@benkhz/context-manager 2.0.0 → 2.0.1

package/README.md

@@ -223,6 +223,11 @@ The manager tracks two parallel message lists: the full **history** (everything
 received, exposed via `getMessages()`) and the **active window** (`getActiveMessages()`) — the
 slice actually sent to the LLM, which compaction and truncation shrink. History is never pruned.
 
+This check runs at the start of every `send()` call, and also between tool-call iterations
+*within* a single turn — a request that triggers several tool calls in a row can grow the active
+window past `contextLimit` well before the turn finishes, so compaction can kick in mid-turn
+rather than waiting for the next `send()`.
+
 When the character count of the active window exceeds `contextLimit`:
 
 1. `onContextLimit` hook is called — returns `'compact'` (default), `'truncate'`, or `'error'`

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@benkhz/context-manager",
-  "version": "2.0.0",
+  "version": "2.0.1",
   "description": "Provider-agnostic LLM context manager with tool execution, auto-compaction, reactive state, and an event bus.",
   "license": "MIT",
   "type": "module",

package/src/AIContextManager.js

@@ -73,13 +73,7 @@ export class AIContextManager {
     }
 
     // Auto-compact when approaching the context limit
-    const chars = this._charCount()
-    if (chars > this._config.contextLimit) {
-      const policy = this._hooks.onContextLimit?.(chars, this._config.contextLimit) ?? 'compact'
-      if      (policy === 'compact')   await this.compact()
-      else if (policy === 'truncate')  this._truncate()
-      else throw new Error(`AIContextManager: context limit exceeded (${chars} chars)`)
-    }
+    await this._enforceContextLimit()
 
     try {
       const assistantMsg = await this._runLoop(opts.system)
@@ -256,9 +250,24 @@ export class AIContextManager {
       this._pushMessage({ role: 'tool', content: JSON.stringify(result), toolCallId: tc.id })
     }
 
+    // Re-check the context limit between tool-call iterations, not just at the
+    // top of send() — a single turn can run several iterations and blow past
+    // the limit long before the next send() call ever re-evaluates it.
+    await this._enforceContextLimit()
+
     return this._runLoop(system, depth + 1)
   }
 
+  /** Check contextLimit and apply the configured policy (compact/truncate/error). */
+  async _enforceContextLimit() {
+    const chars = this._charCount()
+    if (chars <= this._config.contextLimit) return
+    const policy = this._hooks.onContextLimit?.(chars, this._config.contextLimit) ?? 'compact'
+    if      (policy === 'compact')   await this.compact()
+    else if (policy === 'truncate')  this._truncate()
+    else throw new Error(`AIContextManager: context limit exceeded (${chars} chars)`)
+  }
+
   /** Append a message to both the full history and the active LLM-facing window. */
   _pushMessage(msg) {
     this._history.push(msg)