@oh-my-pi/pi-catalog 18.14.1 → 18.14.2

package/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 ## [Unreleased]
 
+## [18.14.2] - 2026-10-31
+
+### Breaking Changes
+
+- Removed `thinkingLoopGuard` from compat records; `thinking-loop-guard` is now a catalog axis that applies on every host, with `"semantic"` for Gemini, DeepSeek, and Grok and `"off"` for Claude
+
+### Fixed
+
+- Fixed GitHub Copilot Claude 5.x base models (`claude-opus-5`, `claude-opus-5.5`, `claude-sonnet-5`, `claude-sonnet-5.5`) reporting the 1M window of their `-1m` variants once the model list was read from cache; discovered limits now survive cached reads instead of reverting to bundled values ([#17481](https://github.com/can1357/oh-my-pi/issues/17481))
+
 ## [18.14.1] - 2026-10-31
 
 ### Changed

package/dist/types/model-manager.d.ts

@@ -129,9 +129,19 @@ export declare function createModelManager<TApi extends Api = Api, TModelsDevPay
  * when at least one configured remote source did not refresh successfully.
  */
 export declare function resolveProviderModels<TApi extends Api = Api, TModelsDevPayload = unknown>(options: ModelManagerOptions<TApi, TModelsDevPayload>, strategy?: ModelRefreshStrategy): Promise<ModelResolutionResult<TApi>>;
+/**
+ * Merge policy that is part of a cache's identity in
+ * {@link fingerprintStaticModels}: `authoritative` when dynamic discovery owns
+ * the catalog, `additive` when the shared catalog may only add ids. The flags
+ * compose; snapshots written under different policies never match.
+ */
+export interface StaticFingerprintPolicy {
+    authoritative?: boolean;
+    additive?: boolean;
+}
 /**
  * Return the versioned content identity for a static provider slice. Callers
  * may mutate or freeze their arrays, so every call observes current content
  * and never attaches a stale sidecar fingerprint to caller-owned data.
  */
-export declare function fingerprintStaticModels<TApi extends Api>(models: readonly (ModelSpec<TApi> | Model<TApi>)[], dynamicModelsAuthoritative?: boolean): string;
+export declare function fingerprintStaticModels<TApi extends Api>(models: readonly (ModelSpec<TApi> | Model<TApi>)[], policy?: StaticFingerprintPolicy): string;

package/dist/types/types.d.ts

@@ -477,8 +477,6 @@ export interface OpenAICompat {
     clampOutputToModelMax?: boolean;
     /** Strip image inputs before encoding (text-only serving of a multimodal id). */
     stripImageInput?: boolean;
-    /** Thinking-loop watchdog guard family applied to streamed reasoning. */
-    thinkingLoopGuard?: "gemini" | "deepseek" | "xai";
 }
 /**
  * Compatibility settings for anthropic-messages API.
@@ -676,8 +674,6 @@ export interface AnthropicCompat {
      * contract override it. Default: 10485760.
      */
     maxImagePayloadBytes?: number;
-    /** Thinking-loop watchdog guard family applied to streamed reasoning. */
-    thinkingLoopGuard?: "gemini" | "deepseek" | "xai";
     /**
      * Drop the enabled `thinking` config and replayed thinking blocks when the
      * conversation is not thinking-led (a proxy contract for non-interleaved
@@ -887,8 +883,6 @@ export interface ResolvedOpenAISharedCompat {
     toolSchemaFlavor?: OpenAICompat["toolSchemaFlavor"];
     /** Strip image inputs before encoding (text-only serving of a multimodal id). */
     stripImageInput: boolean;
-    /** Thinking-loop watchdog guard family applied to streamed reasoning. */
-    thinkingLoopGuard?: OpenAICompat["thinkingLoopGuard"];
     /** Flatten/reject leftover root `anyOf`/`oneOf` unions in strict tool schemas (xAI's function-calling validator 400s on them). */
     rejectRootObjectUnion: boolean;
     /** Retry without strict tools when the host rejects a strict grammar as too large (OpenRouter-Anthropic compiled-grammar overflow). */
@@ -900,7 +894,7 @@ export interface ResolvedOpenAISharedCompat {
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "trustExplicitThinkingOnly" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "statefulResponses" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "stripThinkingHistory" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "sessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "trustExplicitThinkingOnly" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "statefulResponses" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "stripThinkingHistory" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "sessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];
@@ -992,7 +986,7 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
  */
 export type ResolvedOpenRouterCompat = ResolvedOpenAICompat & ResolvedOpenAIResponsesCompat;
 /** Fully-resolved anthropic-messages compat view (same contract as `ResolvedOpenAICompat`). */
-export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "thinkingLoopGuard" | "bedrockMessagesApi" | "effortBeta" | "extendedCacheTtlBeta" | "disabledThinking" | "betweenToolsEffort" | "stripThinkingHistory" | "fastMode" | "maxImageDimension" | "maxImagePayloadBytes">> & {
+export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "bedrockMessagesApi" | "effortBeta" | "extendedCacheTtlBeta" | "disabledThinking" | "betweenToolsEffort" | "stripThinkingHistory" | "fastMode" | "maxImageDimension" | "maxImagePayloadBytes">> & {
     /** Effort-beta override; undefined keeps the transport's legacy heuristic. */
     effortBeta?: AnthropicCompat["effortBeta"];
     /** Extended-cache-ttl beta opt-out; undefined sends it where a 1h breakpoint needs it. */
@@ -1016,8 +1010,6 @@ export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdle
      * request builder.
      */
     maxImagePayloadBytes?: number;
-    /** Thinking-loop watchdog guard family applied to streamed reasoning. */
-    thinkingLoopGuard?: AnthropicCompat["thinkingLoopGuard"];
     /**
      * Bedrock's Anthropic Messages API (see {@link AnthropicCompat.bedrockMessagesApi}).
      * `true` when detected or configured; otherwise unset, so rows baked before
@@ -1116,16 +1108,13 @@ export interface GoogleCompat {
      * {@link AnthropicCompat.maxImagePayloadBytes}).
      */
     maxImagePayloadBytes?: number;
-    /** Thinking-loop watchdog guard family applied to streamed reasoning. */
-    thinkingLoopGuard?: "gemini" | "deepseek" | "xai";
     /** See {@link BedrockCompat.supportsSamplingParams}. */
     supportsSamplingParams?: boolean;
 }
 /** Fully-resolved google-API compat view, materialized once by `buildModel`. */
-export type ResolvedGoogleCompat = Required<Omit<GoogleCompat, "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "thinkingLoopGuard" | "antigravityUsageLabel" | "supportsSamplingParams" | "maxImageDimension" | "maxImagePayloadBytes">> & {
+export type ResolvedGoogleCompat = Required<Omit<GoogleCompat, "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "antigravityUsageLabel" | "supportsSamplingParams" | "maxImageDimension" | "maxImagePayloadBytes">> & {
     streamFirstEventTimeoutMs?: number;
     streamIdleTimeoutMs?: number;
-    thinkingLoopGuard?: GoogleCompat["thinkingLoopGuard"];
     antigravityUsageLabel?: string;
     supportsSamplingParams?: boolean;
     maxImageDimension?: number;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.14.1",
+	"version": "18.14.2",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -88,12 +88,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.14.1",
-		"@oh-my-pi/pi-utils": "18.14.1"
+		"@oh-my-pi/omptype": "18.14.2",
+		"@oh-my-pi/pi-utils": "18.14.2"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.14.1",
+		"@oh-my-pi/pi-ai": "18.14.2",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/axes.ts

@@ -317,11 +317,6 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"strip-image-input": wire("stripImageInput", [...OAI, "anthropic", "google"]),
 	"supports-forced-tool-choice": wire("supportsForcedToolChoice", [...OAI, "anthropic", "bedrock"]),
 	"supports-sampling-params": wire("supportsSamplingParams", [...OAI, "anthropic", "bedrock", "devin", "google"]),
-	"thinking-loop-guard": wire("thinkingLoopGuard", [...OAI, "anthropic", "google"], "scalar", [
-		"gemini",
-		"deepseek",
-		"xai",
-	]),
 
 	// ── thinking control surface ──
 	"thinking-default-level": { key: "defaultLevel", set: "thinking", shape: "scalar", values: EFFORTS },
@@ -456,6 +451,19 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 		shape: "scalar",
 		values: ["output", "signature"],
 	},
+	/**
+	 * Stream loop guard for the model line, on every host and API (read
+	 * through `loopGuardLevel` in pi-ai `utils/thinking-loop`). `semantic`
+	 * adds near-duplicate and stalled-wording paragraph detection to the
+	 * exact repeated-cycle check; `off` disables the guard. Unassigned: exact
+	 * repeated cycles only.
+	 */
+	"thinking-loop-guard": {
+		key: "thinkingLoopGuard",
+		set: "catalog",
+		shape: "scalar",
+		values: ["off", "semantic"],
+	},
 	"delegation-bias": { key: "delegationBias", set: "catalog", shape: "scalar", values: DELEGATION_BIASES },
 	"discovery-api": { key: "discoveryApi", set: "catalog", shape: "scalar" },
 	"edit-prompt-variant": { key: "editPromptVariant", set: "catalog", shape: "scalar", values: ["full", "compact"] },

package/src/compat/resolve.ts

@@ -604,7 +604,6 @@ function detectOpenAICompat(
 		zaiReasoningEffortDialect: false,
 		clampOutputToModelMax: d.isLocalOpenAICompatBackend,
 		stripImageInput: false,
-		thinkingLoopGuard: undefined,
 		rejectRootObjectUnion: false,
 		retryWithoutStrictOnGrammarError: false,
 		supportsPromptCacheKey: false,
@@ -762,7 +761,6 @@ function resolveOpenAIResponsesPolicy(
 		supportsSteering: false,
 		requiresReasoningOffJuiceInstruction: false,
 		stripImageInput: false,
-		thinkingLoopGuard: undefined,
 		reasoningEffortMap: {},
 		supportsReasoningParams: true,
 		supportsSamplingParams: true,
@@ -936,7 +934,6 @@ function resolveAnthropicPolicy(
 		// user-authored `compat` block to override them.
 		maxImageDimension: undefined,
 		maxImagePayloadBytes: undefined,
-		thinkingLoopGuard: undefined,
 		// Present as keys so models.yml `compat` can set them; unset unless a rule assigns them.
 		stripThinkingHistory: undefined,
 		fastMode: undefined,

package/src/compat/rules/classes/anthropic.kdl

@@ -29,6 +29,11 @@ class "anthropic" {
 	// back as text (`reasoning_extraction`, issue #6093).
 	classifier-cot-extraction #true
 	thinking-replay-tokens "output"
+	// Summarized thinking repeats one progress line verbatim while long hidden
+	// reasoning runs ("Still writing the parser..." six-plus times back to
+	// back), which reads as an exact cycle. Claude has no runaway-loop failure
+	// that warrants the guard, so it never discards a turn here.
+	thinking-loop-guard "off"
 	// verified: billing matched this formula on Opus 4.7+ and Fable 5.1.
 	image-tokenization {
 		regime "anthropic-patch"

package/src/compat/rules/classes/deepseek.kdl

@@ -102,7 +102,7 @@ class "deepseek" {
 	// leaks it into content — hosted gateways, local backends, and custom
 	// providers alike. On hosts that parse it the healer sees no markup.
 	stream-markup-healing-pattern "dsml"
-	thinking-loop-guard "deepseek"
+	thinking-loop-guard "semantic"
 	// V4 reasoning reaches omp as its plaintext trace on every host, and the
 	// revision pins the weights (flash 4.0 / 4.1, pro 4.0). Earlier lineages
 	// classify too coarsely to carry: `r1` also matches the R1 distills.

package/src/compat/rules/classes/gemini.kdl

@@ -149,7 +149,7 @@ class "gemini" {
 		}
 	}
 
-	thinking-loop-guard "gemini"
+	thinking-loop-guard "semantic"
 }
 
 // Google Search grounding is a Gemini 2+ lineage capability on every Gemini transport.

package/src/compat/rules/classes/xai.kdl

@@ -67,7 +67,7 @@ class "xai" {
 			}
 		}
 	}
-	thinking-loop-guard "xai"
+	thinking-loop-guard "semantic"
 }
 
 // Same-provider swap targets for hosted search and image generation.