10 minutes before cache clears in free mode

Author: jahooma

agents/base2/base2.ts

@@ -284,22 +284,40 @@ ${PLACEHOLDER.GIT_CHANGES_PROMPT}
         noAskUser,
       }),
 
-    handleSteps: function* ({ params }) {
-      while (true) {
-        // Run context-pruner before each step
-        yield {
-          toolName: 'spawn_agent_inline',
-          input: {
-            agent_type: 'context-pruner',
-            params: params ?? {},
-          },
-          includeToolCall: false,
-        } as any
-
-        const { stepsComplete } = yield 'STEP'
-        if (stepsComplete) break
-      }
-    },
+    // handleSteps is serialized via .toString() and re-eval'd, so closure
+    // variables like `isFree` are not in scope at runtime. Pick the right
+    // literal-baked function here instead.
+    handleSteps: isFree
+      ? function* ({ params }) {
+          while (true) {
+            yield {
+              toolName: 'spawn_agent_inline',
+              input: {
+                agent_type: 'context-pruner',
+                params: { ...(params ?? {}), cacheExpiryMs: 10 * 60 * 1000 },
+              },
+              includeToolCall: false,
+            } as any
+
+            const { stepsComplete } = yield 'STEP'
+            if (stepsComplete) break
+          }
+        }
+      : function* ({ params }) {
+          while (true) {
+            yield {
+              toolName: 'spawn_agent_inline',
+              input: {
+                agent_type: 'context-pruner',
+                params: params ?? {},
+              },
+              includeToolCall: false,
+            } as any
+
+            const { stepsComplete } = yield 'STEP'
+            if (stepsComplete) break
+          }
+        },
   }
 }
 

agents/context-pruner.ts

@@ -31,6 +31,9 @@ const definition: AgentDefinition = {
         userBudget: {
           type: 'number',
         },
+        cacheExpiryMs: {
+          type: 'number',
+        },
       },
       required: [],
     },
@@ -74,8 +77,8 @@ const definition: AgentDefinition = {
     /** Fudge factor for token count threshold to trigger pruning earlier */
     const TOKEN_COUNT_FUDGE_FACTOR = 1_000
 
-    /** Prompt cache expiry time (Anthropic caches for 5 minutes) */
-    const CACHE_EXPIRY_MS = 5 * 60 * 1000
+    /** Prompt cache expiry time (Anthropic caches for 5 minutes by default) */
+    const CACHE_EXPIRY_MS: number = params?.cacheExpiryMs ?? 5 * 60 * 1000
 
     /** Header used in conversation summaries */
     const SUMMARY_HEADER =
@@ -328,6 +331,17 @@ const definition: AgentDefinition = {
       currentMessages.splice(lastSubagentSpawnIndex, 1)
     }
 
+    // Also remove the params USER_PROMPT if params were provided to this agent
+    // (this is the message like <user_message>{"cacheExpiryMs": 600000}</user_message>)
+    if (params && Object.keys(params).length > 0) {
+      const lastUserPromptIndex = currentMessages.findLastIndex((message) =>
+        message.tags?.includes('USER_PROMPT'),
+      )
+      if (lastUserPromptIndex !== -1) {
+        currentMessages.splice(lastUserPromptIndex, 1)
+      }
+    }
+
     // Check for prompt cache miss (>5 min gap before the USER_PROMPT message)
     // The USER_PROMPT is the actual user message; INSTRUCTIONS_PROMPT comes after it
     // We need to find the USER_PROMPT and check the gap between it and the last assistant message