mirror of
https://github.com/catlog22/Claude-Code-Workflow.git
synced 2026-02-04 01:40:45 +08:00
- Introduced merge-plans-with-file prompt for aggregating multiple planning artifacts, resolving conflicts, and synthesizing a unified execution plan. - Implemented detailed execution process including discovery, normalization, conflict detection, resolution, and synthesis phases. - Added quick-plan-with-file prompt for rapid planning with minimal documentation, utilizing multi-agent analysis for actionable outputs. - Both prompts support various input formats and provide structured outputs for execution and documentation.
14 KiB
14 KiB
description, argument-hint
| description | argument-hint |
|---|---|
| Multi-agent rapid planning with minimal documentation, conflict resolution, and actionable synthesis. Lightweight planning from raw task, brainstorm, or analysis artifacts | TOPIC="<planning topic or task>" [--from=brainstorm|analysis|task|raw] [--perspectives=arch,impl,risk,decision] [--auto] [--verbose] |
Codex Quick-Plan-With-File Prompt
Overview
Multi-agent rapid planning workflow with minimal documentation overhead. Coordinates parallel agent analysis (architecture, implementation, validation, decision), synthesizes conflicting perspectives into actionable decisions, and generates an implementation-ready plan.
Core workflow: Parse Input → Parallel Analysis → Conflict Resolution → Plan Synthesis → Output
Key features:
- Format Agnostic: Consumes brainstorm conclusions, analysis recommendations, quick tasks, or raw descriptions
- Minimal Docs: Single plan.md (no lengthy timeline documentation)
- Parallel Multi-Agent: 4 concurrent perspectives for rapid analysis
- Conflict Resolution: Automatic conflict detection and synthesis
- Actionable Output: Direct task breakdown ready for execution
Target Planning
$TOPIC
--from: Input source type (brainstorm | analysis | task | raw) - auto-detected if omitted--perspectives: Which perspectives to use (arch, impl, risk, decision) - all by default--auto: Auto-confirm decisions, minimal user prompts--verbose: Verbose output with all reasoning
Execution Process
Phase 1: Input Validation & Loading
├─ Parse input: topic | artifact reference
├─ Load artifact if referenced (synthesis.json | conclusions.json)
├─ Extract constraints and key requirements
└─ Initialize session folder
Phase 2: Parallel Multi-Agent Analysis (concurrent)
├─ Agent 1 (Architecture): Design decomposition, patterns, scalability
├─ Agent 2 (Implementation): Tech stack, feasibility, effort estimates
├─ Agent 3 (Validation): Risk matrix, testing strategy, monitoring
├─ Agent 4 (Decision): Recommendations, tradeoffs, execution strategy
└─ Aggregate findings into perspectives.json
Phase 3: Conflict Detection & Resolution
├─ Detect: effort conflicts, architecture conflicts, risk conflicts
├─ Analyze rationale for each conflict
├─ Synthesis via arbitration: generate unified recommendation
├─ Document conflicts and resolutions
└─ Update plan.md
Phase 4: Plan Synthesis
├─ Consolidate all insights
├─ Generate task breakdown (5-8 major tasks)
├─ Create execution strategy and dependencies
├─ Document assumptions and risks
└─ Output plan.md + synthesis.json
Output:
├─ .workflow/.planning/{sessionId}/plan.md (minimal, actionable)
├─ .workflow/.planning/{sessionId}/perspectives.json (agent findings)
├─ .workflow/.planning/{sessionId}/conflicts.json (decision points)
└─ .workflow/.planning/{sessionId}/synthesis.json (task breakdown for execution)
Implementation Details
Phase 1: Session Setup & Input Loading
const getUtc8ISOString = () => new Date(Date.now() + 8 * 60 * 60 * 1000).toISOString()
// Parse input
const planSlug = "$TOPIC".toLowerCase()
.replace(/[^a-z0-9\u4e00-\u9fa5]+/g, '-')
.substring(0, 30)
const sessionId = `PLAN-${planSlug}-${getUtc8ISOString().substring(0, 10)}`
const sessionFolder = `.workflow/.planning/${sessionId}`
// Detect input type
let artifact = null
if ($TOPIC.startsWith('BS-') || "$TOPIC".includes('brainstorm')) {
artifact = loadBrainstormArtifact($TOPIC)
} else if ($TOPIC.startsWith('ANL-') || "$TOPIC".includes('analysis')) {
artifact = loadAnalysisArtifact($TOPIC)
}
bash(`mkdir -p ${sessionFolder}`)
Phase 2: Parallel Multi-Agent Analysis
Run 4 agents in parallel using ccw cli:
// Launch all 4 agents concurrently with Bash run_in_background
const agentPromises = []
// Agent 1 - Architecture (Gemini)
agentPromises.push(
Bash({
command: `ccw cli -p "
PURPOSE: Architecture & high-level design for '${planningTopic}'
Success: Clear component decomposition and architectural approach
TASK:
• Decompose problem into major components/modules
• Identify architectural patterns and integration points
• Design component interfaces and data models
• Assess scalability and maintainability implications
• Propose architectural approach with rationale
MODE: analysis
CONTEXT: @**/*
${artifact ? \`| Source artifact: \${artifact.type}\` : ''}
EXPECTED:
- Component decomposition (list with responsibilities)
- Module interfaces and contracts
- Data flow between components
- Architectural patterns applied (e.g., MVC, Event-Driven, etc.)
- Scalability assessment (1-5 rating with rationale)
- Architectural risks identified
CONSTRAINTS: Focus on long-term maintainability and extensibility
" --tool gemini --mode analysis`,
run_in_background: true
})
)
// Agent 2 - Implementation (Codex)
agentPromises.push(
Bash({
command: \`ccw cli -p "
PURPOSE: Implementation approach & technical feasibility for '\${planningTopic}'
Success: Concrete implementation strategy with realistic estimates
TASK:
• Evaluate technical feasibility of proposed approach
• Identify required technologies and dependencies
• Estimate effort: analysis/design/coding/testing/deployment
• Suggest implementation phases and milestones
• Highlight technical blockers and challenges
MODE: analysis
CONTEXT: @**/*
\${artifact ? \`| Source artifact: \${artifact.type}\` : ''}
EXPECTED:
- Technology stack recommendation (languages, frameworks, tools)
- Implementation complexity: high|medium|low (with justification)
- Effort breakdown (hours or complexity: analysis, design, coding, testing, deployment)
- Key technical decisions with tradeoffs explained
- Potential blockers and mitigation strategies
- Suggested implementation phases with sequencing
- Reusable components or libraries identified
CONSTRAINTS: Realistic with current tech stack
" --tool codex --mode analysis\`,
run_in_background: true
})
)
// Agent 3 - Validation & Risk (Claude)
agentPromises.push(
Bash({
command: \`ccw cli -p "
PURPOSE: Risk analysis and validation strategy for '\${planningTopic}'
Success: Comprehensive risk matrix with testing and deployment strategy
TASK:
• Identify technical risks and failure scenarios
• Assess timeline and resource risks
• Define validation/testing strategy (unit, integration, e2e, performance)
• Suggest monitoring and observability requirements
• Propose deployment strategy and rollback plan
MODE: analysis
CONTEXT: @**/*
\${artifact ? \`| Source artifact: \${artifact.type}\` : ''}
EXPECTED:
- Risk matrix (likelihood × impact, each 1-5)
- Top 3 technical risks with mitigation approaches
- Top 3 timeline/resource risks with mitigation
- Testing strategy (what to test, how, when, acceptance criteria)
- Deployment strategy (staged rollout, blue-green, canary, etc.)
- Rollback plan and recovery procedures
- Monitoring/observability requirements (metrics, logs, alerts)
- Overall risk rating: low|medium|high (with confidence)
CONSTRAINTS: Be realistic, not pessimistic
" --tool claude --mode analysis\`,
run_in_background: true
})
)
// Agent 4 - Strategic Decision (Gemini)
agentPromises.push(
Bash({
command: \`ccw cli -p "
PURPOSE: Strategic decisions and execution recommendations for '\${planningTopic}'
Success: Clear recommended approach with tradeoff analysis
TASK:
• Synthesize all perspectives into strategic recommendations
• Identify 2-3 critical decision points with recommended choices
• Clearly outline key tradeoffs (speed vs quality, scope vs timeline, risk vs cost)
• Propose go/no-go decision criteria and success metrics
• Suggest execution strategy and resource sequencing
MODE: analysis
CONTEXT: @**/*
\${artifact ? \`| Source artifact: \${artifact.type}\` : ''}
EXPECTED:
- Primary recommendation with strong rationale (1-2 paragraphs)
- Alternative approaches with pros/cons (2-3 alternatives)
- 2-3 critical decision points:
- What decision needs to be made
- Trade-offs for each option
- Recommended choice and why
- Key trade-offs explained (what we're optimizing for: speed/quality/risk/cost)
- Success metrics and go/no-go criteria
- Resource requirements and critical path items
- Suggested execution sequencing and phases
CONSTRAINTS: Focus on actionable decisions, provide clear rationale
" --tool gemini --mode analysis\`,
run_in_background: true
})
)
// Wait for all agents to complete
const [archResult, implResult, riskResult, decisionResult] = await Promise.all(agentPromises)
// Parse and extract findings from each agent result
const architecture = parseArchitectureResult(archResult)
const implementation = parseImplementationResult(implResult)
const validation = parseValidationResult(riskResult)
const recommendation = parseDecisionResult(decisionResult)
Agent Focus Areas:
| Agent | Perspective | Focus Areas |
|---|---|---|
| Gemini (Design) | Architecture patterns | Components, interfaces, scalability, patterns |
| Codex (Build) | Implementation reality | Tech stack, complexity, effort, blockers |
| Claude (Validate) | Risk & testing | Risk matrix, testing strategy, deployment, monitoring |
| Gemini (Decide) | Strategic synthesis | Recommendations, trade-offs, critical decisions |
Phase 3: Parse & Aggregate Perspectives
const perspectives = {
session_id: sessionId,
topic: "$TOPIC",
timestamp: getUtc8ISOString(),
architecture: {
components: [...],
patterns: [...],
scalability_rating: 3,
risks: [...]
},
implementation: {
technology_stack: [...],
complexity: "medium",
effort_breakdown: { analysis: 2, design: 3, coding: 8, testing: 4 },
blockers: [...]
},
validation: {
risk_matrix: [...],
top_risks: [{ title, impact, mitigation }, ...],
testing_strategy: "...",
monitoring: [...]
},
recommendation: {
primary_approach: "...",
alternatives: [...],
critical_decisions: [...],
tradeoffs: [...]
}
}
Write(`${sessionFolder}/perspectives.json`, JSON.stringify(perspectives, null, 2))
Phase 4: Conflict Detection
Detect conflicts:
- Effort variance: Are estimates consistent?
- Risk disagreement: Do arch and validation agree on risks?
- Scope confusion: Are recommendations aligned?
- Architecture mismatch: Do design and implementation agree?
For each conflict: document it, then run synthesis arbitration.
Phase 5: Generate Plan
# Quick Planning Session
**Session ID**: ${sessionId}
**Topic**: $TOPIC
**Created**: ${timestamp}
---
## Executive Summary
${synthesis.executive_summary}
**Complexity**: ${synthesis.complexity_level}
**Estimated Effort**: ${formatEffort(synthesis.effort_breakdown)}
**Optimization Focus**: ${synthesis.optimization_focus}
---
## Architecture
**Primary Pattern**: ${synthesis.architecture_approach}
**Key Components**:
${synthesis.key_components.map((c, i) => `${i+1}. ${c.name}: ${c.responsibility}`).join('\n')}
---
## Implementation Strategy
**Technology Stack**:
${synthesis.technology_stack.map(t => `- ${t}`).join('\n')}
**Phases**:
${synthesis.phases.map((p, i) => `${i+1}. ${p.name} (${p.effort})`).join('\n')}
---
## Risk Assessment
**Overall Risk**: ${synthesis.overall_risk_level}
**Top 3 Risks**:
${synthesis.top_risks.map((r, i) => `${i+1}. **${r.title}** (Impact: ${r.impact})\n Mitigation: ${r.mitigation}`).join('\n\n')}
---
## Task Breakdown (Ready for Execution)
${synthesis.tasks.map((task, i) => `
${i+1}. **${task.id}: ${task.title}** (Effort: ${task.effort})
${task.description}
Dependencies: ${task.dependencies.join(', ') || 'none'}
`).join('\n')}
---
## Next Steps
**Execute with**:
\`\`\`
/workflow:unified-execute-with-file -p ${sessionFolder}/synthesis.json
\`\`\`
**Detailed planning if needed**:
\`\`\`
/workflow:plan "Based on: $TOPIC"
\`\`\`
Session Folder Structure
.workflow/.planning/{sessionId}/
├── plan.md # Minimal, actionable
├── perspectives.json # Agent findings
├── conflicts.json # Conflicts & resolutions (if any)
└── synthesis.json # Task breakdown for execution
Multi-Agent Coordination
| Agent | Perspective | Tools | Output |
|---|---|---|---|
| Gemini (Design) | Architecture patterns | Design thinking, cross-domain | Components, patterns, scalability |
| Codex (Build) | Implementation reality | Tech stack evaluation | Stack, effort, feasibility |
| Claude (Validate) | Risk & testing | Risk assessment, QA | Risks, testing strategy |
| Gemini (Decide) | Strategic synthesis | Decision analysis | Recommendations, tradeoffs |
Error Handling
| Situation | Action |
|---|---|
| Agents conflict | Arbitration agent synthesizes recommendation |
| Missing blockers | Continue with available context, note gaps |
| Unclear input | Ask for clarification on planning focus |
| Estimate too high | Suggest MVP approach or phasing |
Integration Flow
Raw Task / Brainstorm / Analysis
│
▼
quick-plan-with-file (5-10 min)
│
├─ plan.md
├─ perspectives.json
└─ synthesis.json
│
▼
unified-execute-with-file
│
▼
Implementation
Usage Patterns
Pattern 1: Quick planning from task
TOPIC="实现实时通知系统" --auto
→ Creates actionable plan in ~5 minutes
Pattern 2: Convert brainstorm to execution plan
TOPIC="BS-notifications-2025-01-28" --from=brainstorm
→ Reads synthesis.json from brainstorm
→ Generates implementation plan
Pattern 3: From analysis to plan
TOPIC="ANL-auth-2025-01-28" --from=analysis
→ Converts conclusions.json to executable plan
Now execute quick-plan-with-file for topic: $TOPIC