Claude-Code-Workflow/.claude/commands/workflow/develop-with-file.md

name, description, argument-hint, allowed-tools

name	description	argument-hint	allowed-tools
develop-with-file	Multi-agent development workflow with documented progress, Gemini-guided planning, and incremental iteration support	"feature description or task file.md"	TodoWrite(*), Task(*), AskUserQuestion(*), Read(*), Grep(*), Glob(*), Bash(*), Edit(*), Write(*)

Workflow Develop-With-File Command (/workflow:develop-with-file)

Overview

Production-grade development workflow with 2-document state tracking. Combines multi-agent parallel exploration, Gemini-assisted planning/verification, and incremental iteration in a single coherent workflow.

Core workflow: Explore → Plan → Execute → Verify → Iterate

Key features:

• progress.md: Single source of truth for exploration, planning, execution timeline
• plan.json: Current execution plan (derived from progress.md)
• Multi-agent exploration: Parallel cli-explore-agents from multiple angles
• Multi-CLI execution: Support gemini/codex/agent per task
• Gemini-guided planning: Intelligent task decomposition and validation
• Gemini verification: Post-execution review and correction
• Incremental iteration: Resume from any phase, add new tasks dynamically

Usage

/workflow:develop-with-file [FLAGS] <TASK_DESCRIPTION>

# Flags
-e, --explore              Force exploration phase
--resume <session-id>      Resume existing session

# Arguments
<task-description>         Feature description or path to .md file (required)

Execution Process

Session Detection:
   ├─ Check if session exists (--resume or auto-detect)
   ├─ EXISTS → Continue mode (read progress.md for current phase)
   └─ NOT_FOUND → Explore mode

Explore Mode (Multi-Agent):
   ├─ Assess task complexity (Low/Medium/High)
   ├─ Launch 1-4 parallel cli-explore-agents (angle-based)
   ├─ Aggregate exploration results
   ├─ Document in progress.md → Exploration section
   └─ Transition → Plan mode

Plan Mode (Gemini-Guided):
   ├─ Read progress.md exploration findings
   ├─ Invoke Gemini for task decomposition
   ├─ Generate plan.json (2-7 tasks with dependencies)
   ├─ Assign executor per task (gemini/codex/agent)
   ├─ Document in progress.md → Planning section
   ├─ User confirmation (Allow/Modify/Cancel)
   └─ Transition → Execute mode

Execute Mode (Multi-Agent/CLI):
   ├─ For each task in plan.json:
   │   ├─ Check dependencies (wait if needed)
   │   ├─ Execute via assigned executor (agent/gemini/codex)
   │   ├─ Document progress in progress.md → Execution section
   │   └─ Handle failures (retry/skip/abort)
   ├─ Aggregate execution results
   └─ Transition → Verify mode

Verify Mode (Gemini-Assisted):
   ├─ Invoke Gemini CLI for code review
   ├─ Analyze: correctness, style, potential issues
   ├─ Document in progress.md → Verification section
   ├─ Decision:
   │   ├─ All passed → Complete
   │   ├─ Minor issues → Document corrections, iterate
   │   └─ Major issues → Rollback, re-plan
   └─ Transition → Iterate or Complete

Iterate Mode (Incremental):
   ├─ User provides new requirements or fixes
   ├─ Read progress.md for current state
   ├─ Gemini analyzes delta (what changed?)
   ├─ Generate incremental tasks, append to plan.json
   ├─ Update progress.md → New iteration section
   └─ Transition → Execute mode (incremental tasks only)

Implementation

Session Setup & Mode Detection

const getUtc8ISOString = () => new Date(Date.now() + 8 * 60 * 60 * 1000).toISOString()

const taskSlug = task_description.toLowerCase().replace(/[^a-z0-9]+/g, '-').substring(0, 40)
const dateStr = getUtc8ISOString().substring(0, 10)

const sessionId = flags.resume || `DEV-${taskSlug}-${dateStr}`
const sessionFolder = `.workflow/.develop/${sessionId}`
const progressPath = `${sessionFolder}/progress.md`
const planPath = `${sessionFolder}/plan.json`

// Auto-detect mode
const sessionExists = fs.existsSync(sessionFolder)
const hasProgress = sessionExists && fs.existsSync(progressPath)
const hasPlan = sessionExists && fs.existsSync(planPath)

let mode = 'explore'
if (hasProgress) {
  const progress = Read(progressPath)
  // Parse progress.md to determine current phase
  if (progress.includes('## Verification Results')) mode = 'iterate'
  else if (progress.includes('## Execution Timeline')) mode = 'verify'
  else if (progress.includes('## Planning Results')) mode = 'execute'
  else if (progress.includes('## Exploration Results')) mode = 'plan'
}

if (!sessionExists) {
  bash(`mkdir -p ${sessionFolder}`)
}

console.log(`
## Session Info
- Session ID: ${sessionId}
- Mode: ${mode}
- Folder: ${sessionFolder}
`)

---

Explore Mode (Multi-Agent Parallel)

Step 1.1: Complexity Assessment

// Analyze task complexity based on:
// - Scope: How many systems/modules affected?
// - Depth: Surface change vs architectural impact?
// - Risk: Potential for breaking existing functionality?
// - Dependencies: How interconnected is the change?

const complexity = analyzeTaskComplexity(task_description)
// Returns: 'Low' (1 angle) | 'Medium' (2-3 angles) | 'High' (4 angles)

const ANGLE_PRESETS = {
  architecture: ['architecture', 'dependencies', 'modularity', 'integration-points'],
  feature: ['patterns', 'integration-points', 'testing', 'dependencies'],
  bugfix: ['error-handling', 'dataflow', 'state-management', 'edge-cases'],
  refactor: ['code-quality', 'patterns', 'dependencies', 'testing']
}

function selectAngles(taskDescription, complexity) {
  const text = taskDescription.toLowerCase()
  let preset = 'feature' // default

  if (/refactor|architect|restructure/.test(text)) preset = 'architecture'
  else if (/fix|bug|error|issue/.test(text)) preset = 'bugfix'
  else if (/clean|improve|simplify/.test(text)) preset = 'refactor'

  const count = complexity === 'High' ? 4 : (complexity === 'Medium' ? 3 : 1)
  return ANGLE_PRESETS[preset].slice(0, count)
}

const selectedAngles = selectAngles(task_description, complexity)

Step 1.2: Launch Parallel Explorations

// Launch agents with pre-assigned angles
const explorationTasks = selectedAngles.map((angle, index) =>
  Task(
    subagent_type="cli-explore-agent",
    run_in_background=false,  // ⚠️ MANDATORY: Must wait for results
    description=`Explore: ${angle}`,
    prompt=`
## Task Objective
Execute **${angle}** exploration for development planning context.

## Assigned Context
- **Exploration Angle**: ${angle}
- **Task Description**: ${task_description}
- **Output File**: ${sessionFolder}/exploration-${angle}.json

## MANDATORY FIRST STEPS
1. cat ~/.claude/workflows/cli-templates/schemas/explore-json-schema.json (schema)
2. ccw tool exec get_modules_by_depth '{}' (project structure)
3. Read .workflow/project-tech.json (tech stack)
4. Read .workflow/project-guidelines.json (constraints)

## Exploration Strategy (${angle} focus)

**Structural Scan** (Bash):
- get_modules_by_depth.sh → modules related to ${angle}
- rg -l "{keywords}" --type ts → locate relevant files
- Analyze imports/dependencies from ${angle} perspective

**Semantic Analysis** (Gemini CLI):
- ccw cli -p "PURPOSE: Analyze ${angle} aspect for: ${task_description}
  TASK: • Identify ${angle} patterns • Locate integration points • List constraints
  MODE: analysis
  CONTEXT: @**/* | Task: ${task_description}
  EXPECTED: ${angle} findings with file:line references
  CONSTRAINTS: Focus on ${angle}" --tool gemini --mode analysis

**Output**: ${sessionFolder}/exploration-${angle}.json (follow schema)

## Required Fields
- project_structure: ${angle}-relevant modules
- relevant_files: [{path, relevance (0-1), rationale}]
- patterns: ${angle} patterns with code examples
- integration_points: Where to integrate (file:line)
- constraints: ${angle}-specific limitations
- clarification_needs: [{question, context, options, recommended}]
- _metadata.exploration_angle: "${angle}"

## Success Criteria
- Schema validated
- ≥3 relevant files with rationale
- Integration points with file:line
- Actionable patterns (code examples)
- JSON output matches schema
`
  )
)

// Execute all explorations

Step 1.3: Aggregate Exploration Results

// Auto-discover exploration files
const explorationFiles = bash(`find ${sessionFolder} -name "exploration-*.json" -type f`)
  .split('\n')
  .filter(f => f.trim())

const explorations = explorationFiles.map(file => {
  const data = JSON.parse(Read(file))
  return {
    angle: data._metadata.exploration_angle,
    data: data
  }
})

// Aggregate clarification needs
const allClarifications = []
explorations.forEach(exp => {
  if (exp.data.clarification_needs?.length > 0) {
    exp.data.clarification_needs.forEach(need => {
      allClarifications.push({ ...need, source_angle: exp.angle })
    })
  }
})

// Intelligent deduplication
const dedupedClarifications = intelligentMerge(allClarifications)

// Multi-round clarification (max 4 questions per round)
if (dedupedClarifications.length > 0) {
  const BATCH_SIZE = 4
  for (let i = 0; i < dedupedClarifications.length; i += BATCH_SIZE) {
    const batch = dedupedClarifications.slice(i, i + BATCH_SIZE)
    AskUserQuestion({
      questions: batch.map(need => ({
        question: `[${need.source_angle}] ${need.question}`,
        header: need.source_angle.substring(0, 12),
        multiSelect: false,
        options: need.options.map((opt, idx) => ({
          label: need.recommended === idx ? `${opt} ★` : opt,
          description: need.recommended === idx ? "Recommended" : `Use ${opt}`
        }))
      }))
    })
  }
}

Step 1.4: Document in progress.md

Create or update progress.md:

# Development Progress

**Session ID**: ${sessionId}
**Task Description**: ${task_description}
**Started**: ${getUtc8ISOString()}
**Complexity**: ${complexity}

---

## Exploration Results

### Exploration Summary (${getUtc8ISOString()})

**Angles Explored**: ${selectedAngles.join(', ')}
**Exploration Count**: ${explorations.length}

${explorations.map(exp => `
#### ${exp.angle.toUpperCase()}

**Relevant Files** (top 5):
${exp.data.relevant_files.slice(0, 5).map(f =>
  `- ${f.path} (${(f.relevance * 100).toFixed(0)}%) - ${f.rationale}`
).join('\n')}

**Integration Points**:
${exp.data.integration_points.map(p => `- ${p}`).join('\n')}

**Patterns**:
${exp.data.patterns.map(p => `- ${p}`).join('\n')}

**Constraints**:
${exp.data.constraints.map(c => `- ${c}`).join('\n')}
`).join('\n')}

### Clarifications Collected

${dedupedClarifications.length > 0 ? dedupedClarifications.map((c, i) => `
${i+1}. **[${c.source_angle}]** ${c.question}
   - Answer: ${c.user_answer || 'Pending'}
`).join('\n') : 'No clarifications needed'}

---

## Current State

**Phase**: Exploration complete, ready for planning
**Next Steps**: Invoke Gemini for task decomposition

---

Plan Mode (Gemini-Guided)

Step 2.1: Gemini-Assisted Planning

ccw cli -p "
PURPOSE: Generate implementation plan for: ${task_description}
Success criteria: 2-7 structured tasks with clear dependencies, executor assignments

TASK:
• Read progress.md exploration findings (all angles)
• Analyze task requirements and constraints
• Decompose into 2-7 substantial tasks (15-60min each)
• Assign executor per task (gemini/codex/agent) based on:
  - User explicit mention (\"use gemini for X\")
  - Task nature (analysis → gemini, implementation → agent, git-aware → codex)
  - Default → agent
• Define true dependencies only (Task B needs Task A's output)
• Group by feature/module, NOT by file

MODE: analysis

CONTEXT: @${progressPath} | Exploration findings in progress.md

EXPECTED:
- plan.json following schema (cat ~/.claude/workflows/cli-templates/schemas/plan-json-schema.json)
- tasks: [{id, title, description, files, executor (gemini/codex/agent), depends_on, complexity}]
- executorAssignments: {taskId: {executor, reason}}
- estimated_time, complexity, summary, approach

CONSTRAINTS:
- Respect project-guidelines.json constraints
- Prefer parallel tasks (minimal depends_on)
- Group related changes into single task
- Explicit executor rationale
" --tool gemini --mode analysis --rule planning-breakdown-task-steps

Step 2.2: Parse Gemini Output → plan.json

// Gemini returns plan structure
const planFromGemini = parseGeminiOutput()

const plan = {
  summary: planFromGemini.summary,
  approach: planFromGemini.approach,
  tasks: planFromGemini.tasks.map(t => ({
    id: t.id,
    title: t.title,
    description: t.description,
    files: t.files,
    executor: t.executor || 'agent',  // gemini/codex/agent
    depends_on: t.depends_on || [],
    complexity: t.complexity,
    status: 'pending'  // pending/in_progress/completed/failed
  })),
  executorAssignments: planFromGemini.executorAssignments || {},
  estimated_time: planFromGemini.estimated_time,
  complexity: complexity,
  _metadata: {
    timestamp: getUtc8ISOString(),
    source: 'gemini-planning',
    exploration_angles: selectedAngles
  }
}

Write(planPath, JSON.stringify(plan, null, 2))

Step 2.3: Update progress.md

Append to progress.md:

## Planning Results

### Plan Generated (${getUtc8ISOString()})

**Summary**: ${plan.summary}
**Approach**: ${plan.approach}
**Estimated Time**: ${plan.estimated_time}
**Complexity**: ${plan.complexity}

**Tasks** (${plan.tasks.length}):

${plan.tasks.map((t, i) => `
${i+1}. **${t.id}**: ${t.title}
   - Files: ${t.files.join(', ')}
   - Executor: ${t.executor}
   - Depends on: ${t.depends_on.join(', ') || 'None'}
   - Complexity: ${t.complexity}
`).join('\n')}

**Executor Assignments**:
${Object.entries(plan.executorAssignments).map(([id, assign]) =>
  `- ${id}: ${assign.executor} (Reason: ${assign.reason})`
).join('\n')}

**Gemini Insights**:
${planFromGemini.gemini_insights || 'N/A'}

---

## Current State

**Phase**: Planning complete, awaiting user confirmation
**Next Steps**: User confirms → Execute tasks

Step 2.4: User Confirmation

AskUserQuestion({
  questions: [
    {
      question: `Confirm plan? (${plan.tasks.length} tasks, ${plan.complexity})`,
      header: "Confirm",
      multiSelect: false,
      options: [
        { label: "Allow", description: "Proceed as-is" },
        { label: "Modify", description: "Adjust tasks/executors" },
        { label: "Cancel", description: "Abort workflow" }
      ]
    }
  ]
})

// If Modify → allow editing plan.json, re-confirm
// If Allow → proceed to Execute mode

---

Execute Mode (Multi-Agent/CLI)

Step 3.1: Task Execution Loop

const plan = JSON.parse(Read(planPath))
const completedTasks = []
const failedTasks = []

for (const task of plan.tasks) {
  // Check dependencies
  const depsCompleted = task.depends_on.every(depId =>
    completedTasks.some(t => t.id === depId)
  )

  if (!depsCompleted) {
    console.log(`⏸️ Task ${task.id} waiting for dependencies: ${task.depends_on.join(', ')}`)
    continue
  }

  console.log(`\n## Executing Task: ${task.id} (${task.executor})\n`)

  // Update progress.md
  appendToProgress(`
### Task ${task.id} - ${task.title} (${getUtc8ISOString()})

**Status**: In Progress
**Executor**: ${task.executor}
**Files**: ${task.files.join(', ')}
`)

  // Execute based on assigned executor
  try {
    if (task.executor === 'gemini') {
      // Gemini CLI execution
      bash(`ccw cli -p "
PURPOSE: ${task.description}
TASK: Implement changes in: ${task.files.join(', ')}
MODE: write
CONTEXT: @${task.files.join(' @')} | Memory: Task ${task.id} from ${sessionId}
EXPECTED: Code implementation following project patterns
CONSTRAINTS: ${task.constraints || 'Follow project-guidelines.json'}
" --tool gemini --mode write --cd ${sessionFolder}`, run_in_background=true)

      // Wait for callback

    } else if (task.executor === 'codex') {
      // Codex CLI execution
      bash(`ccw cli -p "
${task.description}

Files to modify: ${task.files.join(', ')}
" --tool codex --mode write`, run_in_background=true)

      // Wait for callback

    } else {
      // Agent execution
      Task(
        subagent_type="code-developer",
        run_in_background=false,
        description=`Execute task ${task.id}`,
        prompt=`
## Task: ${task.title}

${task.description}

## Context from Exploration
Read ${progressPath} for exploration findings relevant to:
${task.files.map(f => `- ${f}`).join('\n')}

## Files to Modify
${task.files.join('\n')}

## Constraints
- Follow patterns identified in exploration
- Respect project-guidelines.json
- Write tests if applicable

## Success Criteria
- Code compiles without errors
- Tests pass (if applicable)
- Follows existing code style
`
      )
    }

    // Mark completed
    task.status = 'completed'
    completedTasks.push(task)

    appendToProgress(`
**Status**: ✅ Completed
**Output**: ${task.output || 'Code changes applied'}
`)

  } catch (error) {
    task.status = 'failed'
    failedTasks.push(task)

    appendToProgress(`
**Status**: ❌ Failed
**Error**: ${error.message}
`)

    // Ask user: retry/skip/abort
    AskUserQuestion({
      questions: [{
        question: `Task ${task.id} failed. How to proceed?`,
        header: "Error",
        multiSelect: false,
        options: [
          { label: "Retry", description: "Retry with same executor" },
          { label: "Skip", description: "Skip and continue" },
          { label: "Abort", description: "Stop workflow" }
        ]
      }]
    })
  }

  // Update plan.json
  Write(planPath, JSON.stringify(plan, null, 2))
}

// Final summary
appendToProgress(`
---

## Execution Summary (${getUtc8ISOString()})

**Total Tasks**: ${plan.tasks.length}
**Completed**: ${completedTasks.length}
**Failed**: ${failedTasks.length}

${failedTasks.length > 0 ? `
**Failed Tasks**:
${failedTasks.map(t => `- ${t.id}: ${t.title}`).join('\n')}
` : ''}

---

## Current State

**Phase**: Execution complete, ready for verification
**Next Steps**: Invoke Gemini for code review
`)

---

Verify Mode (Gemini-Assisted)

Step 4.1: Gemini Code Review

ccw cli -p "
PURPOSE: Review code changes from development session ${sessionId}
Success criteria: Identify issues, validate correctness, suggest improvements

TASK:
• Review all modified files from execution
• Check: correctness, style consistency, potential bugs, test coverage
• Validate against project-guidelines.json
• Provide actionable feedback with file:line references

MODE: review

CONTEXT: @${progressPath} | Execution results in progress.md

EXPECTED:
- Review report with severity levels (Critical/High/Medium/Low)
- Issues with file:line references and fix suggestions
- Overall quality score (1-10)
- Recommendations for improvements

CONSTRAINTS: Evidence-based feedback only
" --tool gemini --mode review

Step 4.2: Parse Review Results

const reviewResults = parseGeminiReview()

const hasBlockers = reviewResults.issues.some(i => i.severity === 'Critical')
const hasMajorIssues = reviewResults.issues.some(i => i.severity === 'High')

Step 4.3: Update progress.md

Append to progress.md:

## Verification Results

### Code Review (${getUtc8ISOString()})

**Reviewer**: Gemini
**Quality Score**: ${reviewResults.quality_score}/10

**Issues Found** (${reviewResults.issues.length}):

${reviewResults.issues.map(issue => `
#### ${issue.severity}: ${issue.title}
- **File**: ${issue.file}:${issue.line}
- **Description**: ${issue.description}
- **Suggested Fix**:
\`\`\`${issue.language || 'typescript'}
${issue.suggested_fix}
\`\`\`
`).join('\n')}

${reviewResults.issues.length === 0 ? '✅ No issues found. Code review passed.' : ''}

**Recommendations**:
${reviewResults.recommendations.map(r => `- ${r}`).join('\n')}

**Gemini Analysis**:
${reviewResults.gemini_analysis}

---

## Current State

**Phase**: Verification complete
**Decision**: ${hasBlockers ? 'BLOCKERS FOUND - Fix required' : (hasMajorIssues ? 'Major issues - Recommend fixing' : 'PASSED - Ready to commit')}

Step 4.4: Decision

if (hasBlockers) {
  // Critical issues → must fix before commit
  console.log(`
❌ Critical issues found. Transitioning to Iterate mode for fixes.
  `)
  mode = 'iterate'

} else if (hasMajorIssues) {
  // Ask user
  AskUserQuestion({
    questions: [{
      question: "High severity issues found. How to proceed?",
      header: "Review",
      multiSelect: false,
      options: [
        { label: "Fix Issues", description: "Iterate to fix high severity issues" },
        { label: "Accept As-Is", description: "Proceed despite issues" },
        { label: "Manual Review", description: "Stop and review manually" }
      ]
    }]
  })

} else {
  // All passed
  console.log(`
✅ Code review passed. Development complete.

Session artifacts saved in: ${sessionFolder}
- progress.md: Full development timeline
- plan.json: Execution plan

Run \`/workflow:develop-with-file --resume ${sessionId}\` to iterate.
  `)
}

---

Iterate Mode (Incremental)

Step 5.1: Analyze Delta

User provides new requirements or fix instructions:

ccw cli -p "
PURPOSE: Analyze incremental changes for session ${sessionId}
Success criteria: Identify minimal tasks to address new requirements

TASK:
• Read progress.md for current state
• Compare new requirements with existing implementation
• Identify delta (what changed?)
• Generate incremental tasks (append to existing plan)

MODE: analysis

CONTEXT:
@${progressPath}
@${planPath}
| New requirements: ${new_requirements}

EXPECTED:
- Delta analysis (what's new, what needs changing)
- Incremental tasks [{id, title, description, files, executor}]
- Updated plan.json with new tasks

CONSTRAINTS: Minimal changes, preserve existing work
" --tool gemini --mode analysis --rule planning-breakdown-task-steps

Step 5.2: Append Incremental Tasks

const plan = JSON.parse(Read(planPath))
const incrementalTasks = parseGeminiDelta()

// Assign new task IDs
const maxId = Math.max(...plan.tasks.map(t => parseInt(t.id.replace('T', ''))))
incrementalTasks.forEach((task, i) => {
  task.id = `T${maxId + i + 1}`
  task.status = 'pending'
  plan.tasks.push(task)
})

Write(planPath, JSON.stringify(plan, null, 2))

Step 5.3: Update progress.md

Append new iteration section:

---

## Iteration ${iteration_number} (${getUtc8ISOString()})

### New Requirements
${new_requirements}

### Delta Analysis
${incrementalTasks.delta_analysis}

### Incremental Tasks
${incrementalTasks.tasks.map(t => `
- **${t.id}**: ${t.title}
  - Files: ${t.files.join(', ')}
  - Executor: ${t.executor}
`).join('\n')}

---

## Current State

**Phase**: Iteration planning complete
**Next Steps**: Execute incremental tasks

Step 5.4: Execute Incremental Tasks

Re-enter Execute Mode, but only process tasks with status: 'pending'.

---

Session Folder Structure

.workflow/.develop/{session-id}/
├── progress.md                    # Timeline: exploration → planning → execution → verification
├── plan.json                      # Current execution plan (tasks, statuses, executors)
└── exploration-{angle}.json       # Temporary exploration results (optional, can delete after planning)

Example:

.workflow/.develop/DEV-implement-jwt-refresh-2025-01-23/
├── progress.md
├── plan.json
├── exploration-architecture.json
├── exploration-patterns.json
└── exploration-testing.json

Progress.md Template

# Development Progress

**Session ID**: DEV-xxx-2025-01-23
**Task Description**: [original description]
**Started**: 2025-01-23T10:00:00+08:00
**Complexity**: Medium

---

## Exploration Results

### Exploration Summary (2025-01-23 10:05)
...

### Clarifications Collected
...

---

## Planning Results

### Plan Generated (2025-01-23 10:15)
...

---

## Execution Timeline

### Task T1 - ... (2025-01-23 10:20)
**Status**: ✅ Completed
...

### Task T2 - ... (2025-01-23 10:35)
**Status**: ✅ Completed
...

---

## Verification Results

### Code Review (2025-01-23 11:00)
...

---

## Iteration 2 (2025-01-23 14:00)

### New Requirements
...

### Delta Analysis
...

---

## Current State

**Phase**: Complete
**Quality Score**: 8/10
**Artifacts**: progress.md, plan.json

Plan.json Schema

{
  "summary": "Implementation summary",
  "approach": "Technical approach description",
  "tasks": [
    {
      "id": "T1",
      "title": "Task title",
      "description": "Detailed description",
      "files": ["src/file1.ts", "src/file2.ts"],
      "executor": "agent",  // gemini/codex/agent
      "depends_on": [],
      "complexity": "Medium",
      "status": "completed"  // pending/in_progress/completed/failed
    }
  ],
  "executorAssignments": {
    "T1": {
      "executor": "agent",
      "reason": "Standard implementation task, agent handles well"
    }
  },
  "estimated_time": "2-3 hours",
  "complexity": "Medium",
  "_metadata": {
    "timestamp": "2025-01-23T10:15:00+08:00",
    "source": "gemini-planning",
    "exploration_angles": ["architecture", "patterns", "testing"]
  }
}

Gemini Integration Points

1. Planning (Plan Mode)

Purpose: Intelligent task decomposition with executor assignment

Prompt Pattern:

PURPOSE: Generate implementation plan + executor assignments
TASK: Analyze exploration → decompose → assign executors (gemini/codex/agent)
CONTEXT: @progress.md (exploration findings)
EXPECTED: plan.json with tasks, executorAssignments, rationale

2. Verification (Verify Mode)

Purpose: Post-execution code review and quality check

Prompt Pattern:

PURPOSE: Review code changes + validate quality
TASK: Check correctness, style, bugs → severity levels → fix suggestions
CONTEXT: @progress.md (execution results)
EXPECTED: Review report with issues (Critical/High/Medium/Low), quality score

3. Iteration (Iterate Mode)

Purpose: Analyze delta and generate incremental tasks

Prompt Pattern:

PURPOSE: Analyze incremental changes + minimal tasks
TASK: Compare new requirements with current state → delta → incremental tasks
CONTEXT: @progress.md @plan.json | New requirements
EXPECTED: Delta analysis + incremental tasks

Error Correction Mechanism

Correction Format in progress.md

When verification finds issues:

### Corrected Understanding (Iteration 2)

- ~~Assumed API returns array~~ → API returns object with data array
  - Why wrong: Misread API response structure
  - Evidence: Runtime error "map is not a function"
  - Fix applied: Changed `response.map` to `response.data.map`

- ~~Thought validation was client-side~~ → Validation is server-side
  - Why wrong: Only checked frontend code
  - Evidence: Backend validation logic found in exploration
  - Fix applied: Removed redundant client validation

Error Handling

Situation	Action
Exploration agent failure	Skip angle, continue with remaining explorations
Gemini planning unavailable	Fallback to direct Claude planning (Low complexity mode)
Task execution failure	Ask user: Retry/Skip/Abort
Verification finds blockers	Force iteration mode, cannot proceed
plan.json corrupted	Regenerate from progress.md + Gemini
>5 iterations	Suggest breaking into sub-sessions

Comparison with /workflow:lite-plan

Feature	/workflow:lite-plan	/workflow:develop-with-file
State tracking	In-memory → lite-execute	progress.md (persistent)
Exploration	Multi-agent parallel	✅ Same
Planning	cli-lite-planning-agent	Gemini CLI (more flexible)
Execution	Delegated to lite-execute	Built-in (multi-agent/CLI)
Verification	None	✅ Gemini code review
Iteration support	❌	✅ Incremental tasks
Document continuity	❌ (session ends)	✅ (progress.md timeline)
Executor assignment	Global only	✅ Per-task

Usage Recommendations

Use /workflow:develop-with-file when:

• Complex features requiring multiple phases (explore → plan → execute → verify)
• Need persistent progress tracking across sessions
• Want Gemini-assisted planning and verification
• Anticipate multiple iterations/refinements
• Team needs to understand development rationale

Use /workflow:lite-plan when:

• Quick one-off tasks
• Simple features (Low complexity)
• Don't need verification
• Session-level documentation not needed