universal-agent-memory 1.0.3 → 1.0.5

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "universal-agent-memory",
-  "version": "1.0.3",
+  "version": "1.0.5",
   "description": "Universal AI agent memory system - CLAUDE.md templates, memory, worktrees for Claude Code, Factory.AI, VSCode, OpenCode",
   "type": "module",
   "main": "dist/index.js",

package/templates/CLAUDE.template.md

@@ -1,15 +1,27 @@
 <!--
-  CLAUDE.md Universal Template - v10.3
+  CLAUDE.md Universal Template - v10.5
   
   CHANGES IN THIS VERSION:
+  - STRENGTHENED Pattern Router: Now requires explicit analysis block output
+  - STRENGTHENED Constraint Extraction: Mandatory checklist with verification
+  - STRENGTHENED Adversarial Thinking: Explicit attack vector enumeration
+  - All pattern activations now require printed confirmation blocks
+  - Pattern Router, Constraint Checklist, and Adversarial Analysis are MANDATORY outputs
+  
+  PREVIOUS (v10.4):
+  - Added MANDATORY COMPLETION GATES section (3 gates must pass before "done")
+  - Gate 1: Output Existence Check (enforces P12)
+  - Gate 2: Constraint Compliance Check (enforces P17)
+  - Gate 3: Test Verification (enforces P13)
+  - Added PATTERN ROUTER as Critical Reminder #0 (auto-selects patterns)
+  
+  PREVIOUS (v10.3):
   - Added 5 new generic patterns (16-20) from deep failure analysis
   - Pattern 16: Task-First Execution (TFE) - prevents analysis without output
   - Pattern 17: Constraint Extraction (CE) - catches format/structure requirements
   - Pattern 18: Multi-Tool Pipeline (MTP) - chains tools for complex tasks
   - Pattern 19: Enhanced Impossible Task Refusal (ITR+) - refuses impossible immediately
   - Pattern 20: Adversarial Thinking (AT) - attack mindset for bypass tasks
-  - Enhanced CRITICAL REMINDERS with 5 new items (12-16)
-  - Total patterns now: 20 (up from 15)
   
   PREVIOUS (v10.2):
   - Added 4 new generic patterns (12-15) from Terminal-Bench 2.0 analysis
@@ -1020,10 +1032,72 @@ gh run view <run-id>
 
 ---
 
+## 🚦 COMPLETION GATES - MANDATORY
+
+**CANNOT say "done" or "complete" until ALL gates pass. This is NOT optional.**
+
+### GATE 1: Output Existence Check
+```bash
+# List ALL expected outputs from task description
+# Verify EACH file exists
+echo "=== GATE 1: OUTPUT EXISTENCE ==="
+for f in /app/expected_output.* /app/result.* /app/*.json; do
+  [ -f "$f" ] && echo "✓ $f exists" || echo "✗ MISSING: $f"
+done
+# If ANY missing: CREATE IT NOW, then re-check
+```
+
+### GATE 2: Constraint Compliance Check
+```
+Extract ALL constraints from task description:
+☐ File count constraint (e.g., "single file")
+☐ Line count constraint (e.g., "exactly one line")  
+☐ Format constraint (e.g., "valid JSON")
+☐ Size constraint (e.g., "under 5KB")
+
+Verify EACH constraint is satisfied before proceeding.
+```
+
+### GATE 3: Test Verification
+```bash
+echo "=== GATE 3: TEST VERIFICATION ==="
+# Run available tests
+pytest /tests/ -v 2>&1 | tail -30
+# If < 100% pass: iterate (fix specific failure, re-run)
+# Reserve 20% of time for iteration
+```
+
+**HARD STOP RULE:**
+- If Gate 1 fails → CREATE missing files immediately
+- If Gate 2 fails → FIX constraint violations immediately  
+- If Gate 3 fails → ITERATE until 100% or timeout
+- NEVER report "complete" with failing gates
+
+---
+
 ## ⚠️ CRITICAL REMINDERS (READ LAST)
 
 **These reminders exploit recency bias - place at END for maximum compliance.**
 
+0. **PATTERN ROUTER (MANDATORY FIRST STEP)**: 
+   BEFORE ANY work, you MUST print this analysis block:
+   ```
+   === PATTERN ROUTER ===
+   Task: [one-line task summary]
+   Classification: [file-creation | constraint-validation | adversarial | pipeline | impossible]
+   
+   SELECTED PATTERNS:
+   - P12 (OEV): [YES/NO] - Does task require creating files?
+   - P17 (CE): [YES/NO] - Does task have "exactly/only/single/must be" constraints?
+   - P20 (AT): [YES/NO] - Does task involve "bypass/break/exploit/filter"?
+   - P11 (Pre-computed): [YES/NO] - Is this a known algorithm (crypto/chess/ML)?
+   - P18 (MTP): [YES/NO] - Does task need multiple tools chained?
+   
+   ACTIVE PATTERNS: [list only YES patterns]
+   === END ROUTER ===
+   ```
+   DO NOT proceed without printing this block first.
+
 1. **BACKUP FIRST**: Before modifying ANY config, database, or critical file:
    ```bash
    cp <file> <file>.bak
@@ -1069,8 +1143,20 @@ gh run view <run-id>
 12. **TASK-FIRST**: Create required output files BEFORE analyzing.
     Touch placeholders immediately, then refine with real content.
 
-13. **EXTRACT CONSTRAINTS**: Before implementing, list ALL constraints from task:
-    Keywords: "exactly", "only", "single", "must be", formats, limits.
+13. **EXTRACT CONSTRAINTS (MANDATORY)**: If P17 is active, you MUST print:
+    ```
+    === CONSTRAINT CHECKLIST ===
+    Extracted from task description:
+    ☐ [constraint 1, e.g., "single file only"]
+    ☐ [constraint 2, e.g., "exactly one line output"]
+    ☐ [constraint 3, e.g., "format: valid JSON"]
+    
+    BEFORE completing, verify EACH constraint:
+    ☑ constraint 1: [PASS/FAIL]
+    ☑ constraint 2: [PASS/FAIL]
+    === END CONSTRAINTS ===
+    ```
+    If ANY constraint FAILS, FIX before completing.
 
 14. **MULTI-TOOL PIPELINE**: For complex tasks, identify ALL tools needed.
     Don't implement from scratch if existing tools exist.
@@ -1078,5 +1164,22 @@ gh run view <run-id>
 15. **REFUSE IMPOSSIBLE**: If task violates information theory (compress 500MB to 5KB),
     REFUSE immediately. Create IMPOSSIBLE.md explaining why. Don't attempt.
 
-16. **ADVERSARIAL MINDSET**: For bypass/exploit tasks, think like an attacker.
-    Enumerate attack vectors (encoding, null bytes, case variation). Test each.
+16. **ADVERSARIAL MINDSET (MANDATORY)**: If P20 is active, you MUST:
+    ```
+    === ADVERSARIAL ANALYSIS ===
+    Target: [what are we trying to bypass/break?]
+    Defense mechanism: [how does the filter/protection work?]
+    
+    ATTACK VECTORS TO TRY:
+    1. Case variation: <ScRiPt>, <SCRIPT>
+    2. Encoding: &#x3C;script&#x3E;, %3Cscript%3E
+    3. Null bytes: <scr\x00ipt>
+    4. Double encoding: %253Cscript%253E
+    5. Context breaking: </style><script>
+    6. Event handlers: <img onerror=alert(1)>
+    7. [add task-specific vectors]
+    
+    TEST EACH vector until one works.
+    === END ADVERSARIAL ===
+    ```
+    DO NOT use the filter "correctly" - your goal is to BREAK it.