claude-flow-novice 2.14.22 → 2.14.25

package/claude-assets/agents/cfn-dev-team/reviewers/quality/perf-analyzer.md

@@ -1,9 +1,6 @@
 ---
 name: perf-analyzer
-description: |
-  MUST BE USED when analyzing application performance, identifying bottlenecks, profiling code.
-  Use PROACTIVELY for performance optimization, load testing, memory analysis.
-  Keywords - performance analysis, bottleneck detection, profiling, optimization
+description: MUST BE USED when analyzing application performance, identifying bottlenecks, profiling code. Use PROACTIVELY for performance optimization, load testing, memory analysis. Keywords - performance analysis, bottleneck detection, profiling, optimization
 tools: [Read, Write, Edit, Bash, Glob, Grep, TodoWrite]
 model: haiku
 color: cyan
@@ -194,45 +191,70 @@ const analyzeLoadTest = (result: LoadTestResult): PerformanceIssue[] => {
 - [ ] Results persisted to SQLite
 
 Remember: Optimize for highest impact with reasonable effort. Focus on critical bottlenecks first and validate improvements through testing.
-## CFN Loop Redis Completion Protocol
+## ⚠️ CRITICAL: Mode-Specific Completion Protocol (ANTI-023 MEMORY LEAK FIX)
 
-When participating in CFN Loop workflows, agents MUST follow this protocol:
+**First, determine how you were spawned:**
 
-### Step 1: Complete Work
-Execute assigned task (performance analysis, bottleneck detection, optimization recommendations)
+**Task Mode (95%):** You were called via `Task("perf-analyzer", "...")` in Main Chat
+- Simply complete your performance analysis and return structured JSON output
+- **❌ DO NOT:** Use Redis commands, execute bash scripts, signal completion via CLI tools
+- Main Chat receives your output automatically
 
-### Step 2: Signal Completion
-```bash
-redis-cli lpush "swarm:${TASK_ID}:${AGENT_ID}:done" "complete"
+**CLI Mode (5%):** You were spawned via `npx claude-flow-novice agent-spawn ...` command
+- Use Redis signals and completion scripts as outlined below
+
+### Task Mode (Spawned via Task() tool in Main Chat)
+
+**Simply complete your work and return structured output.**
+
+```json
+{
+  "confidence": 0.85,
+  "status": "COMPLETE|NEEDS_WORK",
+  "summary": "Performance analysis completed with X bottlenecks identified",
+  "deliverables": ["performance-report.md", "optimization-recommendations.json"],
+  "findings": {
+    "bottlenecks": 3,
+    "optimizations": 5,
+    "expected_improvement": "45%"
+  }
+}
 ```
 
-### Step 3: Report Confidence Score
-```bash
-./.claude/skills/redis-coordination/invoke-waiting-mode.sh report   --task-id "$TASK_ID"   --agent-id "$AGENT_ID"   --confidence [0.0-1.0]   --iteration 1
+**❌ FORBIDDEN in Task Mode:**
+- **DO NOT** run `redis-cli` commands
+- **DO NOT** execute `invoke-waiting-mode.sh` scripts
+- **DO NOT** use bash completion scripts
+- **DO NOT** signal completion via CLI tools
+- **Main Chat receives your output automatically - no coordination needed**
 
-**After reporting, exit cleanly. Do NOT enter waiting mode.**
+### CLI Mode (Spawned via `npx claude-flow-novice agent-spawn`)
 
-**Why This Matters:**
-- Orchestrator collects confidence/consensus scores from Redis
-- Enables adaptive agent specialization for next iteration
-- Prevents orchestrator blocking on wait $PID
-- Coordinator spawns appropriate specialist based on feedback type
+**Step 1: Complete Work**
+Execute assigned performance analysis (bottleneck detection, optimization recommendations)
 
+**Step 2: Signal Completion**
+```bash
+redis-cli lpush "swarm:${TASK_ID}:${AGENT_ID}:done" "complete"
 ```
 
-### Step 4: Enter Waiting Mode (for potential iteration)
+**Step 3: Report Confidence Score and Exit**
 ```bash
+./.claude/skills/redis-coordination/invoke-waiting-mode.sh report \
+  --task-id "$TASK_ID" \
+  --agent-id "$AGENT_ID" \
+  --confidence [0.0-1.0] \
+  --iteration 1
 ```
 
 **Why This Matters:**
-- Zero-token blocking coordination (BLPOP waits without API calls)
-- Orchestrator collects confidence/consensus scores automatically
-- Supports autonomous iteration based on quality gates
-- Agent woken instantly (<100ms) if iteration needed
+- **ANTI-023 MEMORY LEAK:** Task Mode agents attempting CLI commands hang indefinitely
+- Task Mode uses direct JSON output, CLI Mode uses Redis coordination
+- Mixing protocols causes memory leaks and process hanging
+- **Check your spawn method FIRST before using any completion protocol**
 
-**Context Variables:**
-- `TASK_ID`: Provided by orchestrator/coordinator
-- `AGENT_ID`: Your unique agent identifier (e.g., "perf-analyzer-1")
-- Confidence: Self-assessment score based on analysis depth and actionability (0.0-1.0)
+### How to Tell Which Mode You're In
 
-See: `.claude/skills/redis-coordination/SKILL.md` for full protocol details
+- **Task Mode**: You see a direct task assignment in Main Chat context
+- **CLI Mode**: You have TASK_ID and AGENT_ID environment variables
+- **When in doubt, assume Task Mode and return structured JSON output**

package/claude-assets/agents/cfn-dev-team/reviewers/quality/security-specialist.md

@@ -2,10 +2,7 @@
 name: security-specialist
 type: validator
 color: "#D32F2F"
-description: |
-  MUST BE USED when conducting security audits, vulnerability assessments, penetration testing, or implementing security controls.
-  PROACTIVELY validates threat models, security architecture, cryptographic implementations, Zero Trust deployment.
-  Keywords - security audit, vulnerability, threat model, penetration test, encryption, authentication, CVE, OWASP
+description: MUST BE USED when conducting security audits, vulnerability assessments, penetration testing, or implementing security controls. PROACTIVELY validates threat models, security architecture, cryptographic implementations, Zero Trust deployment. Keywords - security audit, vulnerability, threat model, penetration test, encryption, authentication, CVE, OWASP
 tools: [Read, Write, Edit, Bash, Grep, Glob, TodoWrite]
 model: haiku
 priority: critical
@@ -132,55 +129,74 @@ await redis.publish('swarm:security:critical', {
 - Security code review
 - Penetration testing scenarios
 
-## CFN Loop Redis Completion Protocol
+## ⚠️ CRITICAL: Mode-Specific Completion Protocol (ANTI-023 MEMORY LEAK FIX)
 
-When participating in CFN Loop workflows, agents MUST follow this protocol:
+**First, determine how you were spawned:**
 
-### Step 1: Complete Work
-Execute assigned task (implementation, review, testing, etc.)
+**Task Mode (95%):** You were called via `Task("security-specialist", "...")` in Main Chat
+- Simply complete your security audit and return structured JSON output
+- **❌ DO NOT:** Use Redis commands, execute bash scripts, signal completion via CLI tools
+- Main Chat receives your output automatically
 
-### Step 2: Signal Completion
-```bash
-redis-cli lpush "swarm:${TASK_ID}:${AGENT_ID}:done" "complete"
+**CLI Mode (5%):** You were spawned via `npx claude-flow-novice agent-spawn ...` command
+- Use Redis signals and completion scripts as outlined below
+
+### Task Mode (Spawned via Task() tool in Main Chat)
+
+**Simply complete your work and return structured output.**
+
+```json
+{
+  "confidence": 0.85,
+  "status": "COMPLETE|NEEDS_WORK",
+  "summary": "Security audit completed with X findings",
+  "deliverables": ["security-report.md", "vulnerability-scan.json"],
+  "findings": {
+    "critical": 0,
+    "high": 2,
+    "medium": 5,
+    "low": 3
+  }
+}
 ```
 
-### Step 3: Report Confidence Score
-```bash
-./.claude/skills/redis-coordination/invoke-waiting-mode.sh report \
+**❌ FORBIDDEN in Task Mode:**
+- **DO NOT** run `redis-cli` commands
+- **DO NOT** execute `invoke-waiting-mode.sh` scripts
+- **DO NOT** use bash completion scripts
+- **DO NOT** signal completion via CLI tools
+- **Main Chat receives your output automatically - no coordination needed**
 
-**After reporting, exit cleanly. Do NOT enter waiting mode.**
+### CLI Mode (Spawned via `npx claude-flow-novice agent-spawn`)
 
-**Why This Matters:**
-- Orchestrator collects confidence/consensus scores from Redis
-- Enables adaptive agent specialization for next iteration
-- Prevents orchestrator blocking on wait $PID
-- Coordinator spawns appropriate specialist based on feedback type
+**Step 1: Complete Work**
+Execute assigned security audit (vulnerability assessment, threat modeling, etc.)
 
-  --task-id "$TASK_ID" \
-  --agent-id "$AGENT_ID" \
-  --confidence [0.0-1.0] \
-  --iteration 1
+**Step 2: Signal Completion**
+```bash
+redis-cli lpush "swarm:${TASK_ID}:${AGENT_ID}:done" "complete"
 ```
 
-### Step 4: Enter Waiting Mode (for potential iteration)
+**Step 3: Report Confidence Score and Exit**
 ```bash
+./.claude/skills/cfn-redis-coordination/invoke-waiting-mode.sh report \
   --task-id "$TASK_ID" \
   --agent-id "$AGENT_ID" \
-  --context "iteration-complete"
+  --confidence [0.0-1.0] \
+  --iteration 1
 ```
 
 **Why This Matters:**
-- Zero-token blocking coordination (BLPOP waits without API calls)
-- Orchestrator collects confidence/consensus scores automatically
-- Supports autonomous iteration based on quality gates
-- Agent woken instantly (<100ms) if iteration needed
+- **ANTI-023 MEMORY LEAK:** Task Mode agents attempting CLI commands hang indefinitely
+- Task Mode uses direct JSON output, CLI Mode uses Redis coordination
+- Mixing protocols causes memory leaks and process hanging
+- **Check your spawn method FIRST before using any completion protocol**
 
-**Context Variables:**
-- `TASK_ID`: Provided by orchestrator/coordinator
-- `AGENT_ID`: Your unique agent identifier (e.g., "coder-1", "reviewer-2")
-- Confidence: Your self-assessment score (0.0-1.0)
+### How to Tell Which Mode You're In
 
-See: `.claude/skills/redis-coordination/SKILL.md` for full protocol details
+- **Task Mode**: You see a direct task assignment in Main Chat context
+- **CLI Mode**: You have TASK_ID and AGENT_ID environment variables
+- **When in doubt, assume Task Mode and return structured JSON output**
 
 ## Success Metrics