript()` (or append the events to an existing projection). This preserves parser state for paired tool calls, model context, and cumulative token deltas; consumers should not convert raw `response_item` rows themselves.
 
 ### Collect observed sessions with isolation and retry
@@ -181,6 +202,13 @@ const collector = new ObservedSessionCollector({
 await collector.collect(); // scheduling remains the host's responsibility
 ```
 
+Each provider-backed source also exposes the same adapter's `ProviderCapabilitiesV1` record used for its reads and lifecycle:
+
+```typescript
+const source = observedSessionSourceFromProvider(provider, workspacePath);
+applyReadOnlyProjection(source.capabilities);
+```
+
 Discovery and reads are isolated, retries use bounded 30-second-to-5-minute exponential backoff, file changes bypass delay, duplicate identical failures are suppressed, and recovery emits a content-free diagnostic. Clock and fingerprint callbacks are injectable for deterministic hosts/tests.
 
 ## Usage Examples

package/dist/index.d.ts

@@ -10,7 +10,7 @@ export type { TranscriptFidelity, TranscriptSourceProvenance, CanonicalTranscrip
 export { listRecentSessions, readSessionTranscript } from './sessionTranscripts';
 export type { ProviderSessionIndex, ListRecentSessionsOptions } from './sessionTranscripts';
 export { ObservedSessionCollector, observedSessionSourceFromProvider, fileFingerprint, } from './observedSessionCollector';
-export type { ObservedSessionReference, ObservedSessionCollectionSource, ObservedSessionDiagnosticKind, ObservedSessionDiagnostic, ObservedSessionCollection, ObservedSessionCollectorOptions, } from './observedSessionCollector';
+export type { ObservedSessionReference, ObservedSessionCollectionSource, ProviderObservedSessionCollectionSource, ObservedSessionDiagnosticKind, ObservedSessionDiagnostic, ObservedSessionCollection, ObservedSessionCollectorOptions, } from './observedSessionCollector';
 export type { SessionEvent, ClaudeSessionEvent, MessageUsage, SessionMessage, SessionEventBase, MessageSessionEvent, UserSessionEvent, AssistantSessionEvent, SummarySessionEvent, SystemSessionEvent, ToolUseSessionEvent, ToolResultSessionEvent, TokenUsage, ToolCall, ToolAnalytics, TimelineEvent, PendingToolCall, TaskStatus as SessionTaskStatus, TrackedTask, TaskState, SubagentStats, PendingUserRequest, ResponseLatency, LatencyStats, PlanStep, PlanState, SessionStats, CompactionEvent, TruncationEvent, ContextAttribution, TurnAttribution, ContextSizePoint, PermissionMode, PermissionModeChange, } from './types/sessionEvent';
 export type { OpenCodeSession, OpenCodeMessage, OpenCodePart, OpenCodeTextPart, OpenCodeReasoningPart, OpenCodeToolInvocationPart, OpenCodeCompactionPart, OpenCodeDbToolPart, OpenCodeStepStartPart, OpenCodeStepFinishPart, OpenCodePatchPart, OpenCodeSubtaskPart, OpenCodeAgentPart, OpenCodeFilePart, OpenCodeRetryPart, OpenCodeSnapshotPart, OpenCodeProject, DbProject as OpenCodeDbProject, DbSession as OpenCodeDbSession, DbMessage as OpenCodeDbMessage, DbPart as OpenCodeDbPart, } from './types/opencode';
 export type { CodexRolloutLine, CodexSessionMeta, CodexResponseItem, CodexMessageItem, CodexContentPart, CodexReasoningItem, CodexFunctionCallItem, CodexFunctionCallOutputItem, CodexLocalShellCallItem, CodexWebSearchCallItem, CodexCustomToolCallItem, CodexCustomToolCallOutputItem, CodexCompacted, CodexTurnContext, CodexEventMsg, CodexEvent, CodexTurnStartedEvent, CodexTurnCompleteEvent, CodexTaskStartedEvent, CodexTaskCompleteEvent, CodexTurnAbortedEvent, CodexTokenCountEvent, CodexTokenUsage, CodexRateLimits, CodexAgentMessageEvent, CodexAgentReasoningEvent, CodexUserMessageEvent, CodexExecCommandBeginEvent, CodexExecCommandEndEvent, CodexMcpToolCallBeginEvent, CodexMcpToolCallEndEvent, CodexErrorEvent, CodexContextCompactedEvent, CodexPatchAppliedEvent, CodexBackgroundEvent, CodexGenericEvent, CodexDbThread, CodexReasoningSummary, } from './types/codex';
@@ -164,8 +164,10 @@ export type { StatuslineInput, StatuslineSelection } from './statusline/formatte
 export { getModelContextWindowSize, DEFAULT_CONTEXT_WINDOW } from './modelContext';
 export { parseModelId, getModelPricing, getModelInfo, calculateCost, calculateCostWithPricing, calculateCostWithProvenance, mergeCostSources, shortModelName, getModelDisplayInfo, compareModelIds, sortModelIds, formatCost, } from './modelInfo';
 export type { ModelPricing, CostTokenUsage, CostSource, CostProvenanceInput, CostWithProvenance, ModelInfo, ModelProvider, ParsedModelId, ModelDisplayInfo, } from './modelInfo';
-export { hydratePricingCatalog, normalizeLiteLlmCatalog, LITELLM_CATALOG_URL, } from './pricingCatalog';
+export { hydratePricingCatalog, normalizeLiteLlmCatalog, normalizeLiteLlmContextWindows, LITELLM_CATALOG_URL, } from './pricingCatalog';
 export type { HydrateOptions, HydrateResult } from './pricingCatalog';
+export { loadObservedContextWindows, recordObservedContextWindow, getObservedContextWindowPath, } from './observedContextWindows';
+export type { ObservedContextWindowOptions, ObservedContextWindowStore, } from './observedContextWindows';
 export { extractTokenUsage } from './extractors/tokenUsage';
 export { extractUrls, extractFilePaths, extractCommands } from './extractors/sessionAssets';
 export type { AssetAgent, ExtractedAsset, ExtractedAssetProvenance, ExtractedAssetType, ExtractedAssets, SourceAssets, } from './extractors/sessionAssets';

package/dist/index.js

@@ -8,8 +8,8 @@ exports.createSessionContextProjector = exports.calculateSessionContextPressure
 exports.scoreSessionQuality = exports.calculateQualityTrend = exports.PatternExtractor = exports.HeatmapTracker = exports.FrequencyTracker = exports.readErrorHistory = exports.getTopFailingTools = exports.appendErrorHistory = exports.MAX_ERROR_HISTORY_SESSIONS = exports.ERROR_HISTORY_SCHEMA_VERSION = exports.getSnapshotPath = exports.isSnapshotValid = exports.deleteSnapshot = exports.loadSnapshot = exports.saveSnapshot = exports.SessionMonitor = exports.parseTodoDependencies = exports.EventAggregator = exports.getRandomPhrase = exports.PHRASE_CATEGORIES = exports.ALL_PHRASES = exports.HIGHLIGHT_CSS = exports.clearHighlightCache = exports.highlightEvent = exports.formatSessionJson = exports.formatSessionMarkdown = exports.formatSessionText = exports.classifyNoise = exports.shouldMergeWithPrevious = exports.classifyFollowEvent = exports.classifyMessage = exports.getSoftNoiseReason = exports.isHardNoiseFollowEvent = exports.isHardNoise = exports.formatToolSummary = exports.formatTokenCount = exports.formatDurationMs = exports.createJsonlTail = exports.toFollowEvents = exports.createWatcher = exports.parseChangelog = exports.extractProposedPlanShared = exports.parsePlanMarkdownShared = exports.PlanExtractor = exports.segmentAssistantTurn = exports.reasoningSummary = exports.isAssistantTurnSubagentTool = exports.extractTurnSubagents = exports.assistantTurnEventsFromSessionEvents = exports.readSessionContextSnapshot = void 0;
 exports.getCodexProfileHome = exports.getCodexProfilesDir = exports.readClaudeProfileIdentity = exports.isClaudeProfileAuthenticated = exports.claudeKeychainService = exports.claudeKeychainSuffix = exports.getClaudeProfileHome = exports.getClaudeProfilesDir = exports.getActiveAccountStatus = exports.removeSavedAccountProfile = exports.replaceSavedAccountProfiles = exports.setActiveSavedAccount = exports.upsertSavedAccountProfile = exports.getActiveSavedAccount = exports.listSavedAccountProfiles = exports.writeSavedAccountRegistry = exports.readSavedAccountRegistry = exports.getAccountsDir = exports.resolveClaudeLoginCommand = exports.listAllAccounts = exports.switchAccount = exports.spawnAccountLogin = exports.finalizeAccountLogin = exports.getAccountLoginStatus = exports.beginAccountLogin = exports.isMultiAccountEnabled = exports.resolveActiveClaudeAccount = exports.getActiveAccount = exports.listAccounts = exports.removeAccount = exports.reconcileClaudeAuthState = exports.applyActiveClaudeToLiveHome = exports.resolveActiveClaudeHome = exports.switchToAccount = exports.addCurrentAccount = exports.readActiveClaudeAccount = exports.writeAccountRegistry = exports.readAccountRegistry = exports.ensureDefaultAccounts = exports.readClaudeMaxAccessTokenSync = exports.readClaudeMaxCredentials = exports.writeActiveCredentials = exports.readActiveCredentials = exports.openInBrowser = exports.parseTranscriptFromEvents = exports.parseTranscript = exports.generateHtmlReport = exports.formatCompactionLedger = exports.calculateCompactionLedger = exports.calculateCodeImpact = void 0;
 exports.selectStatuslineAccount = exports.formatStatusline = exports.estimateTimeToQuota = exports.BurnRateCalculator = exports.MultiProviderQuotaService = exports.resolveZaiQuota = exports.readZaiCredentials = exports.quotaStateFromZaiQuotaLimitPayload = exports.fetchZaiQuotaFromApi = exports.CodexQuotaWatcher = exports.resolveCodexQuotaFromLocalSources = exports.resolveCodexQuota = exports.readLatestCodexQuotaFromRollouts = exports.quotaFromCodexRateLimits = exports.fetchCodexQuotaFromApi = exports.fetchCodexResetCreditsFromApi = exports.getWorkspaceIdFromPath = exports.pruneQuotaHistory = exports.readQuotaHistoryDailyBuckets = exports.readQuotaHistoryRange = exports.appendQuotaHistorySample = exports.writeQuotaSnapshot = exports.readQuotaSnapshot = exports.QuotaPoller = exports.describeQuotaFailure = exports.withQuotaProjections = exports.projectQuotaWindow = exports.fetchQuota = exports.SEVEN_DAY_WINDOW_MS = exports.FIVE_HOUR_WINDOW_MS = exports.AutoSwitchController = exports.decideAutoSwitch = exports.DEFAULT_AUTO_SWITCH_CONFIG = exports.removeLauncher = exports.writeLauncher = exports.isShellHookInstalled = exports.uninstallShellHook = exports.installShellHook = exports.setTerminalActiveProfile = exports.removeCodexAccount = exports.switchToCodexAccount = exports.finalizeCodexAccount = exports.prepareCodexAccount = exports.getCodexExecutionEnv = exports.resolveSidekickCodexHome = exports.resolveActiveCodexAccount = exports.getActiveCodexAccount = exports.listCodexAccounts = exports.getSystemCodexHome = exports.getCodexMonitoringHomes = void 0;
-exports.zaiProviderQuotaStateSchema = exports.codexProviderQuotaStateSchema = exports.claudeProviderQuotaStateSchema = exports.providerQuotaStateSchema = exports.runtimeQuotaProviderSchema = exports.quotaFailureDescriptorSchema = exports.peakHoursStateSchema = exports.codexResetCreditsSnapshotSchema = exports.codexResetCreditSchema = exports.quotaSourceSchema = exports.quotaProviderIdSchema = exports.quotaFailureKindSchema = exports.quotaStateSchema = exports.quotaWindowSchema = exports.extractSessionEvents = exports.permissionModeSchema = exports.sessionEventSchema = exports.sessionMessageSchema = exports.messageUsageSchema = exports.ToolCallTracker = exports.extractToolCalls = exports.extractToolCall = exports.extractErrorMessage = exports.categorizeError = exports.gatherAssetsForCwd = exports.codexSessions = exports.readCodexAssets = exports.claudeSessions = exports.readClaudeAssets = exports.extractCommands = exports.extractFilePaths = exports.extractUrls = exports.extractTokenUsage = exports.LITELLM_CATALOG_URL = exports.normalizeLiteLlmCatalog = exports.hydratePricingCatalog = exports.formatCost = exports.sortModelIds = exports.compareModelIds = exports.getModelDisplayInfo = exports.shortModelName = exports.mergeCostSources = exports.calculateCostWithProvenance = exports.calculateCostWithPricing = exports.calculateCost = exports.getModelInfo = exports.getModelPricing = exports.parseModelId = exports.DEFAULT_CONTEXT_WINDOW = exports.getModelContextWindowSize = void 0;
-exports.scopePeakHoursToSessionProvider = exports.isClaudeCodeSessionProvider = exports.fetchPeakHoursStatus = exports.createPeakHoursNotApplicableState = exports.runDoctor = exports.getSessionDiagnostics = exports.formatHealthReport = exports.fetchOpenAIStatus = exports.fetchProviderStatus = exports.assistantTurnToolRefSchema = exports.assistantTurnToolGroupStepSchema = exports.assistantTurnTimelineItemSchema = exports.assistantTurnSubagentStatusSchema = exports.assistantTurnSubagentSchema = exports.assistantTurnReasoningTimelineItemSchema = exports.assistantTurnProjectionSchema = exports.assistantTurnProcessStepSchema = exports.assistantTurnProcessSchema = exports.assistantTurnNarrationStepSchema = exports.assistantTurnEventTypeSchema = exports.assistantTurnEventSchema = exports.listAllAccountsResultSchema = exports.savedAccountProfileSchema = exports.accountEntrySchema = exports.accountLoginStatusSchema = exports.beginAccountLoginResultSchema = exports.accountManagerResultSchema = exports.accountProviderIdSchema = exports.activeAccountStatusSchema = exports.activeProviderAccountStatusSchema = exports.quotaHistoryDailyBucketSchema = exports.quotaHistorySampleSchema = exports.quotaHistoryRuntimeProviderSchema = exports.providerQuotaMapSchema = void 0;
+exports.runtimeQuotaProviderSchema = exports.quotaFailureDescriptorSchema = exports.peakHoursStateSchema = exports.codexResetCreditsSnapshotSchema = exports.codexResetCreditSchema = exports.quotaSourceSchema = exports.quotaProviderIdSchema = exports.quotaFailureKindSchema = exports.quotaStateSchema = exports.quotaWindowSchema = exports.extractSessionEvents = exports.permissionModeSchema = exports.sessionEventSchema = exports.sessionMessageSchema = exports.messageUsageSchema = exports.ToolCallTracker = exports.extractToolCalls = exports.extractToolCall = exports.extractErrorMessage = exports.categorizeError = exports.gatherAssetsForCwd = exports.codexSessions = exports.readCodexAssets = exports.claudeSessions = exports.readClaudeAssets = exports.extractCommands = exports.extractFilePaths = exports.extractUrls = exports.extractTokenUsage = exports.getObservedContextWindowPath = exports.recordObservedContextWindow = exports.loadObservedContextWindows = exports.LITELLM_CATALOG_URL = exports.normalizeLiteLlmContextWindows = exports.normalizeLiteLlmCatalog = exports.hydratePricingCatalog = exports.formatCost = exports.sortModelIds = exports.compareModelIds = exports.getModelDisplayInfo = exports.shortModelName = exports.mergeCostSources = exports.calculateCostWithProvenance = exports.calculateCostWithPricing = exports.calculateCost = exports.getModelInfo = exports.getModelPricing = exports.parseModelId = exports.DEFAULT_CONTEXT_WINDOW = exports.getModelContextWindowSize = void 0;
+exports.scopePeakHoursToSessionProvider = exports.isClaudeCodeSessionProvider = exports.fetchPeakHoursStatus = exports.createPeakHoursNotApplicableState = exports.runDoctor = exports.getSessionDiagnostics = exports.formatHealthReport = exports.fetchOpenAIStatus = exports.fetchProviderStatus = exports.assistantTurnToolRefSchema = exports.assistantTurnToolGroupStepSchema = exports.assistantTurnTimelineItemSchema = exports.assistantTurnSubagentStatusSchema = exports.assistantTurnSubagentSchema = exports.assistantTurnReasoningTimelineItemSchema = exports.assistantTurnProjectionSchema = exports.assistantTurnProcessStepSchema = exports.assistantTurnProcessSchema = exports.assistantTurnNarrationStepSchema = exports.assistantTurnEventTypeSchema = exports.assistantTurnEventSchema = exports.listAllAccountsResultSchema = exports.savedAccountProfileSchema = exports.accountEntrySchema = exports.accountLoginStatusSchema = exports.beginAccountLoginResultSchema = exports.accountManagerResultSchema = exports.accountProviderIdSchema = exports.activeAccountStatusSchema = exports.activeProviderAccountStatusSchema = exports.quotaHistoryDailyBucketSchema = exports.quotaHistorySampleSchema = exports.quotaHistoryRuntimeProviderSchema = exports.providerQuotaMapSchema = exports.zaiProviderQuotaStateSchema = exports.codexProviderQuotaStateSchema = exports.claudeProviderQuotaStateSchema = exports.providerQuotaStateSchema = void 0;
 var usageNormalization_1 = require("./usageNormalization");
 Object.defineProperty(exports, "normalizeProviderUsage", { enumerable: true, get: function () { return usageNormalization_1.normalizeProviderUsage; } });
 Object.defineProperty(exports, "extractNormalizedUsage", { enumerable: true, get: function () { return usageNormalization_1.extractNormalizedUsage; } });
@@ -397,7 +397,13 @@ Object.defineProperty(exports, "formatCost", { enumerable: true, get: function (
 var pricingCatalog_1 = require("./pricingCatalog");
 Object.defineProperty(exports, "hydratePricingCatalog", { enumerable: true, get: function () { return pricingCatalog_1.hydratePricingCatalog; } });
 Object.defineProperty(exports, "normalizeLiteLlmCatalog", { enumerable: true, get: function () { return pricingCatalog_1.normalizeLiteLlmCatalog; } });
+Object.defineProperty(exports, "normalizeLiteLlmContextWindows", { enumerable: true, get: function () { return pricingCatalog_1.normalizeLiteLlmContextWindows; } });
 Object.defineProperty(exports, "LITELLM_CATALOG_URL", { enumerable: true, get: function () { return pricingCatalog_1.LITELLM_CATALOG_URL; } });
+// Observed context windows (provider-reported, persisted) — Node-only.
+var observedContextWindows_1 = require("./observedContextWindows");
+Object.defineProperty(exports, "loadObservedContextWindows", { enumerable: true, get: function () { return observedContextWindows_1.loadObservedContextWindows; } });
+Object.defineProperty(exports, "recordObservedContextWindow", { enumerable: true, get: function () { return observedContextWindows_1.recordObservedContextWindow; } });
+Object.defineProperty(exports, "getObservedContextWindowPath", { enumerable: true, get: function () { return observedContextWindows_1.getObservedContextWindowPath; } });
 // Extractors — per-event token usage and tool call extraction
 var tokenUsage_1 = require("./extractors/tokenUsage");
 Object.defineProperty(exports, "extractTokenUsage", { enumerable: true, get: function () { return tokenUsage_1.extractTokenUsage; } });

package/dist/modelContext.d.ts

@@ -1,19 +1,66 @@
 /**
  * Centralized model ID → context window size lookup.
  * Single source of truth used by all providers and the CLI dashboard.
+ *
+ * Resolution is layered, most-trusted first:
+ *   1. Runtime-reported — providers that self-report their window (Codex) apply
+ *      it in `getContextWindowLimit`, ahead of this module.
+ *   2. Observed overrides — the last window a provider actually reported for a
+ *      model, persisted across runs by `observedContextWindows.ts`. Reflects the
+ *      effective window for this account/tier, which can be well below the
+ *      model's published maximum.
+ *   3. Catalog overrides — `max_input_tokens` from the LiteLLM catalog, hydrated
+ *      by `pricingCatalog.ts`. Keeps new models correct without a code change.
+ *   4. The static table below — offline baseline of last resort.
+ *
+ * Within that order, match quality comes first: every layer is tried for an
+ * exact hit before any layer is tried for a prefix hit. See
+ * `getModelContextWindowSize`.
+ *
+ * Layers 2 and 3 are pushed in by Node-only modules through the `_set*` hooks,
+ * so this file stays browser-safe (no node:fs, no fetch) and can be bundled for
+ * webviews. Same pattern as `_setPricingOverrides` in `modelInfo.ts`.
  */
 /** Default context window size when model is unknown. */
 export declare const DEFAULT_CONTEXT_WINDOW = 200000;
+/**
+ * Internal: replace the observed-context-window map. Called by
+ * `observedContextWindows.ts` on load and on each newly observed value.
+ * Not part of the stable public API.
+ */
+export declare function _setObservedContextWindows(overrides: Record<string, number>): void;
+/** Internal: snapshot observed overrides (test + diagnostics). */
+export declare function _getObservedContextWindows(): Record<string, number>;
+/** Internal: clear observed overrides (used by tests). */
+export declare function _clearObservedContextWindows(): void;
+/**
+ * Internal: replace the catalog-context-window map. Called by
+ * `pricingCatalog.ts` after a successful LiteLLM hydration.
+ * Not part of the stable public API.
+ */
+export declare function _setCatalogContextWindows(overrides: Record<string, number>): void;
+/** Internal: snapshot catalog overrides (test + diagnostics). */
+export declare function _getCatalogContextWindows(): Record<string, number>;
+/** Internal: clear catalog overrides (used by tests and when disabling hydration). */
+export declare function _clearCatalogContextWindows(): void;
 /**
  * Returns the context window size for a model ID.
  *
  * Input is trimmed and lowercased before lookup, so padded or mixed-case
- * IDs (e.g. "Claude-Opus-4-8 ") resolve without caller-side normalization.
+ * IDs (e.g. "Claude-Opus-5 ") resolve without caller-side normalization.
  *
  * Lookup order:
  * 1. Explicit "[1m]" suffix (Claude Code's 1M-variant marker) → 1_000_000
- * 2. Exact match against MODEL_CONTEXT_SIZES
- * 3. Longest-prefix match (e.g. "claude-opus-4-6-20250414" → "claude-opus-4-6")
+ * 2. Exact match, most-trusted layer first: observed → catalog → static
+ * 3. Longest-prefix match, same layer order
+ *    (e.g. "claude-opus-4-6-20250414" → "claude-opus-4-6")
  * 4. DEFAULT_CONTEXT_WINDOW
+ *
+ * Match quality outranks layer trust: an exact hit in *any* layer beats a
+ * prefix guess in *every* layer. Resolving each layer fully before moving on
+ * would let a catalog key that merely happens to be a prefix shadow a curated
+ * static entry — e.g. the catalog has no "claude-sonnet-4-7", and prefix-first
+ * would hand it "claude-sonnet-4" (128K, a GitHub Copilot deployment) instead
+ * of the static table's exact 1M.
  */
 export declare function getModelContextWindowSize(modelId?: string): number;

package/dist/modelContext.js

@@ -2,17 +2,44 @@
 /**
  * Centralized model ID → context window size lookup.
  * Single source of truth used by all providers and the CLI dashboard.
+ *
+ * Resolution is layered, most-trusted first:
+ *   1. Runtime-reported — providers that self-report their window (Codex) apply
+ *      it in `getContextWindowLimit`, ahead of this module.
+ *   2. Observed overrides — the last window a provider actually reported for a
+ *      model, persisted across runs by `observedContextWindows.ts`. Reflects the
+ *      effective window for this account/tier, which can be well below the
+ *      model's published maximum.
+ *   3. Catalog overrides — `max_input_tokens` from the LiteLLM catalog, hydrated
+ *      by `pricingCatalog.ts`. Keeps new models correct without a code change.
+ *   4. The static table below — offline baseline of last resort.
+ *
+ * Within that order, match quality comes first: every layer is tried for an
+ * exact hit before any layer is tried for a prefix hit. See
+ * `getModelContextWindowSize`.
+ *
+ * Layers 2 and 3 are pushed in by Node-only modules through the `_set*` hooks,
+ * so this file stays browser-safe (no node:fs, no fetch) and can be bundled for
+ * webviews. Same pattern as `_setPricingOverrides` in `modelInfo.ts`.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.DEFAULT_CONTEXT_WINDOW = void 0;
+exports._setObservedContextWindows = _setObservedContextWindows;
+exports._getObservedContextWindows = _getObservedContextWindows;
+exports._clearObservedContextWindows = _clearObservedContextWindows;
+exports._setCatalogContextWindows = _setCatalogContextWindows;
+exports._getCatalogContextWindows = _getCatalogContextWindows;
+exports._clearCatalogContextWindows = _clearCatalogContextWindows;
 exports.getModelContextWindowSize = getModelContextWindowSize;
 /** Known model context window sizes (in tokens). */
 const MODEL_CONTEXT_SIZES = {
     // Claude — native 1M context (Fable 5, Opus 4.6+, Sonnet 4.6+)
     'claude-fable-5': 1_000_000,
+    'claude-opus-5': 1_000_000,
     'claude-opus-4-8': 1_000_000,
     'claude-opus-4-7': 1_000_000,
     'claude-opus-4-6': 1_000_000,
+    'claude-sonnet-5': 1_000_000,
     'claude-sonnet-4-7': 1_000_000,
     'claude-sonnet-4-6': 1_000_000,
     // Claude 4 family
@@ -30,7 +57,14 @@ const MODEL_CONTEXT_SIZES = {
     // OpenAI GPT-4.1 series (1M context)
     'gpt-4.1': 1_048_576,
     // OpenAI GPT-5 series (keys sorted longest-first below; explicit entries
-    // for every variant so prefix matching can't misclassify a new one)
+    // for every variant so prefix matching can't misclassify a new one).
+    // These are published model maxima. Codex reports a smaller *effective*
+    // window per account tier via token_count; that value wins through the
+    // observed-override layer.
+    'gpt-5.6-sol': 1_050_000,
+    'gpt-5.6-terra': 1_050_000,
+    'gpt-5.6': 1_050_000,
+    'gpt-5.5': 1_050_000,
     'gpt-5.4': 1_050_000,
     'gpt-5.3-codex-spark': 128_000,
     'gpt-5.3-codex': 400_000,
@@ -49,20 +83,100 @@ const MODEL_CONTEXT_SIZES = {
     deepseek: 128_000,
 };
 /** Prefix keys sorted longest-first for correct prefix matching. */
-const SORTED_KEYS = Object.keys(MODEL_CONTEXT_SIZES).sort((a, b) => b.length - a.length);
+const SORTED_KEYS = sortKeysLongestFirst(MODEL_CONTEXT_SIZES);
 /** Default context window size when model is unknown. */
 exports.DEFAULT_CONTEXT_WINDOW = 200_000;
+// ── Override Tables (populated at runtime by Node-only modules) ──
+let observedTable = {};
+let observedSortedKeys = [];
+let catalogTable = {};
+let catalogSortedKeys = [];
+function sortKeysLongestFirst(table) {
+    return Object.keys(table).sort((a, b) => b.length - a.length);
+}
+/**
+ * Lowercase keys so lookups match regardless of how the source spelled them.
+ * The LiteLLM catalog publishes some IDs mixed-case; the static table and the
+ * lookup path are both lowercase. Drops non-positive and non-finite values.
+ */
+function normalizeOverrides(overrides) {
+    const out = {};
+    for (const [key, value] of Object.entries(overrides)) {
+        if (!Number.isFinite(value) || value <= 0)
+            continue;
+        out[key.trim().toLowerCase()] = value;
+    }
+    return out;
+}
+/**
+ * Internal: replace the observed-context-window map. Called by
+ * `observedContextWindows.ts` on load and on each newly observed value.
+ * Not part of the stable public API.
+ */
+function _setObservedContextWindows(overrides) {
+    observedTable = normalizeOverrides(overrides);
+    observedSortedKeys = sortKeysLongestFirst(observedTable);
+}
+/** Internal: snapshot observed overrides (test + diagnostics). */
+function _getObservedContextWindows() {
+    return { ...observedTable };
+}
+/** Internal: clear observed overrides (used by tests). */
+function _clearObservedContextWindows() {
+    observedTable = {};
+    observedSortedKeys = [];
+}
+/**
+ * Internal: replace the catalog-context-window map. Called by
+ * `pricingCatalog.ts` after a successful LiteLLM hydration.
+ * Not part of the stable public API.
+ */
+function _setCatalogContextWindows(overrides) {
+    catalogTable = normalizeOverrides(overrides);
+    catalogSortedKeys = sortKeysLongestFirst(catalogTable);
+}
+/** Internal: snapshot catalog overrides (test + diagnostics). */
+function _getCatalogContextWindows() {
+    return { ...catalogTable };
+}
+/** Internal: clear catalog overrides (used by tests and when disabling hydration). */
+function _clearCatalogContextWindows() {
+    catalogTable = {};
+    catalogSortedKeys = [];
+}
+// ── Lookup ──
+/** Exact match against one table. */
+function exactWindow(table, modelId) {
+    const hit = table[modelId];
+    return hit === undefined ? null : hit;
+}
+/** Longest-prefix match against one table. */
+function prefixWindow(table, sortedKeys, modelId) {
+    for (const key of sortedKeys) {
+        if (modelId.startsWith(key))
+            return table[key];
+    }
+    return null;
+}
 /**
  * Returns the context window size for a model ID.
  *
  * Input is trimmed and lowercased before lookup, so padded or mixed-case
- * IDs (e.g. "Claude-Opus-4-8 ") resolve without caller-side normalization.
+ * IDs (e.g. "Claude-Opus-5 ") resolve without caller-side normalization.
  *
  * Lookup order:
  * 1. Explicit "[1m]" suffix (Claude Code's 1M-variant marker) → 1_000_000
- * 2. Exact match against MODEL_CONTEXT_SIZES
- * 3. Longest-prefix match (e.g. "claude-opus-4-6-20250414" → "claude-opus-4-6")
+ * 2. Exact match, most-trusted layer first: observed → catalog → static
+ * 3. Longest-prefix match, same layer order
+ *    (e.g. "claude-opus-4-6-20250414" → "claude-opus-4-6")
  * 4. DEFAULT_CONTEXT_WINDOW
+ *
+ * Match quality outranks layer trust: an exact hit in *any* layer beats a
+ * prefix guess in *every* layer. Resolving each layer fully before moving on
+ * would let a catalog key that merely happens to be a prefix shadow a curated
+ * static entry — e.g. the catalog has no "claude-sonnet-4-7", and prefix-first
+ * would hand it "claude-sonnet-4" (128K, a GitHub Copilot deployment) instead
+ * of the static table's exact 1M.
  */
 function getModelContextWindowSize(modelId) {
     if (!modelId)
@@ -72,21 +186,24 @@ function getModelContextWindowSize(modelId) {
     if (/\[1m\]/i.test(modelId))
         return 1_000_000;
     // Strip the suffix if present, so the normal lookup still succeeds when
-    // a caller passes e.g. "claude-opus-4-7[1m]" and we've already handled it.
+    // a caller passes e.g. "claude-opus-5[1m]" and we've already handled it.
     // Trim/lowercase so padded or mixed-case IDs match the lowercase table keys.
     const normalized = modelId
         .replace(/\[1m\]/gi, '')
         .trim()
         .toLowerCase();
-    // Exact match
-    if (MODEL_CONTEXT_SIZES[normalized] !== undefined) {
-        return MODEL_CONTEXT_SIZES[normalized];
-    }
-    // Longest-prefix match
-    for (const key of SORTED_KEYS) {
-        if (normalized.startsWith(key)) {
-            return MODEL_CONTEXT_SIZES[key];
-        }
-    }
+    // Pass 1 — exact matches, most-trusted layer first.
+    const exact = exactWindow(observedTable, normalized) ??
+        exactWindow(catalogTable, normalized) ??
+        exactWindow(MODEL_CONTEXT_SIZES, normalized);
+    if (exact !== null)
+        return exact;
+    // Pass 2 — longest-prefix matches, same layer order. Resolves versioned IDs
+    // that no layer spells out, e.g. "claude-opus-4-6-20250414".
+    const prefix = prefixWindow(observedTable, observedSortedKeys, normalized) ??
+        prefixWindow(catalogTable, catalogSortedKeys, normalized) ??
+        prefixWindow(MODEL_CONTEXT_SIZES, SORTED_KEYS, normalized);
+    if (prefix !== null)
+        return prefix;
     return exports.DEFAULT_CONTEXT_WINDOW;
 }

package/dist/modelInfo.js

@@ -76,6 +76,14 @@ const PRICING_TABLE = {
         cacheWriteCostPerMillion: 1.0,
         cacheReadCostPerMillion: 0.08,
     },
+    // Sonnet 5 carries the standard rate here. The introductory $2/$10 (through
+    // 2026-08-31) arrives via catalog hydration, which supersedes this baseline.
+    'claude-sonnet-5': {
+        inputCostPerMillion: 3.0,
+        outputCostPerMillion: 15.0,
+        cacheWriteCostPerMillion: 3.75,
+        cacheReadCostPerMillion: 0.3,
+    },
     'claude-sonnet-4-6': {
         inputCostPerMillion: 3.0,
         outputCostPerMillion: 15.0,
@@ -106,6 +114,12 @@ const PRICING_TABLE = {
         cacheWriteCostPerMillion: 3.75,
         cacheReadCostPerMillion: 0.3,
     },
+    'claude-opus-5': {
+        inputCostPerMillion: 5.0,
+        outputCostPerMillion: 25.0,
+        cacheWriteCostPerMillion: 6.25,
+        cacheReadCostPerMillion: 0.5,
+    },
     'claude-opus-4-8': {
         inputCostPerMillion: 5.0,
         outputCostPerMillion: 25.0,
@@ -186,14 +200,45 @@ const PRICING_TABLE = {
         cacheWriteCostPerMillion: 0,
         cacheReadCostPerMillion: 0,
     },
-    // ── OpenAI: GPT-5 family (baseline estimate; LiteLLM catalog overrides) ──
-    // Codex emits `gpt-5.4`, `gpt-5.3-codex`, `gpt-5`. Anchoring the static
-    // estimate to gpt-4o-tier rates so offline users see a reasonable ballpark.
+    // ── OpenAI: GPT-5 family ──
+    // Published rates, verified against the LiteLLM catalog. Runtime hydration
+    // still supersedes these; they are the offline baseline. Entries below
+    // `gpt-5.4` predate the published-rate pass and remain rough estimates.
+    'gpt-5.6-sol': {
+        inputCostPerMillion: 5.0,
+        outputCostPerMillion: 30.0,
+        cacheWriteCostPerMillion: 6.25,
+        cacheReadCostPerMillion: 0.5,
+    },
+    'gpt-5.6-terra': {
+        inputCostPerMillion: 2.5,
+        outputCostPerMillion: 15.0,
+        cacheWriteCostPerMillion: 3.13,
+        cacheReadCostPerMillion: 0.25,
+    },
+    'gpt-5.6': {
+        inputCostPerMillion: 5.0,
+        outputCostPerMillion: 30.0,
+        cacheWriteCostPerMillion: 6.25,
+        cacheReadCostPerMillion: 0.5,
+    },
+    'gpt-5.5': {
+        inputCostPerMillion: 5.0,
+        outputCostPerMillion: 30.0,
+        cacheWriteCostPerMillion: 0,
+        cacheReadCostPerMillion: 0.5,
+    },
+    'gpt-5.4-mini': {
+        inputCostPerMillion: 0.75,
+        outputCostPerMillion: 4.5,
+        cacheWriteCostPerMillion: 0,
+        cacheReadCostPerMillion: 0.07,
+    },
     'gpt-5.4': {
-        inputCostPerMillion: 1.25,
-        outputCostPerMillion: 10.0,
+        inputCostPerMillion: 2.5,
+        outputCostPerMillion: 15.0,
         cacheWriteCostPerMillion: 0,
-        cacheReadCostPerMillion: 0.125,
+        cacheReadCostPerMillion: 0.25,
     },
     'gpt-5.3-codex': {
         inputCostPerMillion: 1.25,

package/dist/node.d.ts

@@ -5,9 +5,11 @@
  * and `node:path`; safe for extension-host and CLI consumers. Not safe
  * for browser bundles — use `sidekick-shared/browser` there.
  */
-export { hydratePricingCatalog, normalizeLiteLlmCatalog, LITELLM_CATALOG_URL, } from './pricingCatalog';
+export { hydratePricingCatalog, normalizeLiteLlmCatalog, normalizeLiteLlmContextWindows, LITELLM_CATALOG_URL, } from './pricingCatalog';
 export type { HydrateOptions, HydrateResult } from './pricingCatalog';
+export { loadObservedContextWindows, recordObservedContextWindow, getObservedContextWindowPath, } from './observedContextWindows';
+export type { ObservedContextWindowOptions, ObservedContextWindowStore, } from './observedContextWindows';
 export { listRecentSessions, readSessionTranscript } from './sessionTranscripts';
 export type { ProviderSessionIndex, ListRecentSessionsOptions } from './sessionTranscripts';
 export { ObservedSessionCollector, observedSessionSourceFromProvider, fileFingerprint, } from './observedSessionCollector';
-export type { ObservedSessionReference, ObservedSessionCollectionSource, ObservedSessionDiagnosticKind, ObservedSessionDiagnostic, ObservedSessionCollection, ObservedSessionCollectorOptions, } from './observedSessionCollector';
+export type { ObservedSessionReference, ObservedSessionCollectionSource, ProviderObservedSessionCollectionSource, ObservedSessionDiagnosticKind, ObservedSessionDiagnostic, ObservedSessionCollection, ObservedSessionCollectorOptions, } from './observedSessionCollector';

package/dist/node.js

@@ -7,11 +7,16 @@
  * for browser bundles — use `sidekick-shared/browser` there.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.fileFingerprint = exports.observedSessionSourceFromProvider = exports.ObservedSessionCollector = exports.readSessionTranscript = exports.listRecentSessions = exports.LITELLM_CATALOG_URL = exports.normalizeLiteLlmCatalog = exports.hydratePricingCatalog = void 0;
+exports.fileFingerprint = exports.observedSessionSourceFromProvider = exports.ObservedSessionCollector = exports.readSessionTranscript = exports.listRecentSessions = exports.getObservedContextWindowPath = exports.recordObservedContextWindow = exports.loadObservedContextWindows = exports.LITELLM_CATALOG_URL = exports.normalizeLiteLlmContextWindows = exports.normalizeLiteLlmCatalog = exports.hydratePricingCatalog = void 0;
 var pricingCatalog_1 = require("./pricingCatalog");
 Object.defineProperty(exports, "hydratePricingCatalog", { enumerable: true, get: function () { return pricingCatalog_1.hydratePricingCatalog; } });
 Object.defineProperty(exports, "normalizeLiteLlmCatalog", { enumerable: true, get: function () { return pricingCatalog_1.normalizeLiteLlmCatalog; } });
+Object.defineProperty(exports, "normalizeLiteLlmContextWindows", { enumerable: true, get: function () { return pricingCatalog_1.normalizeLiteLlmContextWindows; } });
 Object.defineProperty(exports, "LITELLM_CATALOG_URL", { enumerable: true, get: function () { return pricingCatalog_1.LITELLM_CATALOG_URL; } });
+var observedContextWindows_1 = require("./observedContextWindows");
+Object.defineProperty(exports, "loadObservedContextWindows", { enumerable: true, get: function () { return observedContextWindows_1.loadObservedContextWindows; } });
+Object.defineProperty(exports, "recordObservedContextWindow", { enumerable: true, get: function () { return observedContextWindows_1.recordObservedContextWindow; } });
+Object.defineProperty(exports, "getObservedContextWindowPath", { enumerable: true, get: function () { return observedContextWindows_1.getObservedContextWindowPath; } });
 var sessionTranscripts_1 = require("./sessionTranscripts");
 Object.defineProperty(exports, "listRecentSessions", { enumerable: true, get: function () { return sessionTranscripts_1.listRecentSessions; } });
 Object.defineProperty(exports, "readSessionTranscript", { enumerable: true, get: function () { return sessionTranscripts_1.readSessionTranscript; } });

package/dist/observedContextWindows.d.ts

@@ -0,0 +1,54 @@
+/**
+ * Persisted record of context windows providers actually reported (Node-only).
+ *
+ * Some agents report the real context window for the running session — Codex
+ * emits `model_context_window` on every `token_count` event. That number is the
+ * *effective* window for the account and tier, which can sit well below the
+ * model's published maximum (e.g. 258,400 observed for `gpt-5.6-sol` against a
+ * catalog maximum of 1,050,000).
+ *
+ * A live session uses that value directly. This module keeps it across runs, so
+ * historical and static views don't fall back to a published maximum that would
+ * make the context gauge read far too empty.
+ *
+ * Precedence, most-trusted first: runtime → observed (here) → LiteLLM catalog →
+ * static table. See `modelContext.ts`.
+ *
+ * @module observedContextWindows
+ */
+export interface ObservedContextWindowOptions {
+    /** Directory for the store (defaults to the Sidekick config dir). */
+    cacheDir?: string;
+    /** Optional logger (useful for diagnostics without pulling a logger dep). */
+    logger?: (msg: string) => void;
+}
+export interface ObservedContextWindowEntry {
+    contextWindow: number;
+    /** ISO timestamp of the most recent observation. */
+    observedAt: string;
+}
+export interface ObservedContextWindowStore {
+    version: 1;
+    models: Record<string, ObservedContextWindowEntry>;
+}
+/** Resolve the on-disk store path. */
+export declare function getObservedContextWindowPath(cacheDir?: string): string;
+/**
+ * Load persisted observations and apply them as context-window overrides.
+ *
+ * Always resolves — never throws. A missing, unreadable, or malformed store
+ * leaves the overrides empty and lookups fall through to catalog/static.
+ */
+export declare function loadObservedContextWindows(options?: ObservedContextWindowOptions): Promise<Record<string, number>>;
+/**
+ * Record a provider-reported context window for a model.
+ *
+ * Applies in memory immediately, then persists — but only when the value
+ * actually changed. Codex reports a window on every `token_count` event, so an
+ * unconditional write would mean disk I/O many times per session for a value
+ * that almost never moves.
+ *
+ * Always resolves — never throws. Persistence failures are logged and dropped;
+ * the in-memory override still stands for this process.
+ */
+export declare function recordObservedContextWindow(modelId: string, contextWindow: number, options?: ObservedContextWindowOptions): Promise<void>;