@proxysoul/soulforge 2.16.0 → 2.16.1

package/dist/index.js

@@ -63415,7 +63415,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@proxysoul/soulforge",
-    version: "2.16.0",
+    version: "2.16.1",
     description: "Graph-powered code intelligence \u2014 multi-agent coding with codebase-aware AI",
     repository: {
       type: "git",
@@ -95850,6 +95850,22 @@ function buildContextEdits(config2, contextWindow, thinkingEnabled) {
   }
   return edits.length > 0 ? edits : null;
 }
+function buildCacheProviderOptions(ttl) {
+  const cache = ttl === "1h" ? CACHE_EPHEMERAL_1H : CACHE_EPHEMERAL_5M;
+  return {
+    anthropic: cache,
+    google: cache,
+    proxy: cache,
+    llmgateway: cache,
+    opencode_zen: cache,
+    opencode_go: cache,
+    openrouter: cache,
+    vercel_gateway: cache
+  };
+}
+function getEphemeralCache(ttl = "5m") {
+  return ttl === "1h" ? EPHEMERAL_CACHE_1H : EPHEMERAL_CACHE_5M;
+}
 async function buildAnthropicOptions(modelId, caps, config2) {
   const opts = {};
   const headers = {};
@@ -96186,7 +96202,7 @@ function buildGroqOptions(config2) {
   }
   return { opts };
 }
-var NO_SUPPORT, ANTHROPIC_FULL, OPENAI_FULL, GOOGLE_FULL, XAI_FULL, DEEPSEEK_FULL, OPENROUTER_FULL, GATEWAY_FULL, COMPAT_ONLY, PROVIDER_CONSTRAINTS, LEGACY_PREFIXES, CACHE_EPHEMERAL, EPHEMERAL_CACHE;
+var NO_SUPPORT, ANTHROPIC_FULL, OPENAI_FULL, GOOGLE_FULL, XAI_FULL, DEEPSEEK_FULL, OPENROUTER_FULL, GATEWAY_FULL, COMPAT_ONLY, PROVIDER_CONSTRAINTS, LEGACY_PREFIXES, CACHE_EPHEMERAL_5M, CACHE_EPHEMERAL_1H, EPHEMERAL_CACHE_5M, EPHEMERAL_CACHE_1H, EPHEMERAL_CACHE;
 var init_provider_options = __esm(() => {
   init_models();
   init_providers();
@@ -96300,17 +96316,15 @@ var init_provider_options = __esm(() => {
     "claude-2",
     "claude-instant"
   ];
-  CACHE_EPHEMERAL = { cacheControl: { type: "ephemeral" } };
-  EPHEMERAL_CACHE = {
-    anthropic: CACHE_EPHEMERAL,
-    google: CACHE_EPHEMERAL,
-    proxy: CACHE_EPHEMERAL,
-    llmgateway: CACHE_EPHEMERAL,
-    opencode_zen: CACHE_EPHEMERAL,
-    opencode_go: CACHE_EPHEMERAL,
-    openrouter: CACHE_EPHEMERAL,
-    vercel_gateway: CACHE_EPHEMERAL
+  CACHE_EPHEMERAL_5M = {
+    cacheControl: { type: "ephemeral", ttl: "5m" }
   };
+  CACHE_EPHEMERAL_1H = {
+    cacheControl: { type: "ephemeral", ttl: "1h" }
+  };
+  EPHEMERAL_CACHE_5M = buildCacheProviderOptions("5m");
+  EPHEMERAL_CACHE_1H = buildCacheProviderOptions("1h");
+  EPHEMERAL_CACHE = EPHEMERAL_CACHE_5M;
 });
 
 // src/core/prompts/families/shared-rules.ts
@@ -96329,7 +96343,7 @@ A turn is tool calls followed by exactly one final answer. Between tool calls: z
 
 After the last tool: speak. The final answer is mandatory \u2014 every turn ends with text, never on a tool result. Speak only when (a) the task is complete, (b) a destructive/irreversible action needs confirmation, (c) genuine ambiguity blocks progress, or (d) an unrecoverable error makes further tool calls pointless. Warning about a destructive action: the warning IS the answer \u2014 full sentences, no tool chain first.
 
-Commit boundary: when a turn has 2+ tool calls, call \`set_lockin({on:false})\` as your LAST tool immediately before the final answer. The renderer folds prior tool work into a collapsed rail and streams the answer text visibly. Skip the call for pure-chat turns (no tool work) and for one-tool turns. Don't call it before another tool \u2014 it must be last.
+Commit boundary \u2014 MANDATORY whenever a turn uses 2+ tool calls (parallel batches in one step count as 2+). Call \`set_lockin({on:false})\` as your LAST tool, after every other tool, immediately before the final answer. Not optional. Not "if convenient." Every multi-tool turn ends with set_lockin \u2192 text. Skip ONLY for pure-chat turns (zero tools) and single-tool turns. Never call before another tool \u2014 it must be the absolute last tool of the turn.
 </tool_loop>
 
 <answer_voice>
@@ -390901,7 +390915,10 @@ function createCodeAgent(model, options) {
     disablePruning: options?.disablePruning,
     tabId: options?.tabId
   });
-  const { maxTransientRetries: retryMaxRetries } = resolveRetrySettings(loadConfig().retry, {
+  const cfg = loadConfig();
+  const cacheTtl = cfg.cache?.ttl ?? "5m";
+  const cacheOpts = getEphemeralCache(cacheTtl);
+  const { maxTransientRetries: retryMaxRetries } = resolveRetrySettings(cfg.retry, {
     agent: true
   });
   return new ToolLoopAgent({
@@ -390927,7 +390944,7 @@ Ownership: you own files you edit first. check_edit_conflicts before touching an
 If another agent owns the file: report_finding with the exact edit instead.
 Coordination: report_finding after significant changes (paths, what changed, new exports). Peer findings appear in tool results.`;
       })(),
-      providerOptions: EPHEMERAL_CACHE
+      providerOptions: cacheOpts
     },
     stopWhen: stopConditions,
     prepareStep,
@@ -390936,7 +390953,7 @@ Coordination: report_finding after significant changes (paths, what changed, new
       ...options?.providerOptions,
       anthropic: {
         ...options?.providerOptions?.anthropic ?? {},
-        cacheControl: { type: "ephemeral" },
+        cacheControl: { type: "ephemeral", ttl: cacheTtl },
         max_tokens: MAX_OUTPUT_TOKENS
       }
     },
@@ -391010,7 +391027,10 @@ function createExploreAgent(model, options) {
     disablePruning: options?.disablePruning,
     tabId: options?.tabId
   });
-  const { maxTransientRetries: retryMaxRetries } = resolveRetrySettings(loadConfig().retry, {
+  const cfg = loadConfig();
+  const cacheTtl = cfg.cache?.ttl ?? "5m";
+  const cacheOpts = getEphemeralCache(cacheTtl);
+  const { maxTransientRetries: retryMaxRetries } = resolveRetrySettings(cfg.retry, {
     agent: true
   });
   return new ToolLoopAgent({
@@ -391034,7 +391054,7 @@ function createExploreAgent(model, options) {
         return `${base}
 Coordination: report_finding after discoveries \u2014 especially shared symbols/configs with peer targets. check_findings for peer detail.`;
       })(),
-      providerOptions: EPHEMERAL_CACHE
+      providerOptions: cacheOpts
     },
     stopWhen: stopConditions,
     prepareStep,
@@ -391043,7 +391063,7 @@ Coordination: report_finding after discoveries \u2014 especially shared symbols/
       ...options?.providerOptions,
       anthropic: {
         ...options?.providerOptions?.anthropic ?? {},
-        cacheControl: { type: "ephemeral" },
+        cacheControl: { type: "ephemeral", ttl: cacheTtl },
         max_tokens: MAX_OUTPUT_TOKENS
       }
     },
@@ -396777,14 +396797,10 @@ function lastStepHadPlanCall(messages) {
   }
   return false;
 }
-function buildForgePrepareStep(isPlanMode, drainSteering, contextManager, tabId, codeExecution3, parentMessagesRef, proxyInstructions) {
+function buildForgePrepareStep(isPlanMode, drainSteering, contextManager, tabId, codeExecution3, parentMessagesRef, proxyInstructions, cacheOpts = EPHEMERAL_CACHE) {
   const previousInjects = [];
   const recallInjects = [];
   let lastUserTurnCount = 0;
-  let lockInToolCallsThisTurn = 0;
-  let lockInNudgedOffThisTurn = false;
-  let lockInLastUserTurnCount = 0;
-  let lockInCommittedThisTurn = false;
   const proxyInstructionsMessage = proxyInstructions ? {
     role: "user",
     content: [
@@ -396795,7 +396811,7 @@ ${proxyInstructions}
 </system-instructions>`
       }
     ],
-    providerOptions: EPHEMERAL_CACHE
+    providerOptions: cacheOpts
   } : null;
   return async ({
     stepNumber,
@@ -396845,12 +396861,12 @@ ${proxyInstructions}
                   {
                     role: "user",
                     content: pair[0].content,
-                    providerOptions: EPHEMERAL_CACHE
+                    providerOptions: cacheOpts
                   },
                   {
                     role: "assistant",
                     content: pair[1].content,
-                    providerOptions: EPHEMERAL_CACHE
+                    providerOptions: cacheOpts
                   }
                 ]
               });
@@ -396915,26 +396931,30 @@ ${proxyInstructions}
         hints.push(crossTab);
     }
     {
-      const utc = countUserTurns(sanitized);
-      if (utc > lockInLastUserTurnCount) {
-        lockInLastUserTurnCount = utc;
-        lockInToolCallsThisTurn = 0;
-        lockInNudgedOffThisTurn = false;
-        lockInCommittedThisTurn = false;
-      }
-      const prev = prevStep;
-      const lastHadTools = !!prev?.toolCalls?.length;
-      if (lastHadTools) {
-        lockInToolCallsThisTurn += prev.toolCalls?.length ?? 0;
-        for (const tc of prev.toolCalls ?? []) {
-          if (tc.toolName === "set_lockin" && tc.input?.on === false) {
-            lockInCommittedThisTurn = true;
+      let toolCallsThisTurn = 0;
+      let committedThisTurn = false;
+      for (let i2 = sanitized.length - 1;i2 >= 0; i2--) {
+        const m = sanitized[i2];
+        if (!m)
+          continue;
+        if (m.role === "user")
+          break;
+        if (m.role !== "assistant" || !Array.isArray(m.content))
+          continue;
+        for (const part of m.content) {
+          if (typeof part !== "object" || part === null || !("type" in part))
+            continue;
+          const p = part;
+          if (p.type !== "tool-call")
+            continue;
+          toolCallsThisTurn++;
+          if (p.toolName === "set_lockin" && p.input?.on === false) {
+            committedThisTurn = true;
           }
         }
       }
-      if (!lockInCommittedThisTurn && !lockInNudgedOffThisTurn && lockInToolCallsThisTurn >= 2) {
+      if (!committedThisTurn && toolCallsThisTurn >= 2) {
         hints.push("Multiple tool calls this turn without a commit boundary. Call set_lockin({on:false}) as your LAST tool before your final answer so prior tool work collapses into the rail and your text streams visibly.");
-        lockInNudgedOffThisTurn = true;
       }
     }
     if (stepNumber >= PERSONA_NUDGE_START && (stepNumber - PERSONA_NUDGE_START) % PERSONA_NUDGE_INTERVAL === 0) {
@@ -397319,11 +397339,13 @@ function createForgeAgent({
     }
     return names.length < allToolNames.length ? names : undefined;
   };
+  const cacheTtl = loadConfig().cache?.ttl ?? "5m";
+  const cacheOpts = getEphemeralCache(cacheTtl);
   const wrappedProviderOptions = {
     ...providerOptions,
     anthropic: {
       ...providerOptions?.anthropic ?? {},
-      cacheControl: { type: "ephemeral" },
+      cacheControl: { type: "ephemeral", ttl: cacheTtl },
       max_tokens: MAX_OUTPUT_TOKENS
     }
   };
@@ -397344,7 +397366,7 @@ function createForgeAgent({
     instructions: isProxyClaude ? undefined : {
       role: "system",
       content: buildInstructions(contextManager, modelId),
-      providerOptions: EPHEMERAL_CACHE
+      providerOptions: cacheOpts
     },
     callOptionsSchema: exports_external.object({
       userMessage: exports_external.string().nullable()
@@ -397356,7 +397378,7 @@ function createForgeAgent({
         ...activeTools ? { activeTools } : {}
       };
     },
-    prepareStep: buildForgePrepareStep(forgeMode === "plan", drainSteering, contextManager, tabId, canUseCodeExecution, parentMessagesRef, isProxyClaude ? buildInstructions(contextManager, modelId) : undefined),
+    prepareStep: buildForgePrepareStep(forgeMode === "plan", drainSteering, contextManager, tabId, canUseCodeExecution, parentMessagesRef, isProxyClaude ? buildInstructions(contextManager, modelId) : undefined, cacheOpts),
     experimental_repairToolCall: repairToolCall,
     providerOptions: wrappedProviderOptions,
     ...subagentHeaders ? { headers: subagentHeaders } : {}
@@ -413405,6 +413427,20 @@ Project commands: ${parts2.join(" \xB7 ")}` : "";
     hasSoulMapDiff() {
       return this.soulMapDiffChangedFiles.size > 0 && this.isRepoMapReady();
     }
+    lastTurnAt = 0;
+    maybeReseedExpiredCache(idleMs = 270000) {
+      const now2 = Date.now();
+      const prev = this.lastTurnAt;
+      this.lastTurnAt = now2;
+      if (prev === 0)
+        return false;
+      if (now2 - prev < idleMs)
+        return false;
+      if (this.soulMapDiffChangedFiles.size === 0)
+        return false;
+      this.resetForCompaction();
+      return true;
+    }
   };
 });
 
@@ -496419,6 +496455,11 @@ Proceeding without it will significantly reduce capabilities \u2014 no soul tool
           }
         }
         setLoadingStartedAt(Date.now());
+        {
+          const cacheTtl = effectiveConfig2.cache?.ttl ?? "5m";
+          const idleMs = cacheTtl === "1h" ? 3540000 : 270000;
+          contextManager.maybeReseedExpiredCache(idleMs);
+        }
         const agent2 = createForgeAgent({
           model,
           fullModelId: modelId,
@@ -526331,6 +526372,8 @@ function readValuesFromLayer(layer) {
     v4.pruning = layer.contextManagement.pruningTarget;
   else if (layer.contextManagement?.disablePruning !== undefined)
     v4.pruning = layer.contextManagement.disablePruning ? "none" : "subagents";
+  if (layer.cache?.ttl !== undefined)
+    v4.cacheTtl = layer.cache.ttl;
   return v4;
 }
 function effectiveValues(global2, project2) {
@@ -526406,6 +526449,8 @@ function buildPatch(key3, value) {
       return {
         contextManagement: { pruningTarget: value }
       };
+    case "cacheTtl":
+      return { cache: { ttl: value } };
     default:
       return {};
   }
@@ -526882,6 +526927,14 @@ var init_ProviderSettings = __esm(async () => {
       label: "Clear old tool uses",
       desc: "Drop old tool results at 65% ctx. Busts prompt cache when it fires",
       type: "toggle"
+    },
+    { type: "section", label: "Cache" },
+    {
+      key: "cacheTtl",
+      label: "Cache TTL",
+      desc: "5m = free \xB7 1h = ~2\xD7 cache-write cost, 12\xD7 lifetime (long agentic runs)",
+      type: "cycle",
+      options: ["5m", "1h"]
     }
   ];
   OPENAI_ITEMS = [
@@ -527097,7 +527150,8 @@ var init_ProviderSettings = __esm(async () => {
     groqReasoningEffort: "off",
     openaiReasoningSummary: "off",
     openaiVerbosity: "off",
-    groqReasoningFormat: "off"
+    groqReasoningFormat: "off",
+    cacheTtl: "5m"
   };
 });
 

package/dist/workers/io.worker.js

@@ -50577,7 +50577,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@proxysoul/soulforge",
-    version: "2.16.0",
+    version: "2.16.1",
     description: "Graph-powered code intelligence \u2014 multi-agent coding with codebase-aware AI",
     repository: {
       type: "git",
@@ -81645,7 +81645,20 @@ function supportsTemperature(modelId) {
     return true;
   return v.major < 5 && (v.major < 4 || v.minor < 7);
 }
-var NO_SUPPORT, ANTHROPIC_FULL, OPENAI_FULL, GOOGLE_FULL, XAI_FULL, DEEPSEEK_FULL, OPENROUTER_FULL, GATEWAY_FULL, COMPAT_ONLY, PROVIDER_CONSTRAINTS;
+function buildCacheProviderOptions(ttl) {
+  const cache = ttl === "1h" ? CACHE_EPHEMERAL_1H : CACHE_EPHEMERAL_5M;
+  return {
+    anthropic: cache,
+    google: cache,
+    proxy: cache,
+    llmgateway: cache,
+    opencode_zen: cache,
+    opencode_go: cache,
+    openrouter: cache,
+    vercel_gateway: cache
+  };
+}
+var NO_SUPPORT, ANTHROPIC_FULL, OPENAI_FULL, GOOGLE_FULL, XAI_FULL, DEEPSEEK_FULL, OPENROUTER_FULL, GATEWAY_FULL, COMPAT_ONLY, PROVIDER_CONSTRAINTS, CACHE_EPHEMERAL_5M, CACHE_EPHEMERAL_1H, EPHEMERAL_CACHE_5M, EPHEMERAL_CACHE_1H;
 var init_provider_options = __esm(() => {
   init_models();
   init_providers();
@@ -81754,6 +81767,20 @@ var init_provider_options = __esm(() => {
       interleavedThinking: false
     }
   };
+  CACHE_EPHEMERAL_5M = {
+    cacheControl: {
+      type: "ephemeral",
+      ttl: "5m"
+    }
+  };
+  CACHE_EPHEMERAL_1H = {
+    cacheControl: {
+      type: "ephemeral",
+      ttl: "1h"
+    }
+  };
+  EPHEMERAL_CACHE_5M = buildCacheProviderOptions("5m");
+  EPHEMERAL_CACHE_1H = buildCacheProviderOptions("1h");
 });
 
 // src/core/compaction/summarize.ts

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@proxysoul/soulforge",
-  "version": "2.16.0",
+  "version": "2.16.1",
   "description": "Graph-powered code intelligence — multi-agent coding with codebase-aware AI",
   "repository": {
     "type": "git",