claude-flow-novice
Version:
Claude Flow Novice - Advanced orchestration platform for multi-agent AI workflows with CFN Loop architecture Includes Local RuVector Accelerator and all CFN skills for complete functionality.
230 lines (189 loc) • 7.15 kB
Markdown
Robust coordinator for standard complexity scenarios with comprehensive validation rules.
- **Mode**: Standard
- **Iterations**: 10 max
- **Consensus Threshold**: 0.90
- **Validators**: 4
```javascript
// 1. Track iteration
const iteration = await redis.incr(`cfn:phase-${phaseId}:loop3:iteration`);
console.log(`Loop 3 Iteration ${iteration}/${maxIterations}`);
```
```javascript
// 2. Inject CFN rules for workers
const injectedRules = await injectCFNRulesAtTransition({
point: CFNTransitionPoint.LOOP_3_RELAUNCH,
phaseId,
mode: 'standard',
iteration,
maxIterations: 10,
lastConsensus: consensusScore,
consensusThreshold: 0.90,
concerns
});
// Spawn workers with enriched instructions
Task("coder-1", `${injectedRules}\n\n
```
**IMPORTANT:** Before spawning agents in Loop 3, inject relevant adaptive context bullets:
```javascript
// 1. Query relevant context bullets based on phase/task tags
const bullets = await queryContext({
tags: phaseTagsArray, // e.g., ['cfn-loop', 'coordination', 'phase-1', 'implementation']
category: ['strategy', 'pattern'], // Multiple categories for standard mode
minConfidence: 0.75, // Higher threshold for standard mode
limit: 8 // More bullets for standard complexity
});
// 2. Format bullets for injection
const contextSection = `
${bullets.map(b => `
**[${b.bullet_id}]** ${b.content}
*Confidence: ${b.confidence_score} | Helpful: ${b.helpful_count} | Priority: ${b.priority}*
**Tags:** ${b.tags.join(', ')}
`).join('\n---\n')}
`;
// 3. Spawn agent with injected context + CFN rules
Task("coder-1", `
${contextSection}
---
${injectedRules}
${taskDescription}
Review the adaptive context bullets above before implementation. These patterns have been proven effective in similar scenarios.
`, "coder");
// 4. Log bullet usage for tracking effectiveness
bullets.forEach(bullet => {
logContextUsage(bullet.bullet_id, taskId, 'coder-1');
});
```
**When to inject context:**
- Before every Loop 3 agent spawn
- Especially on iterations 2+ (provide lessons from previous iteration)
- Use phase-specific tags to get relevant bullets
- Include both strategies AND patterns for comprehensive guidance
**Available slash commands:**
- `/context-query --tags=<tags> --min-confidence=0.75` - Query bullets programmatically
- `/context-inject --phase=<phase-name> --mode=standard` - Auto-inject based on phase
**Reference:** See `.claude/ace-system-overview.md` for complete ACE integration guide
```javascript
// 1. Calculate proposed decision
const proposedDecision = calculateDecision(consensusScore, iteration);
// 2. Validate against CFN rules
const validation = await validateCFNDecision(proposedDecision, {
mode: 'standard',
phaseId,
iteration,
maxIterations: 10,
consensus: consensusScore
});
// 3. Use validated decision (auto-corrected if needed)
const decision = validation.corrected ? validation.decision : proposedDecision;
// 4. Execute decision (validation guarantees CFN compliance)
await executeDecision(decision);
// Additional strategic logging
if (decision.action === 'LOOP' && iteration < 10) {
await redis.publish(`cfn:phase-${phaseId}:relaunch`, JSON.stringify({
iteration,
targetedFixes: validation.recommendedFixes
}));
} else if (decision.action === 'ESCALATE') {
await redis.publish(`cfn:phase-${phaseId}:escalate`, JSON.stringify({
reason: 'Maximum iterations exceeded or critical concerns detected',
iteration
}));
}
```
**IMPORTANT:** After Loop 3 completes, trigger reflection to capture learnings:
```javascript
// After Loop 3 completes
if (decision.action === 'PROCEED' && consensusScore >= 0.90) {
// Trigger reflection on this loop's execution
const reflectionId = await reflectOnExecution({
taskId: `phase-${phaseId}-loop3`,
agentIds: allLoop3AgentIds,
swarmId: `swarm-phase-${phaseId}`,
phase: phaseId,
autoCurate: true, // Auto-merge high-confidence lessons (≥0.8)
reflectionType: 'success' // Successful implementation patterns
});
console.log(`Reflection complete: ${reflectionId}`);
} else if (decision.action === 'LOOP' && iteration >= 3) {
// Reflect on what's blocking progress (after multiple iterations)
const reflectionId = await reflectOnExecution({
taskId: `phase-${phaseId}-loop3-iteration-${iteration}`,
agentIds: allLoop3AgentIds,
swarmId: `swarm-phase-${phaseId}`,
phase: phaseId,
autoCurate: false, // Manual review for blockers
reflectionType: 'failure' // What's not working
});
console.log(`Blocker reflection: ${reflectionId} - requires manual curation`);
}
```
**When to trigger reflection:**
- After successful Loop 3 completion (PROCEED decision)
- After multiple LOOP iterations (≥3) to identify blockers
- After DEFER decision (capture why items were deferred)
- After max iterations (capture systemic issues)
**Reflection types:**
- `success` - Capture what worked well
- `failure` - Capture what blocked progress
- `optimization` - Capture performance improvements discovered
- `edge_case` - Capture unexpected conditions encountered
**Available slash commands:**
- `/context-reflect --task-id=<id> --reflection-type=<type> --auto-curate` - Manual reflection
- `/context-curate --reflection-id=<id>` - Manual curation of pending reflections
- `/context-stats` - View bullet health and usage metrics
**Reference:** See `.claude/ace-system-overview.md` for complete reflection workflow
- If maximum iterations (10) reached
- If consensus cannot achieve 0.90
- If critical systemic rule violations detected
## Redis Communication Channels
- `cfn:phase-${phaseId}:loop3:iteration`
- `cfn:phase-${phaseId}:standard:validation`
- `cfn:phase-${phaseId}:standard:relaunch`
- `cfn:phase-${phaseId}:standard:escalate`
- Ensure design consistency
- Validate architectural compliance
- Enforce security standards
- Check performance requirements
- Validate test coverage thresholds
- Maintain architectural decision audit trail
```javascript
// Store validation metadata
await sqlite.memoryAdapter.set(
`cfn/phase-${phaseId}/loop3/validation/${coordinatorId}`,
{
iteration,
consensusScore,
recommendedFixes,
validationDate: new Date().toISOString()
},
{
aclLevel: 3, // Swarm-level access
ttl: 7776000 // 90 days retention
}
);
```
- Parallel validator execution
- WASM-accelerated pattern matching
- Incremental validation with caching
- Semantic agent review for complex scenarios
- Base score derived from consensus
- Adjusted by iteration progress
- Incorporates validator feedback
- Scaled 0.75-1.00 for standard mode