@averyyy/pi-coding-agent 0.87.1-piclient.1 → 0.87.1-piclient.2

package/dist/core/agent-session.js

@@ -33,7 +33,8 @@ import { ExtensionRunner, wrapRegisteredTools, } from "./extensions/index.js";
 import { emitSessionShutdownEvent } from "./extensions/runner.js";
 import { convertToLlm } from "./messages.js";
 import { ModelRegistry } from "./model-registry.js";
-import { cancelPiServerOperations, compactPiServer, } from "./pi-server-client.js";
+import { cancelPiServerOperations, compactPiServer, hashStaticContext, } from "./pi-server-client.js";
+import { hashPiServerProviderMessages } from "./pi-server-protocol.js";
 import { expandPromptTemplate } from "./prompt-templates.js";
 import { CURRENT_SESSION_VERSION, getLatestCompactionEntry, SessionManager, } from "./session-manager.js";
 import { createSyntheticSourceInfo } from "./source-info.js";
@@ -195,6 +196,7 @@ export class AgentSession {
     _prePromptCompletionPromise = undefined;
     _overflowRecoveryAttempted = false;
     _lastServerCompactionTimestamp = undefined;
+    _cachePrefixUnavailableReason;
     // Branch summarization state
     _branchSummaryAbortController = undefined;
     _branchSummaryCompletionPromise = undefined;
@@ -427,6 +429,8 @@ export class AgentSession {
     }
     _installAgentRequestProjection() {
         const previousPrepareRequest = this.agent.prepareRequest;
+        if (previousPrepareRequest)
+            this._cachePrefixUnavailableReason ??= "custom_prepare_request";
         this.agent.prepareRequest = async (request, signal) => {
             const canonicalContext = {
                 ...request.context,
@@ -1264,6 +1268,8 @@ export class AgentSession {
      */
     _installAgentForcedPromptProjection() {
         const previousTransformContext = this.agent.transformContext;
+        if (previousTransformContext)
+            this._cachePrefixUnavailableReason ??= "custom_context_transform";
         this.agent.transformContext = async (messages, signal) => {
             const transformed = previousTransformContext ? await previousTransformContext(messages, signal) : messages;
             const forced = this._runSystemPromptOptions?.forceSystemPrompt;
@@ -2280,6 +2286,58 @@ export class AgentSession {
     // =========================================================================
     // Compaction
     // =========================================================================
+    async _buildCompactionCacheInput(model, preparation) {
+        if (this._cachePrefixUnavailableReason)
+            return { reason: this._cachePrefixUnavailableReason };
+        if (this.agent.onPayload)
+            return { reason: "request_payload_hook_active" };
+        if (this._runSystemPromptOptions?.forceSystemPrompt !== undefined) {
+            return { reason: "forced_system_prompt_active" };
+        }
+        const activeModel = this.agent.state.model;
+        if (!modelsAreEqual(model, activeModel) ||
+            model.api !== activeModel.api ||
+            model.baseUrl !== activeModel.baseUrl) {
+            return { reason: "summary_model_mismatch" };
+        }
+        if (!this.agent.sessionId)
+            return { reason: "session_id_unavailable" };
+        const projectedMessages = this.sessionManager.buildSessionProjection().messages;
+        if (JSON.stringify(projectedMessages) !== JSON.stringify(this.agent.state.messages)) {
+            return { reason: "request_context_mismatch" };
+        }
+        const fullMessages = await this.agent.convertToLlm(projectedMessages);
+        const fullContext = normalizeContext({ messages: fullMessages });
+        const cacheOptions = {
+            sessionId: this.agent.sessionId,
+            transport: this.agent.transport,
+            thinkingBudgets: this.agent.thinkingBudgets,
+            maxRetryDelayMs: this.agent.maxRetryDelayMs,
+        };
+        if (!preparation) {
+            return {
+                cacheContextHash: hashPiServerProviderMessages(fullMessages),
+                cacheStaticContextHash: hashStaticContext({
+                    systemPrompt: this.systemPrompt,
+                    tools: this.agent.state.tools,
+                    messages: [],
+                }),
+                cacheOptions,
+                reason: "exact_prefix_unavailable",
+            };
+        }
+        const prefixMessages = await this.agent.convertToLlm(preparation.cachePrefixMessages);
+        const cachePrefixContext = normalizeContext({ messages: prefixMessages });
+        if (cachePrefixContext.messages.length > fullContext.messages.length ||
+            cachePrefixContext.messages.some((message, index) => JSON.stringify(message) !== JSON.stringify(fullContext.messages[index]))) {
+            return { reason: "projected_prefix_mismatch" };
+        }
+        return {
+            cachePrefixContext,
+            cacheOptions,
+            reason: "exact_prefix_unavailable",
+        };
+    }
     async _applyPiServerCompactionResult(result, reason, willRetry) {
         this.sessionManager.replaceTree(result.entries, result.leafId);
         const sessionContext = this.sessionManager.buildSessionContext();
@@ -2338,15 +2396,20 @@ export class AgentSession {
             const { model: requestModel, apiKey, headers, env, } = await this._getSummarizationRequestAuth(model, this._compactionAbortController.signal);
             if (isPiServerMode()) {
                 this._detachPiServerTerminalAssistantFailure();
+                const cacheInput = await this._buildCompactionCacheInput(requestModel);
                 const piServerResult = await compactPiServer(requestModel, {
                     systemPrompt: this.systemPrompt,
                     messages: this.agent.state.messages,
                     tools: this.agent.state.tools,
                 }, {
+                    ...cacheInput.cacheOptions,
                     sessionId: this.sessionId,
                     apiKey,
                     headers,
                     customInstructions,
+                    cacheContextHash: cacheInput.cacheContextHash,
+                    cacheStaticContextHash: cacheInput.cacheStaticContextHash,
+                    cacheFallbackReason: cacheInput.reason,
                     settings: this.settingsManager.getCompactionSettings(),
                     sessionTree: buildPiServerSessionTree(this.sessionManager.getEntries(), this.sessionManager.getLeafId()),
                     signal: this._compactionAbortController.signal,
@@ -2409,7 +2472,8 @@ export class AgentSession {
             }
             else {
                 // Generate compaction result
-                const result = await compact(preparation, requestModel, apiKey, headers, customInstructions, this._compactionAbortController.signal, this.thinkingLevel, this.agent.streamFunction, env, this.settingsManager.getRetrySettings(), this._summarizationRetryCallbacks({ source: "compaction", reason: "manual" }));
+                const cacheInput = await this._buildCompactionCacheInput(requestModel, preparation);
+                const result = await compact(preparation, requestModel, apiKey, headers, customInstructions, this._compactionAbortController.signal, this.thinkingLevel, this.agent.streamFunction, env, this.settingsManager.getRetrySettings(), this._summarizationRetryCallbacks({ source: "compaction", reason: "manual" }), undefined, cacheInput.cachePrefixContext, cacheInput.cacheOptions, cacheInput.reason);
                 summary = result.summary;
                 firstKeptEntryId = result.firstKeptEntryId;
                 tokensBefore = result.tokensBefore;
@@ -2650,14 +2714,19 @@ export class AgentSession {
                 abortController.signal.throwIfAborted();
                 const { model: requestModel, apiKey, headers, } = await this._getSummarizationRequestAuth(model, abortController.signal);
                 abortController.signal.throwIfAborted();
+                const cacheInput = await this._buildCompactionCacheInput(requestModel);
                 const piServerResult = await compactPiServer(requestModel, {
                     systemPrompt: this.systemPrompt,
                     messages: this.agent.state.messages,
                     tools: this.agent.state.tools,
                 }, {
+                    ...cacheInput.cacheOptions,
                     sessionId: this.sessionId,
                     apiKey,
                     headers,
+                    cacheContextHash: cacheInput.cacheContextHash,
+                    cacheStaticContextHash: cacheInput.cacheStaticContextHash,
+                    cacheFallbackReason: cacheInput.reason,
                     settings,
                     preparation: preparationOptions,
                     sessionTree: buildPiServerSessionTree(this.sessionManager.getEntries(), this.sessionManager.getLeafId()),
@@ -2726,7 +2795,8 @@ export class AgentSession {
                 details = extensionCompaction.details;
             }
             else {
-                const compactResult = await compact(preparation, requestModel, apiKey, headers, undefined, abortController.signal, this.thinkingLevel, this.agent.streamFunction, env, this.settingsManager.getRetrySettings(), this._summarizationRetryCallbacks({ source: "compaction", reason }));
+                const cacheInput = await this._buildCompactionCacheInput(requestModel, preparation);
+                const compactResult = await compact(preparation, requestModel, apiKey, headers, undefined, abortController.signal, this.thinkingLevel, this.agent.streamFunction, env, this.settingsManager.getRetrySettings(), this._summarizationRetryCallbacks({ source: "compaction", reason }), undefined, cacheInput.cachePrefixContext, cacheInput.cacheOptions, cacheInput.reason);
                 summary = compactResult.summary;
                 firstKeptEntryId = compactResult.firstKeptEntryId;
                 tokensBefore = compactResult.tokensBefore;