@code-yeongyu/senpi 2026.8.29 → 2026.8.30-3

package/dist/core/extensions/builtin/compaction/speculative-summary.js

@@ -0,0 +1,131 @@
+import { sanitizeAnthropicToolPairs as sanitizeAnthropicPayload, } from "@earendil-works/pi-ai";
+import { stream } from "@earendil-works/pi-ai/compat";
+import { consumeStreamWithIdleTimeout, DEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS, DEFAULT_SUMMARIZATION_MAX_DURATION_MS, } from "../../../compaction/stream-watchdog.js";
+import { convertToLlm } from "../../../messages.js";
+import { repairOrphanedToolResults } from "./repair-tool-pairs.js";
+import { normalizeSummarizationTurnOrder } from "./summarization-turn-order.js";
+const SUMMARY_TOKEN_HEADROOM = 32_768;
+const SUMMARY_CONTEXT_WINDOW_RESERVE_RATIO = 0.5;
+function summaryMaxTokens(model, contextWindow) {
+    const headroom = model.maxTokens > 0 ? Math.min(SUMMARY_TOKEN_HEADROOM, model.maxTokens) : SUMMARY_TOKEN_HEADROOM;
+    if (contextWindow > 0) {
+        return Math.min(headroom, Math.floor(contextWindow * SUMMARY_CONTEXT_WINDOW_RESERVE_RATIO));
+    }
+    return headroom;
+}
+/**
+ * Reasoning override for summarization requests. Compaction must be fast: a
+ * summarization request that inherits the provider's default reasoning mode
+ * burns its latency (and output budget) on invisible thinking before emitting
+ * the summary. Disable or minimize reasoning per wire family; adapters ignore
+ * options their provider does not support. Mirrors how OpenAI Codex keeps its
+ * compaction turn cheap.
+ */
+function summarizationReasoningOptions(model) {
+    if (!model.reasoning)
+        return {};
+    if (model.api === "anthropic-messages")
+        return { thinkingEnabled: false };
+    const reasoningEffort = ["low", "medium", "high"].find((level) => model.thinkingLevelMap?.[level] !== null);
+    if (!reasoningEffort)
+        return {};
+    switch (model.api) {
+        case "openai-responses":
+        case "openai-codex-responses":
+        case "azure-openai-responses":
+            return { reasoningEffort, reasoningSummary: null };
+        case "openai-completions":
+            return { reasoningEffort };
+        default:
+            return {};
+    }
+}
+export function getSummaryText(message) {
+    const content = Array.isArray(message.content)
+        ? message.content
+        : [{ type: "text", text: message.content }];
+    return content
+        .filter((content) => content.type === "text")
+        .map((content) => content.text)
+        .join("\n")
+        .trim();
+}
+export function isAssistantMessage(message) {
+    return message.role === "assistant" && "stopReason" in message;
+}
+/**
+ * Providers registered through `pi.registerProvider()` (claude-sdk-oauth, Kiro, any
+ * extension provider) exist only in Senpi's ModelRuntime, never in compat's builtin
+ * api-registry, which rejects their api id outright. Dispatch through the runtime
+ * whenever it is reachable and keep compat for contexts constructed without a registry.
+ */
+function summarizationStream(context, model, requestContext, options) {
+    const runtime = context.modelRegistry?.modelRuntime;
+    return runtime ? runtime.stream(model, requestContext, options) : stream(model, requestContext, options);
+}
+export async function generateSummaryMessage(options) {
+    // Send the conversation as native LLM messages with the summarization
+    // instruction as a trailing user message, mirroring normal agent traffic.
+    // A single serialized `<conversation>` text dump of a large session is
+    // deterministically refused by Anthropic's anti-distillation classifier
+    // ("reverse engineering or duplicating model outputs"), while the same
+    // content as native blocks with the agent's system prompt and tools passes.
+    // Request-local controller: the idle watchdog must be able to tear down a
+    // stalled summarization request without aborting the caller's own signal.
+    const requestController = new AbortController();
+    const onCallerAbort = () => requestController.abort(options.signal?.reason);
+    if (options.signal) {
+        if (options.signal.aborted)
+            onCallerAbort();
+        else
+            options.signal.addEventListener("abort", onCallerAbort, { once: true });
+    }
+    try {
+        const requestMessages = [
+            ...options.messages,
+            {
+                role: "user",
+                content: [{ type: "text", text: options.prompt.user }],
+                timestamp: Date.now(),
+            },
+        ];
+        const providerRequest = await options.context.prepareProviderRequest?.(requestMessages);
+        const requestContext = {
+            systemPrompt: options.snapshot.systemPrompt ?? options.prompt.system,
+            messages: repairOrphanedToolResults(normalizeSummarizationTurnOrder(convertToLlm(providerRequest?.messages ?? requestMessages))),
+            ...(options.snapshot.tools && options.snapshot.tools.length > 0 ? { tools: options.snapshot.tools } : {}),
+        };
+        const headers = providerRequest
+            ? await providerRequest.transformHeaders(options.auth.headers ?? {})
+            : options.auth.headers;
+        const responseStream = summarizationStream(options.context, options.snapshot.model, requestContext, {
+            apiKey: options.auth.apiKey,
+            headers,
+            extraBody: options.auth.extraBody,
+            onPayload: async (payload, model) => {
+                const sanitized = model.api === "anthropic-messages" ? sanitizeAnthropicPayload(payload) : payload;
+                return providerRequest ? await providerRequest.transformPayload(sanitized) : sanitized;
+            },
+            maxTokens: summaryMaxTokens(options.snapshot.model, options.snapshot.contextWindow),
+            signal: requestController.signal,
+            ...summarizationReasoningOptions(options.snapshot.model),
+        });
+        await consumeStreamWithIdleTimeout(responseStream, {
+            idleTimeoutMs: DEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS,
+            maxDurationMs: DEFAULT_SUMMARIZATION_MAX_DURATION_MS,
+            abort: () => requestController.abort(),
+            signal: options.signal,
+            onEvent: (event) => {
+                if (event.type === "text_delta" && event.delta) {
+                    options.onProgress?.(event.delta);
+                }
+            },
+        });
+        return await responseStream.result();
+    }
+    finally {
+        if (options.signal)
+            options.signal.removeEventListener("abort", onCallerAbort);
+    }
+}
+//# sourceMappingURL=speculative-summary.js.map

package/dist/core/extensions/builtin/compaction/speculative-summary.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"speculative-summary.js","sourceRoot":"","sources":["../../../../../src/core/extensions/builtin/compaction/speculative-summary.ts"],"names":[],"mappings":"AACA,OAAO,EAQN,0BAA0B,IAAI,wBAAwB,GAEtD,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,MAAM,EAAE,MAAM,8BAA8B,CAAC;AACtD,OAAO,EACN,4BAA4B,EAC5B,qCAAqC,EACrC,qCAAqC,GACrC,MAAM,wCAAwC,CAAC;AAChD,OAAO,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAEpD,OAAO,EAAE,yBAAyB,EAAE,MAAM,wBAAwB,CAAC;AAEnE,OAAO,EAAE,+BAA+B,EAAE,MAAM,+BAA+B,CAAC;AAEhF,MAAM,sBAAsB,GAAG,MAAM,CAAC;AACtC,MAAM,oCAAoC,GAAG,GAAG,CAAC;AAGjD,SAAS,gBAAgB,CAAC,KAAiB,EAAE,aAAqB;IACjE,MAAM,QAAQ,GAAG,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,sBAAsB,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC;IAClH,IAAI,aAAa,GAAG,CAAC,EAAE,CAAC;QACvB,OAAO,IAAI,CAAC,GAAG,CAAC,QAAQ,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,oCAAoC,CAAC,CAAC,CAAC;IAC7F,CAAC;IACD,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,6BAA6B,CAAC,KAAiB;IACvD,IAAI,CAAC,KAAK,CAAC,SAAS;QAAE,OAAO,EAAE,CAAC;IAChC,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB;QAAE,OAAO,EAAE,eAAe,EAAE,KAAK,EAAE,CAAC;IAC1E,MAAM,eAAe,GAAI,CAAC,KAAK,EAAE,QAAQ,EAAE,MAAM,CAAW,CAAC,IAAI,CAChE,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC,KAAK,CAAC,KAAK,IAAI,CACnD,CAAC;IACF,IAAI,CAAC,eAAe;QAAE,OAAO,EAAE,CAAC;IAChC,QAAQ,KAAK,CAAC,GAAG,EAAE,CAAC;QACnB,KAAK,kBAAkB,CAAC;QACxB,KAAK,wBAAwB,CAAC;QAC9B,KAAK,wBAAwB;YAC5B,OAAO,EAAE,eAAe,EAAE,gBAAgB,EAAE,IAAI,EAAE,CAAC;QACpD,KAAK,oBAAoB;YACxB,OAAO,EAAE,eAAe,EAAE,CAAC;QAC5B;YACC,OAAO,EAAE,CAAC;IACZ,CAAC;AACF,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,OAAgB;IAC9C,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC;QAC7C,CAAC,CAAC,OAAO,CAAC,OAAO;QACjB,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;IACtD,OAAO,OAAO;SACZ,MAAM,CAAC,CAAC,OAAO,EAA0B,EAAE,CAAC,OAAO,CAAC,IAAI,KAAK,MAAM,CAAC;SACpE,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC;SAC9B,IAAI,CAAC,IAAI,CAAC;SACV,IAAI,EAAE,CAAC;AACV,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,OAAgB;IAClD,OAAO,OAAO,CAAC,IAAI,KAAK,WAAW,IAAI,YAAY,IAAI,OAAO,CAAC;AAChE,CAAC;AAED;;;;;GAKG;AACH,SAAS,mBAAmB,CAC3B,OAAqC,EACrC,KAAiB,EACjB,cAAuB,EACvB,OAAgD;IAEhD,MAAM,OAAO,GAAG,OAAO,CAAC,aAAa,EAAE,YAAY,CAAC;IACpD,OAAO,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,EAAE,cAAc,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,EAAE,cAAc,EAAE,OAAO,CAAC,CAAC;AAC1G,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAAC,OAY5C;IACA,sEAAsE;IACtE,0EAA0E;IAC1E,uEAAuE;IACvE,wEAAwE;IACxE,uEAAuE;IACvE,4EAA4E;IAC5E,0EAA0E;IAC1E,0EAA0E;IAC1E,MAAM,iBAAiB,GAAG,IAAI,eAAe,EAAE,CAAC;IAChD,MAAM,aAAa,GAAG,GAAG,EAAE,CAAC,iBAAiB,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5E,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;QACpB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;YAAE,aAAa,EAAE,CAAC;;YACvC,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,aAAa,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAC9E,CAAC;IACD,IAAI,CAAC;QACJ,MAAM,eAAe,GAAmB;YACvC,GAAG,OAAO,CAAC,QAAQ;YACnB;gBACC,IAAI,EAAE,MAAM;gBACZ,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;gBACtD,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;aACrB;SACD,CAAC;QACF,MAAM,eAAe,GAAG,MAAM,OAAO,CAAC,OAAO,CAAC,sBAAsB,EAAE,CAAC,eAAe,CAAC,CAAC;QACxF,MAAM,cAAc,GAAG;YACtB,YAAY,EAAE,OAAO,CAAC,QAAQ,CAAC,YAAY,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM;YACpE,QAAQ,EAAE,yBAAyB,CAClC,+BAA+B,CAAC,YAAY,CAAC,eAAe,EAAE,QAAQ,IAAI,eAAe,CAAC,CAAC,CAC3F;YACD,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACzG,CAAC;QACF,MAAM,OAAO,GAAG,eAAe;YAC9B,CAAC,CAAC,MAAM,eAAe,CAAC,gBAAgB,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC;YACpE,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC;QACxB,MAAM,cAAc,GAAG,mBAAmB,CAAC,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,QAAQ,CAAC,KAAK,EAAE,cAAc,EAAE;YACnG,MAAM,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM;YAC3B,OAAO;YACP,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,SAAS;YACjC,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE;gBACnC,MAAM,SAAS,GAAG,KAAK,CAAC,GAAG,KAAK,oBAAoB,CAAC,CAAC,CAAC,wBAAwB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;gBACnG,OAAO,eAAe,CAAC,CAAC,CAAC,MAAM,eAAe,CAAC,gBAAgB,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;YACxF,CAAC;YACD,SAAS,EAAE,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;YACnF,MAAM,EAAE,iBAAiB,CAAC,MAAM;YAChC,GAAG,6BAA6B,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC;SACxD,CAAC,CAAC;QACH,MAAM,4BAA4B,CAAC,cAAc,EAAE;YAClD,aAAa,EAAE,qCAAqC;YACpD,aAAa,EAAE,qCAAqC;YACpD,KAAK,EAAE,GAAG,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE;YACtC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE;gBAClB,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;oBAChD,OAAO,CAAC,UAAU,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;gBACnC,CAAC;YACF,CAAC;SACD,CAAC,CAAC;QACH,OAAO,MAAM,cAAc,CAAC,MAAM,EAAE,CAAC;IACtC,CAAC;YAAS,CAAC;QACV,IAAI,OAAO,CAAC,MAAM;YAAE,OAAO,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,aAAa,CAAC,CAAC;IAChF,CAAC;AACF,CAAC","sourcesContent":["import type { AgentMessage } from \"@earendil-works/pi-agent-core\";\nimport {\n\ttype AssistantMessage,\n\ttype AssistantMessageEventStream,\n\ttype Context,\n\ttype Message,\n\ttype Model,\n\ttype ProviderHeaders,\n\ttype StreamOptions,\n\tsanitizeAnthropicToolPairs as sanitizeAnthropicPayload,\n\ttype TextContent,\n} from \"@earendil-works/pi-ai\";\nimport { stream } from \"@earendil-works/pi-ai/compat\";\nimport {\n\tconsumeStreamWithIdleTimeout,\n\tDEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS,\n\tDEFAULT_SUMMARIZATION_MAX_DURATION_MS,\n} from \"../../../compaction/stream-watchdog.ts\";\nimport { convertToLlm } from \"../../../messages.ts\";\nimport type { buildPrompt } from \"./prompts.ts\";\nimport { repairOrphanedToolResults } from \"./repair-tool-pairs.ts\";\nimport type { SpeculativeCompactionContext, SpeculativeCompactionSnapshot } from \"./speculative.ts\";\nimport { normalizeSummarizationTurnOrder } from \"./summarization-turn-order.ts\";\n\nconst SUMMARY_TOKEN_HEADROOM = 32_768;\nconst SUMMARY_CONTEXT_WINDOW_RESERVE_RATIO = 0.5;\ntype CompactionProgressCallback = (delta: string) => void;\n\nfunction summaryMaxTokens(model: Model<any>, contextWindow: number): number {\n\tconst headroom = model.maxTokens > 0 ? Math.min(SUMMARY_TOKEN_HEADROOM, model.maxTokens) : SUMMARY_TOKEN_HEADROOM;\n\tif (contextWindow > 0) {\n\t\treturn Math.min(headroom, Math.floor(contextWindow * SUMMARY_CONTEXT_WINDOW_RESERVE_RATIO));\n\t}\n\treturn headroom;\n}\n\n/**\n * Reasoning override for summarization requests. Compaction must be fast: a\n * summarization request that inherits the provider's default reasoning mode\n * burns its latency (and output budget) on invisible thinking before emitting\n * the summary. Disable or minimize reasoning per wire family; adapters ignore\n * options their provider does not support. Mirrors how OpenAI Codex keeps its\n * compaction turn cheap.\n */\nfunction summarizationReasoningOptions(model: Model<any>): Record<string, unknown> {\n\tif (!model.reasoning) return {};\n\tif (model.api === \"anthropic-messages\") return { thinkingEnabled: false };\n\tconst reasoningEffort = ([\"low\", \"medium\", \"high\"] as const).find(\n\t\t(level) => model.thinkingLevelMap?.[level] !== null,\n\t);\n\tif (!reasoningEffort) return {};\n\tswitch (model.api) {\n\t\tcase \"openai-responses\":\n\t\tcase \"openai-codex-responses\":\n\t\tcase \"azure-openai-responses\":\n\t\t\treturn { reasoningEffort, reasoningSummary: null };\n\t\tcase \"openai-completions\":\n\t\t\treturn { reasoningEffort };\n\t\tdefault:\n\t\t\treturn {};\n\t}\n}\n\nexport function getSummaryText(message: Message): string {\n\tconst content = Array.isArray(message.content)\n\t\t? message.content\n\t\t: [{ type: \"text\" as const, text: message.content }];\n\treturn content\n\t\t.filter((content): content is TextContent => content.type === \"text\")\n\t\t.map((content) => content.text)\n\t\t.join(\"\\n\")\n\t\t.trim();\n}\n\nexport function isAssistantMessage(message: Message): message is AssistantMessage {\n\treturn message.role === \"assistant\" && \"stopReason\" in message;\n}\n\n/**\n * Providers registered through `pi.registerProvider()` (claude-sdk-oauth, Kiro, any\n * extension provider) exist only in Senpi's ModelRuntime, never in compat's builtin\n * api-registry, which rejects their api id outright. Dispatch through the runtime\n * whenever it is reachable and keep compat for contexts constructed without a registry.\n */\nfunction summarizationStream(\n\tcontext: SpeculativeCompactionContext,\n\tmodel: Model<any>,\n\trequestContext: Context,\n\toptions: StreamOptions & Record<string, unknown>,\n): AssistantMessageEventStream {\n\tconst runtime = context.modelRegistry?.modelRuntime;\n\treturn runtime ? runtime.stream(model, requestContext, options) : stream(model, requestContext, options);\n}\n\nexport async function generateSummaryMessage(options: {\n\tcontext: SpeculativeCompactionContext;\n\tmessages: AgentMessage[];\n\tonProgress?: CompactionProgressCallback;\n\tprompt: ReturnType<typeof buildPrompt>;\n\tsignal?: AbortSignal;\n\tsnapshot: SpeculativeCompactionSnapshot;\n\tauth: {\n\t\tapiKey?: string;\n\t\theaders?: ProviderHeaders;\n\t\textraBody?: Record<string, unknown>;\n\t};\n}): Promise<Message | undefined> {\n\t// Send the conversation as native LLM messages with the summarization\n\t// instruction as a trailing user message, mirroring normal agent traffic.\n\t// A single serialized `<conversation>` text dump of a large session is\n\t// deterministically refused by Anthropic's anti-distillation classifier\n\t// (\"reverse engineering or duplicating model outputs\"), while the same\n\t// content as native blocks with the agent's system prompt and tools passes.\n\t// Request-local controller: the idle watchdog must be able to tear down a\n\t// stalled summarization request without aborting the caller's own signal.\n\tconst requestController = new AbortController();\n\tconst onCallerAbort = () => requestController.abort(options.signal?.reason);\n\tif (options.signal) {\n\t\tif (options.signal.aborted) onCallerAbort();\n\t\telse options.signal.addEventListener(\"abort\", onCallerAbort, { once: true });\n\t}\n\ttry {\n\t\tconst requestMessages: AgentMessage[] = [\n\t\t\t...options.messages,\n\t\t\t{\n\t\t\t\trole: \"user\",\n\t\t\t\tcontent: [{ type: \"text\", text: options.prompt.user }],\n\t\t\t\ttimestamp: Date.now(),\n\t\t\t},\n\t\t];\n\t\tconst providerRequest = await options.context.prepareProviderRequest?.(requestMessages);\n\t\tconst requestContext = {\n\t\t\tsystemPrompt: options.snapshot.systemPrompt ?? options.prompt.system,\n\t\t\tmessages: repairOrphanedToolResults(\n\t\t\t\tnormalizeSummarizationTurnOrder(convertToLlm(providerRequest?.messages ?? requestMessages)),\n\t\t\t),\n\t\t\t...(options.snapshot.tools && options.snapshot.tools.length > 0 ? { tools: options.snapshot.tools } : {}),\n\t\t};\n\t\tconst headers = providerRequest\n\t\t\t? await providerRequest.transformHeaders(options.auth.headers ?? {})\n\t\t\t: options.auth.headers;\n\t\tconst responseStream = summarizationStream(options.context, options.snapshot.model, requestContext, {\n\t\t\tapiKey: options.auth.apiKey,\n\t\t\theaders,\n\t\t\textraBody: options.auth.extraBody,\n\t\t\tonPayload: async (payload, model) => {\n\t\t\t\tconst sanitized = model.api === \"anthropic-messages\" ? sanitizeAnthropicPayload(payload) : payload;\n\t\t\t\treturn providerRequest ? await providerRequest.transformPayload(sanitized) : sanitized;\n\t\t\t},\n\t\t\tmaxTokens: summaryMaxTokens(options.snapshot.model, options.snapshot.contextWindow),\n\t\t\tsignal: requestController.signal,\n\t\t\t...summarizationReasoningOptions(options.snapshot.model),\n\t\t});\n\t\tawait consumeStreamWithIdleTimeout(responseStream, {\n\t\t\tidleTimeoutMs: DEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS,\n\t\t\tmaxDurationMs: DEFAULT_SUMMARIZATION_MAX_DURATION_MS,\n\t\t\tabort: () => requestController.abort(),\n\t\t\tsignal: options.signal,\n\t\t\tonEvent: (event) => {\n\t\t\t\tif (event.type === \"text_delta\" && event.delta) {\n\t\t\t\t\toptions.onProgress?.(event.delta);\n\t\t\t\t}\n\t\t\t},\n\t\t});\n\t\treturn await responseStream.result();\n\t} finally {\n\t\tif (options.signal) options.signal.removeEventListener(\"abort\", onCallerAbort);\n\t}\n}\n"]}

package/dist/core/extensions/builtin/compaction/speculative.d.ts

@@ -6,6 +6,7 @@ import type { ModelRegistry } from "../../../model-registry.ts";
 import type { ReadonlySessionManager } from "../../../session-manager.ts";
 import type { ApplyCompactionResult, ContextUsage, ProviderRequestPreparation } from "../../types.ts";
 import { type MergedCompactionPromptVariant } from "./prompts.ts";
+export { createEmergencyPruneLatch, type EmergencyPruneLatch, hardLimitEmergencyPrune, truncateContextMessages, } from "./emergency-prune.ts";
 type CompactionProgressCallback = (delta: string) => void;
 export interface SpeculativeCompactionContext {
     model: Model<any> | undefined;
@@ -64,20 +65,15 @@ export declare class SummaryGenerationError extends Error {
     readonly kind: "auth" | "empty-summary";
     constructor(kind: "auth" | "empty-summary", message: string);
 }
-export declare function truncateContextMessages(messages: AgentMessage[]): AgentMessage[];
 /**
- * Sticky engage/release state for {@link hardLimitEmergencyPrune}. Callers that
- * issue many requests for one session share a single latch so the emitted context
- * shape stays stable while the estimate hovers around the engage threshold.
+ * Output budget for one summarization request. Adaptive-thinking models emit
+ * reasoning tokens before the summary text, so the legacy flat 8192 cap could
+ * be consumed entirely by thinking, ending the stream with zero text blocks.
+ * Grant generous headroom, clamped to what the model can actually emit and to
+ * half the context window: providers that enforce input + output <= window
+ * would otherwise reject the request for models advertising contextWindow ==
+ * maxTokens (every summarization request also carries conversation input).
  */
-export interface EmergencyPruneLatch {
-    engaged: boolean;
-}
-export declare function createEmergencyPruneLatch(): EmergencyPruneLatch;
-export declare function hardLimitEmergencyPrune(messages: AgentMessage[], contextWindow: number, latch?: EmergencyPruneLatch): {
-    messages: AgentMessage[];
-    needsAggressiveCompaction: boolean;
-};
 export declare function getPromptVariant(options: {
     reason: string;
     preparation: {
@@ -102,5 +98,4 @@ export declare function createSpeculativeCompactionSnapshot(context: Speculative
 export declare function runExtensionCompaction(context: SpeculativeCompactionContext, snapshot: SpeculativeCompactionSnapshot, signal?: AbortSignal, onProgress?: CompactionProgressCallback): Promise<CompactionResult | undefined>;
 export declare function applyGeneratedCompaction(context: SpeculativeCompactionContext, snapshot: SpeculativeCompactionSnapshot | undefined, getCurrentGeneration: () => number, compaction: CompactionResult | undefined, signal?: AbortSignal): Promise<SpeculativeCompactionResult>;
 export declare function applySpeculativeCompaction(context: SpeculativeCompactionContext, snapshot: SpeculativeCompactionSnapshot | undefined, getCurrentGeneration: () => number, generate: () => Promise<CompactionResult | undefined>): Promise<SpeculativeCompactionResult>;
-export {};
 //# sourceMappingURL=speculative.d.ts.map

package/dist/core/extensions/builtin/compaction/speculative.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"speculative.d.ts","sourceRoot":"","sources":["../../../../../src/core/extensions/builtin/compaction/speculative.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,+BAA+B,CAAC;AAClE,OAAO,EASN,KAAK,KAAK,EAMV,KAAK,IAAI,EACT,MAAM,uBAAuB,CAAC;AAE/B,OAAO,EACN,KAAK,qBAAqB,EAC1B,KAAK,gBAAgB,EAIrB,MAAM,8BAA8B,CAAC;AAQtC,OAAO,EAGN,KAAK,kBAAkB,EACvB,MAAM,oCAAoC,CAAC;AAE5C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAChE,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAC1E,OAAO,KAAK,EAAE,qBAAqB,EAAE,YAAY,EAAE,0BAA0B,EAAE,MAAM,gBAAgB,CAAC;AAWtG,OAAO,EAAe,KAAK,6BAA6B,EAAE,MAAM,cAAc,CAAC;AAmB/E,KAAK,0BAA0B,GAAG,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;AAE1D,MAAM,WAAW,4BAA4B;IAC5C,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC;IAC9B,cAAc,EAAE,sBAAsB,CAAC;IACvC,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,eAAe,IAAI,YAAY,GAAG,SAAS,CAAC;IAC5C,qBAAqB,CAAC,IAAI,qBAAqB,CAAC,UAAU,CAAC,CAAC;IAC5D,kBAAkB,IAAI,MAAM,CAAC;IAC7B,eAAe,CAAC,IAAI,MAAM,CAAC;IAC3B,sBAAsB,CAAC,CAAC,QAAQ,EAAE,YAAY,EAAE,GAAG,OAAO,CAAC,0BAA0B,CAAC,CAAC;IACvF,MAAM,CAAC,IAAI,OAAO,CAAC;IACnB,eAAe,CACd,WAAW,EAAE,gBAAgB,EAC7B,OAAO,EAAE;QACR,MAAM,EAAE,WAAW,CAAC;QACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,kBAAkB,CAAC,EAAE,kBAAkB,CAAC;QACxC,MAAM,CAAC,EAAE,WAAW,CAAC;KACrB,GACC,OAAO,CAAC,qBAAqB,CAAC,CAAC;CAClC;AAED,MAAM,WAAW,6BAA6B;IAC7C,UAAU,EAAE,MAAM,CAAC;IACnB,gBAAgB,EAAE,MAAM,CAAC;IACzB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,qBAAqB,CAAC;IACnC,aAAa,CAAC,EAAE,UAAU,CAAC,sBAAsB,CAAC,WAAW,CAAC,CAAC,CAAC;IAChE,aAAa,EAAE,6BAA6B,CAAC;IAC7C,MAAM,CAAC,EAAE,aAAa,GAAG,UAAU,GAAG,YAAY,CAAC;IACnD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,kGAAkG;IAClG,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,2FAA2F;IAC3F,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf;AAED,MAAM,MAAM,2BAA2B,GAAG,qBAAqB,GAAG;IAAE,OAAO,EAAE,KAAK,CAAC;IAAC,MAAM,EAAE,aAAa,GAAG,QAAQ,CAAA;CAAE,CAAC;AAMvH;;;;GAIG;AACH;;;;;GAKG;AACH,MAAM,MAAM,yBAAyB,GAAG,2BAA2B,CAAC;AAEpE,qBAAa,mBAAoB,SAAQ,KAAK;IAC7C,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC;IAC5B,QAAQ,CAAC,WAAW,CAAC,EAAE,yBAAyB,CAAC;gBAErC,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,EAAE,WAAW,CAAC,EAAE,yBAAyB;CAMxF;AA6BD,qBAAa,sBAAuB,SAAQ,KAAK;IAChD,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,eAAe,CAAC;gBAE5B,IAAI,EAAE,MAAM,GAAG,eAAe,EAAE,OAAO,EAAE,MAAM;CAK3D;AAyKD,wBAAgB,uBAAuB,CAAC,QAAQ,EAAE,YAAY,EAAE,GAAG,YAAY,EAAE,CAchF;AAED;;;;GAIG;AACH,MAAM,WAAW,mBAAmB;IACnC,OAAO,EAAE,OAAO,CAAC;CACjB;AAED,wBAAgB,yBAAyB,IAAI,mBAAmB,CAE/D;AAED,wBAAgB,uBAAuB,CACtC,QAAQ,EAAE,YAAY,EAAE,EACxB,aAAa,EAAE,MAAM,EACrB,KAAK,CAAC,EAAE,mBAAmB,GACzB;IACF,QAAQ,EAAE,YAAY,EAAE,CAAC;IACzB,yBAAyB,EAAE,OAAO,CAAC;CACnC,CAwBA;AAED,wBAAgB,gBAAgB,CAAC,OAAO,EAAE;IACzC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE;QAAE,eAAe,CAAC,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,OAAO,CAAA;KAAE,CAAC;CAChE,GAAG,6BAA6B,CAKhC;AAED,wBAAgB,mCAAmC,CAClD,OAAO,EAAE,4BAA4B,EACrC,OAAO,EAAE;IACR,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,aAAa,GAAG,UAAU,GAAG,YAAY,CAAC;IACnD,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf,GACC,6BAA6B,GAAG,SAAS,CA4B3C;AAED;;;;;;;GAOG;AACH,wBAAsB,sBAAsB,CAC3C,OAAO,EAAE,4BAA4B,EACrC,QAAQ,EAAE,6BAA6B,EACvC,MAAM,CAAC,EAAE,WAAW,EACpB,UAAU,CAAC,EAAE,0BAA0B,GACrC,OAAO,CAAC,gBAAgB,GAAG,SAAS,CAAC,CAkKvC;AAED,wBAAsB,wBAAwB,CAC7C,OAAO,EAAE,4BAA4B,EACrC,QAAQ,EAAE,6BAA6B,GAAG,SAAS,EACnD,oBAAoB,EAAE,MAAM,MAAM,EAClC,UAAU,EAAE,gBAAgB,GAAG,SAAS,EACxC,MAAM,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,2BAA2B,CAAC,CA2BtC;AAED,wBAAsB,0BAA0B,CAC/C,OAAO,EAAE,4BAA4B,EACrC,QAAQ,EAAE,6BAA6B,GAAG,SAAS,EACnD,oBAAoB,EAAE,MAAM,MAAM,EAClC,QAAQ,EAAE,MAAM,OAAO,CAAC,gBAAgB,GAAG,SAAS,CAAC,GACnD,OAAO,CAAC,2BAA2B,CAAC,CAKtC"}
+{"version":3,"file":"speculative.d.ts","sourceRoot":"","sources":["../../../../../src/core/extensions/builtin/compaction/speculative.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,+BAA+B,CAAC;AAClE,OAAO,EAON,KAAK,KAAK,EAEV,KAAK,IAAI,EACT,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACN,KAAK,qBAAqB,EAC1B,KAAK,gBAAgB,EAIrB,MAAM,8BAA8B,CAAC;AAEtC,OAAO,EAGN,KAAK,kBAAkB,EACvB,MAAM,oCAAoC,CAAC;AAE5C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAChE,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAC;AAC1E,OAAO,KAAK,EAAE,qBAAqB,EAAE,YAAY,EAAE,0BAA0B,EAAE,MAAM,gBAAgB,CAAC;AAUtG,OAAO,EAAe,KAAK,6BAA6B,EAAE,MAAM,cAAc,CAAC;AAO/E,OAAO,EACN,yBAAyB,EACzB,KAAK,mBAAmB,EACxB,uBAAuB,EACvB,uBAAuB,GACvB,MAAM,sBAAsB,CAAC;AAW9B,KAAK,0BAA0B,GAAG,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;AAE1D,MAAM,WAAW,4BAA4B;IAC5C,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC;IAC9B,cAAc,EAAE,sBAAsB,CAAC;IACvC,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,eAAe,IAAI,YAAY,GAAG,SAAS,CAAC;IAC5C,qBAAqB,CAAC,IAAI,qBAAqB,CAAC,UAAU,CAAC,CAAC;IAC5D,kBAAkB,IAAI,MAAM,CAAC;IAC7B,eAAe,CAAC,IAAI,MAAM,CAAC;IAC3B,sBAAsB,CAAC,CAAC,QAAQ,EAAE,YAAY,EAAE,GAAG,OAAO,CAAC,0BAA0B,CAAC,CAAC;IACvF,MAAM,CAAC,IAAI,OAAO,CAAC;IACnB,eAAe,CACd,WAAW,EAAE,gBAAgB,EAC7B,OAAO,EAAE;QACR,MAAM,EAAE,WAAW,CAAC;QACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,kBAAkB,CAAC,EAAE,kBAAkB,CAAC;QACxC,MAAM,CAAC,EAAE,WAAW,CAAC;KACrB,GACC,OAAO,CAAC,qBAAqB,CAAC,CAAC;CAClC;AAED,MAAM,WAAW,6BAA6B;IAC7C,UAAU,EAAE,MAAM,CAAC;IACnB,gBAAgB,EAAE,MAAM,CAAC;IACzB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,qBAAqB,CAAC;IACnC,aAAa,CAAC,EAAE,UAAU,CAAC,sBAAsB,CAAC,WAAW,CAAC,CAAC,CAAC;IAChE,aAAa,EAAE,6BAA6B,CAAC;IAC7C,MAAM,CAAC,EAAE,aAAa,GAAG,UAAU,GAAG,YAAY,CAAC;IACnD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,kGAAkG;IAClG,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,2FAA2F;IAC3F,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf;AAED,MAAM,MAAM,2BAA2B,GAAG,qBAAqB,GAAG;IAAE,OAAO,EAAE,KAAK,CAAC;IAAC,MAAM,EAAE,aAAa,GAAG,QAAQ,CAAA;CAAE,CAAC;AAMvH;;;;GAIG;AACH;;;;;GAKG;AACH,MAAM,MAAM,yBAAyB,GAAG,2BAA2B,CAAC;AAEpE,qBAAa,mBAAoB,SAAQ,KAAK;IAC7C,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC;IAC5B,QAAQ,CAAC,WAAW,CAAC,EAAE,yBAAyB,CAAC;gBAErC,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,EAAE,WAAW,CAAC,EAAE,yBAAyB;CAMxF;AA6BD,qBAAa,sBAAuB,SAAQ,KAAK;IAChD,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,eAAe,CAAC;gBAE5B,IAAI,EAAE,MAAM,GAAG,eAAe,EAAE,OAAO,EAAE,MAAM;CAK3D;AAED;;;;;;;;GAQG;AAEH,wBAAgB,gBAAgB,CAAC,OAAO,EAAE;IACzC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE;QAAE,eAAe,CAAC,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,OAAO,CAAA;KAAE,CAAC;CAChE,GAAG,6BAA6B,CAKhC;AAED,wBAAgB,mCAAmC,CAClD,OAAO,EAAE,4BAA4B,EACrC,OAAO,EAAE;IACR,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,aAAa,GAAG,UAAU,GAAG,YAAY,CAAC;IACnD,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf,GACC,6BAA6B,GAAG,SAAS,CAyB3C;AAED;;;;;;;GAOG;AACH,wBAAsB,sBAAsB,CAC3C,OAAO,EAAE,4BAA4B,EACrC,QAAQ,EAAE,6BAA6B,EACvC,MAAM,CAAC,EAAE,WAAW,EACpB,UAAU,CAAC,EAAE,0BAA0B,GACrC,OAAO,CAAC,gBAAgB,GAAG,SAAS,CAAC,CAkKvC;AAED,wBAAsB,wBAAwB,CAC7C,OAAO,EAAE,4BAA4B,EACrC,QAAQ,EAAE,6BAA6B,GAAG,SAAS,EACnD,oBAAoB,EAAE,MAAM,MAAM,EAClC,UAAU,EAAE,gBAAgB,GAAG,SAAS,EACxC,MAAM,CAAC,EAAE,WAAW,GAClB,OAAO,CAAC,2BAA2B,CAAC,CA2BtC;AAED,wBAAsB,0BAA0B,CAC/C,OAAO,EAAE,4BAA4B,EACrC,QAAQ,EAAE,6BAA6B,GAAG,SAAS,EACnD,oBAAoB,EAAE,MAAM,MAAM,EAClC,QAAQ,EAAE,MAAM,OAAO,CAAC,gBAAgB,GAAG,SAAS,CAAC,GACnD,OAAO,CAAC,2BAA2B,CAAC,CAKtC"}

package/dist/core/extensions/builtin/compaction/speculative.js

@@ -1,28 +1,23 @@
-import { hasCredentialHeaders, isContextOverflow, isRetryableAssistantError, isRetryableErrorMessage, retryTransientCall, sanitizeAnthropicToolPairs as sanitizeAnthropicPayload, } from "@earendil-works/pi-ai";
-import { stream } from "@earendil-works/pi-ai/compat";
+import { hasCredentialHeaders, isContextOverflow, isRetryableAssistantError, isRetryableErrorMessage, retryTransientCall, } from "@earendil-works/pi-ai";
 import { DEFAULT_COMPACTION_SETTINGS, estimateContextTokens, prepareCompaction, } from "../../../compaction/index.js";
-import { consumeStreamWithIdleTimeout, DEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS, DEFAULT_SUMMARIZATION_MAX_DURATION_MS, StreamDurationBudgetError, StreamIdleTimeoutError, } from "../../../compaction/stream-watchdog.js";
+import { StreamDurationBudgetError, StreamIdleTimeoutError } from "../../../compaction/stream-watchdog.js";
 import { createWarmAnchorSnapshot, isWarmSummaryAnchorValid, } from "../../../compaction/warm-anchor.js";
 import { convertToLlm } from "../../../messages.js";
-import { allowOverflowRetry, boundSummarizationInput, estimateTotalTokens, pruneOldMessagesToBudget, SUMMARIZATION_INPUT_BUDGET_RATIO, SummarizationOverflowExhaustedError, shrinkSummarizationInputForOverflowRetry, } from "./overflow-retry.js";
+import { pruneToolResults } from "./emergency-prune.js";
+import { allowOverflowRetry, boundSummarizationInput, SUMMARIZATION_INPUT_BUDGET_RATIO, SummarizationOverflowExhaustedError, shrinkSummarizationInputForOverflowRetry, } from "./overflow-retry.js";
 import { computeEffectiveKeepRecentTokens, computeEffectiveThreshold } from "./policy.js";
 import { buildPrompt } from "./prompts.js";
-import { repairOrphanedToolResults } from "./repair-tool-pairs.js";
+import { generateSummaryMessage, getSummaryText, isAssistantMessage } from "./speculative-summary.js";
 import { allowSummarizationRetry, DEFAULT_SUMMARIZATION_RETRY_POLICY } from "./summarization-retry.js";
-import { normalizeSummarizationTurnOrder } from "./summarization-turn-order.js";
 import { extractTaskIntent, resolveInheritedTaskIntent } from "./task-intent.js";
-import * as truncation from "./tool-truncation.js";
+export { createEmergencyPruneLatch, hardLimitEmergencyPrune, truncateContextMessages, } from "./emergency-prune.js";
 import { computeStructuralYield } from "./yield.js";
 const DEFAULT_CONTEXT_WINDOW = 200_000;
-const EMERGENCY_CONTEXT_TARGET_RATIO = 0.95;
 // Hysteresis: the emergency prune engages at EMERGENCY_CONTEXT_TARGET_RATIO but only
 // releases once the context falls below this lower ratio. A single threshold makes a
 // session parked near the limit alternate between the pruned and un-pruned history on
 // consecutive requests; because pruning rewrites old tool results, every alternation
 // invalidates the provider prompt-cache prefix and re-bills the whole conversation.
-const EMERGENCY_CONTEXT_RELEASE_RATIO = 0.85;
-const SUMMARY_TOKEN_HEADROOM = 32_768;
-const SUMMARY_CONTEXT_WINDOW_RESERVE_RATIO = 0.5;
 const SUMMARY_SCHEMA = "senpi.compaction.summary.v1";
 function approxTokens(text) {
     return Math.ceil(text.length / 4);
@@ -81,187 +76,6 @@ export class SummaryGenerationError extends Error {
  * would otherwise reject the request for models advertising contextWindow ==
  * maxTokens (every summarization request also carries conversation input).
  */
-function summaryMaxTokens(model, contextWindow) {
-    const headroom = model.maxTokens > 0 ? Math.min(SUMMARY_TOKEN_HEADROOM, model.maxTokens) : SUMMARY_TOKEN_HEADROOM;
-    if (contextWindow > 0) {
-        return Math.min(headroom, Math.floor(contextWindow * SUMMARY_CONTEXT_WINDOW_RESERVE_RATIO));
-    }
-    return headroom;
-}
-/**
- * Reasoning override for summarization requests. Compaction must be fast: a
- * summarization request that inherits the provider's default reasoning mode
- * burns its latency (and output budget) on invisible thinking before emitting
- * the summary. Disable or minimize reasoning per wire family; adapters ignore
- * options their provider does not support. Mirrors how OpenAI Codex keeps its
- * compaction turn cheap.
- */
-function summarizationReasoningOptions(model) {
-    if (!model.reasoning)
-        return {};
-    if (model.api === "anthropic-messages")
-        return { thinkingEnabled: false };
-    const reasoningEffort = ["low", "medium", "high"].find((level) => model.thinkingLevelMap?.[level] !== null);
-    if (!reasoningEffort)
-        return {};
-    switch (model.api) {
-        case "openai-responses":
-        case "openai-codex-responses":
-        case "azure-openai-responses":
-            return { reasoningEffort, reasoningSummary: null };
-        case "openai-completions":
-            return { reasoningEffort };
-        default:
-            return {};
-    }
-}
-function getSummaryText(message) {
-    const content = Array.isArray(message.content)
-        ? message.content
-        : [{ type: "text", text: message.content }];
-    return content
-        .filter((content) => content.type === "text")
-        .map((content) => content.text)
-        .join("\n")
-        .trim();
-}
-function isAssistantMessage(message) {
-    return message.role === "assistant" && "stopReason" in message;
-}
-/**
- * Providers registered through `pi.registerProvider()` (claude-sdk-oauth, Kiro, any
- * extension provider) exist only in Senpi's ModelRuntime, never in compat's builtin
- * api-registry, which rejects their api id outright. Dispatch through the runtime
- * whenever it is reachable and keep compat for contexts constructed without a registry.
- */
-function summarizationStream(context, model, requestContext, options) {
-    const runtime = context.modelRegistry?.modelRuntime;
-    return runtime ? runtime.stream(model, requestContext, options) : stream(model, requestContext, options);
-}
-async function generateSummaryMessage(options) {
-    // Send the conversation as native LLM messages with the summarization
-    // instruction as a trailing user message, mirroring normal agent traffic.
-    // A single serialized `<conversation>` text dump of a large session is
-    // deterministically refused by Anthropic's anti-distillation classifier
-    // ("reverse engineering or duplicating model outputs"), while the same
-    // content as native blocks with the agent's system prompt and tools passes.
-    // Request-local controller: the idle watchdog must be able to tear down a
-    // stalled summarization request without aborting the caller's own signal.
-    const requestController = new AbortController();
-    const onCallerAbort = () => requestController.abort(options.signal?.reason);
-    if (options.signal) {
-        if (options.signal.aborted)
-            onCallerAbort();
-        else
-            options.signal.addEventListener("abort", onCallerAbort, { once: true });
-    }
-    try {
-        const requestMessages = [
-            ...options.messages,
-            {
-                role: "user",
-                content: [{ type: "text", text: options.prompt.user }],
-                timestamp: Date.now(),
-            },
-        ];
-        const providerRequest = await options.context.prepareProviderRequest?.(requestMessages);
-        const requestContext = {
-            systemPrompt: options.snapshot.systemPrompt ?? options.prompt.system,
-            messages: repairOrphanedToolResults(normalizeSummarizationTurnOrder(convertToLlm(providerRequest?.messages ?? requestMessages))),
-            ...(options.snapshot.tools && options.snapshot.tools.length > 0 ? { tools: options.snapshot.tools } : {}),
-        };
-        const headers = providerRequest
-            ? await providerRequest.transformHeaders(options.auth.headers ?? {})
-            : options.auth.headers;
-        const responseStream = summarizationStream(options.context, options.snapshot.model, requestContext, {
-            apiKey: options.auth.apiKey,
-            headers,
-            extraBody: options.auth.extraBody,
-            onPayload: async (payload, model) => {
-                const sanitized = model.api === "anthropic-messages" ? sanitizeAnthropicPayload(payload) : payload;
-                return providerRequest ? await providerRequest.transformPayload(sanitized) : sanitized;
-            },
-            maxTokens: summaryMaxTokens(options.snapshot.model, options.snapshot.contextWindow),
-            signal: requestController.signal,
-            ...summarizationReasoningOptions(options.snapshot.model),
-        });
-        await consumeStreamWithIdleTimeout(responseStream, {
-            idleTimeoutMs: DEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS,
-            maxDurationMs: DEFAULT_SUMMARIZATION_MAX_DURATION_MS,
-            abort: () => requestController.abort(),
-            signal: options.signal,
-            onEvent: (event) => {
-                if (event.type === "text_delta" && event.delta) {
-                    options.onProgress?.(event.delta);
-                }
-            },
-        });
-        return await responseStream.result();
-    }
-    finally {
-        if (options.signal)
-            options.signal.removeEventListener("abort", onCallerAbort);
-    }
-}
-function pruneToolResults(messages, contextWindow, budgetRatio) {
-    const toolResults = messages
-        .filter((message) => message.role === "toolResult")
-        .map((message) => ({ content: message.content, details: undefined }));
-    if (toolResults.length === 0)
-        return messages;
-    const prunedResults = truncation.prePruneToolOutputsToBudget(toolResults, contextWindow * budgetRatio);
-    let resultIndex = 0;
-    return messages.map((message) => {
-        if (message.role !== "toolResult")
-            return message;
-        const pruned = prunedResults[resultIndex];
-        resultIndex++;
-        return pruned ? { ...message, content: pruned.content } : message;
-    });
-}
-export function truncateContextMessages(messages) {
-    const toolResults = messages
-        .filter((message) => message.role === "toolResult")
-        .map((message) => ({ content: message.content, details: undefined }));
-    if (toolResults.length === 0)
-        return messages;
-    const truncatedResults = truncation.truncateOversizedToolResults(toolResults);
-    let resultIndex = 0;
-    return messages.map((message) => {
-        if (message.role !== "toolResult")
-            return message;
-        const truncated = truncatedResults[resultIndex];
-        resultIndex++;
-        return truncated ? { ...message, content: truncated.content } : message;
-    });
-}
-export function createEmergencyPruneLatch() {
-    return { engaged: false };
-}
-export function hardLimitEmergencyPrune(messages, contextWindow, latch) {
-    const targetTokens = Math.floor(contextWindow * EMERGENCY_CONTEXT_TARGET_RATIO);
-    const releaseTokens = Math.floor(contextWindow * EMERGENCY_CONTEXT_RELEASE_RATIO);
-    const totalTokens = estimateTotalTokens(messages);
-    // Without a latch this keeps the historical single-threshold behaviour.
-    const engaged = latch
-        ? latch.engaged
-            ? totalTokens > releaseTokens
-            : totalTokens > targetTokens
-        : totalTokens > targetTokens;
-    if (latch)
-        latch.engaged = engaged;
-    if (!engaged) {
-        return { messages, needsAggressiveCompaction: false };
-    }
-    const noLlmPruned = truncateContextMessages(pruneToolResults(messages, contextWindow, SUMMARIZATION_INPUT_BUDGET_RATIO));
-    if (estimateTotalTokens(noLlmPruned) <= targetTokens) {
-        return { messages: noLlmPruned, needsAggressiveCompaction: false };
-    }
-    return {
-        messages: pruneOldMessagesToBudget(noLlmPruned, targetTokens),
-        needsAggressiveCompaction: true,
-    };
-}
 export function getPromptVariant(options) {
     if (options.reason === "branch")
         return "branch";
@@ -287,17 +101,14 @@ export function createSpeculativeCompactionSnapshot(context, options) {
     if (!preparation)
         return undefined;
     return {
-        generation: options.generation,
+        ...options,
         expectedRevision,
         model,
         contextWindow,
         preparation,
         branchEntries,
         promptVariant: getPromptVariant({ reason: "extension", preparation }),
-        ...(options.origin ? { origin: options.origin } : {}),
-        customInstructions: options.customInstructions,
         systemPrompt: context.getSystemPrompt?.(),
-        tools: options.tools,
     };
 }
 /**