open-agents-ai 0.138.65 → 0.138.67

package/dist/index.js

@@ -22332,6 +22332,28 @@ Integrate this guidance into your current approach. Continue working on the task
             });
           }
           const turnTier = this.options.modelTier ?? "large";
+          if (turn === 1 && (turnTier === "small" || turnTier === "medium")) {
+            const goal = this._taskState.goal || "";
+            const wordCount = goal.split(/\s+/).length;
+            const hasMultipleActions = /\band\b.*\band\b|then.*then|also.*also/i.test(goal);
+            const hasMultipleFiles = /files?.*files?|\.ts.*\.ts|create.*write|modify.*create/i.test(goal);
+            const isComplex = wordCount > 40 || hasMultipleActions || hasMultipleFiles;
+            if (isComplex) {
+              messages.push({
+                role: "user",
+                content: `[TASK DECOMPOSITION \u2014 Complete these steps IN ORDER]
+
+This task has multiple parts. Do them one at a time:
+1. LOCATE: Find and read the relevant file(s). List what you found.
+2. UNDERSTAND: Describe the current behavior in one sentence.
+3. PLAN: State exactly what changes you'll make (file, function, change).
+4. IMPLEMENT: Make the changes one file at a time. Test after each file.
+5. COMPLETE: Call task_complete only after ALL parts are verified.
+
+Start with step 1 \u2014 read the relevant files NOW.`
+              });
+            }
+          }
           const turnBudget = turnTier === "small" ? 5 : turnTier === "medium" ? 8 : 0;
           if (turnBudget > 0 && turn > 0 && turn % turnBudget === 0) {
             const repetitionRatio = this.detectRepetition(toolCallLog);
@@ -23606,7 +23628,10 @@ ${trimmedNew}`;
             return `The path "${lastFilePath}" does not exist. Call list_directory(".") to find correct paths, then try again.`;
           }
           if (/FAIL|test.*fail|assert/i.test(lastError)) {
-            return `Tests are failing. Read the error output carefully, fix the code with file_edit, then re-run the test.`;
+            return `Tests are failing. Do these TWO steps in order:
+STEP 1: Write a 5-line script isolating JUST the failing case, run it with shell, and observe the actual output.
+STEP 2: Based on what you observed (not assumed), edit ONLY the specific line(s) causing the failure.
+Do NOT rewrite the entire function. Patch the specific fault, then re-run tests.`;
           }
           if (/permission denied/i.test(lastError)) {
             return `Permission denied on "${lastFilePath}". Try a different approach or check file permissions.`;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.138.65",
+  "version": "0.138.67",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",

package/prompts/agentic/system-small.md

@@ -32,3 +32,16 @@ Debugging — OBSERVE before reasoning:
 - For type questions: run `typeof x` or `x instanceof Date` in shell to check.
 - For refactoring: before changing code, write the existing behavior as a test. Run it. Then change code. Run test again. If it breaks, your refactor has a bug.
 - NEVER reason about 10+ lines of code in your head. Use shell to execute and observe instead.
+
+When a test fails — TWO-STEP debug:
+1. ISOLATE: Write a 5-line script reproducing JUST the failing case. Run it. Read the output.
+2. PATCH: Based on what you SAW (not guessed), edit ONLY the failing line(s). Re-run test.
+Do NOT rewrite whole functions. Patch the specific fault.
+
+Complex tasks — DECOMPOSE first:
+1. LOCATE: Find and read the relevant file(s).
+2. UNDERSTAND: Describe current behavior in one sentence.
+3. PLAN: State what changes you'll make (which file, which function, what change).
+4. IMPLEMENT: Make changes one at a time.
+5. TEST: Run tests after each change.
+6. COMPLETE: Call task_complete when ALL parts are done.