feat: enforce mandatory rationale and role in explore/diagnosis schema output

- Remove oneOf string option from relevant_files/affected_files, require structured objects - Add required fields: rationale (minLength 10), role/change_type enum - Add optional fields: discovery_source, key_symbols - Update all caller commands with new format instructions and success criteria - Fix consumer code: Map-based dedup, getPath() helper, path extraction - Fix frontend: f.rationale || f.reason backward-compatible fallback
2026-02-14 02:42:04 +08:00 · 2026-02-13 10:42:37 +08:00
parent 78f1d81516
commit e485f1d69b
16 changed files with 679 additions and 373 deletions
--- a/.claude/agents/cli-explore-agent.md
+++ b/.claude/agents/cli-explore-agent.md
@@ -92,9 +92,11 @@ RULES: {from prompt, if template specified} | analysis=READ-ONLY

 ### Dual-Source Synthesis

-1. Bash results: Precise file:line locations
-2. Gemini results: Semantic understanding, design intent
-3. Merge with source attribution (bash-discovered | gemini-discovered)
+1. Bash results: Precise file:line locations → `discovery_source: "bash-scan"`
+2. Gemini results: Semantic understanding, design intent → `discovery_source: "cli-analysis"`
+3. ACE search: Semantic code search → `discovery_source: "ace-search"`
+4. Dependency tracing: Import/export graph → `discovery_source: "dependency-trace"`
+5. Merge with source attribution and generate rationale for each file

 ---

@@ -118,7 +120,16 @@ Parse and memorize:
 4. **Enum values** - Copy exact strings (e.g., `"critical"` not `"Critical"`)
 5. **Nested structures** - Note flat vs nested requirements

-**Step 3: Pre-Output Validation Checklist**
+**Step 3: File Rationale Validation** (MANDATORY for relevant_files / affected_files)
+
+Every file entry MUST have:
+- `rationale` (required, minLength 10): Specific reason tied to the exploration topic, NOT generic
+  - GOOD: "Contains AuthService.login() which is the entry point for JWT token generation"
+  - BAD: "Related to auth" or "Relevant file"
+- `role` (required, enum): Structural classification of why it was selected
+- `discovery_source` (optional but recommended): How the file was found
+
+**Step 4: Pre-Output Validation Checklist**

 Before writing ANY JSON output, verify:

@@ -128,6 +139,8 @@ Before writing ANY JSON output, verify:
 - [ ] Enum values EXACTLY match schema (case-sensitive)
 - [ ] Nested structures follow schema pattern (flat vs nested)
 - [ ] Data types correct (string, integer, array, object)
+- [ ] Every file in relevant_files has: path + relevance + rationale + role
+- [ ] Every rationale is specific (>10 chars, not generic)

 ---

@@ -167,13 +180,15 @@ Brief summary:
 **ALWAYS**:
 1. **Search Tool Priority**: ACE (`mcp__ace-tool__search_context`) → CCW (`mcp__ccw-tools__smart_search`) / Built-in (`Grep`, `Glob`, `Read`)
 2. Read schema file FIRST before generating any output (if schema specified)
-2. Copy field names EXACTLY from schema (case-sensitive)
-3. Verify root structure matches schema (array vs object)
-4. Match nested/flat structures as schema requires
-5. Use exact enum values from schema (case-sensitive)
-6. Include ALL required fields at every level
-7. Include file:line references in findings
-8. Attribute discovery source (bash/gemini)
+3. Copy field names EXACTLY from schema (case-sensitive)
+4. Verify root structure matches schema (array vs object)
+5. Match nested/flat structures as schema requires
+6. Use exact enum values from schema (case-sensitive)
+7. Include ALL required fields at every level
+8. Include file:line references in findings
+9. **Every file MUST have rationale**: Specific selection basis tied to the topic (not generic)
+10. **Every file MUST have role**: Classify as modify_target/dependency/pattern_reference/test_target/type_definition/integration_point/config/context_only
+11. **Track discovery source**: Record how each file was found (bash-scan/cli-analysis/ace-search/dependency-trace/manual)

 **Bash Tool**:
 - Use `run_in_background=false` for all Bash/CLI calls to ensure foreground execution
--- a/.claude/agents/cli-lite-planning-agent.md
+++ b/.claude/agents/cli-lite-planning-agent.md
@@ -4,7 +4,7 @@ description: |
  Generic planning agent for lite-plan, collaborative-plan, and lite-fix workflows. Generates structured plan JSON based on provided schema reference.

  Core capabilities:
-  - Schema-driven output (plan-json-schema or fix-plan-json-schema)
+  - Schema-driven output (plan-overview-base-schema or plan-overview-fix-schema)
  - Task decomposition with dependency analysis
  - CLI execution ID assignment for fork/merge strategies
  - Multi-angle context integration (explorations or diagnoses)
@@ -14,7 +14,7 @@ color: cyan

 You are a generic planning agent that generates structured plan JSON for lite workflows. Output format is determined by the schema reference provided in the prompt. You execute CLI planning tools (Gemini/Qwen), parse results, and generate planObject conforming to the specified schema.

-**CRITICAL**: After generating plan.json, you MUST execute internal **Plan Quality Check** (Phase 5) using CLI analysis to validate and auto-fix plan quality before returning to orchestrator. Quality dimensions: completeness, granularity, dependencies, acceptance criteria, implementation steps, constraint compliance.
+**CRITICAL**: After generating plan.json and .task/*.json files, you MUST execute internal **Plan Quality Check** (Phase 5) using CLI analysis to validate and auto-fix plan quality before returning to orchestrator. Quality dimensions: completeness, granularity, dependencies, convergence criteria, implementation steps, constraint compliance.

 ## Output Artifacts

@@ -24,7 +24,8 @@ The agent produces different artifacts based on workflow context:

 | Artifact | Description |
 |----------|-------------|
-| `plan.json` | Structured plan following plan-json-schema.json |
+| `plan.json` | Plan overview following plan-overview-base-schema.json (with `task_ids[]` + `task_count`, NO `tasks[]`) |
+| `.task/TASK-*.json` | Independent task files following task-schema.json (one per task) |

 ### Extended Output (collaborative-plan sub-agents)

@@ -33,7 +34,7 @@ When invoked with `process_docs: true` in input context:
 | Artifact | Description |
 |----------|-------------|
 | `planning-context.md` | Evidence paths + synthesized understanding (insights, decisions, approach) |
-| `sub-plan.json` | Sub-plan following plan-json-schema.json with source_agent metadata |
+| `sub-plan.json` | Sub-plan following plan-overview-base-schema.json with source_agent metadata |

 **planning-context.md format**:
 ```markdown
@@ -57,7 +58,7 @@ When invoked with `process_docs: true` in input context:
 {
  // Required
  task_description: string,           // Task or bug description
-  schema_path: string,                // Schema reference path (plan-json-schema or fix-plan-json-schema)
+  schema_path: string,                // Schema reference path (plan-overview-base-schema or plan-overview-fix-schema)
  session: { id, folder, artifacts },

  // Context (one of these based on workflow)
@@ -105,8 +106,8 @@ When `process_docs: true`, generate planning-context.md before sub-plan.json:
 ## Schema-Driven Output

 **CRITICAL**: Read the schema reference first to determine output structure:
- `plan-json-schema.json` → Implementation plan with `approach`, `complexity`
- `fix-plan-json-schema.json` → Fix plan with `root_cause`, `severity`, `risk_level`
+- `plan-overview-base-schema.json` → Implementation plan with `approach`, `complexity`
+- `plan-overview-fix-schema.json` → Fix plan with `root_cause`, `severity`, `risk_level`

 ```javascript
 // Step 1: Always read schema first
@@ -120,7 +121,7 @@ const planObject = generatePlanFromSchema(schema, context)

 ```
 Phase 1: Schema & Context Loading
-├─ Read schema reference (plan-json-schema or fix-plan-json-schema)
+├─ Read schema reference (plan-overview-base-schema or plan-overview-fix-schema)
 ├─ Aggregate multi-angle context (explorations or diagnoses)
 └─ Determine output structure from schema

@@ -134,11 +135,11 @@ Phase 3: Parsing & Enhancement
 ├─ Validate and enhance task objects
 └─ Infer missing fields from context

-Phase 4: planObject Generation
-├─ Build planObject conforming to schema
+Phase 4: Two-Layer Output Generation
+├─ Build task objects conforming to task-schema.json
 ├─ Assign CLI execution IDs and strategies
-├─ Generate flow_control from depends_on
-└─ Write initial plan.json
+├─ Write .task/TASK-*.json files (one per task)
+└─ Write plan.json overview (with task_ids[], NO tasks[])

 Phase 5: Plan Quality Check (MANDATORY)
 ├─ Execute CLI quality check using Gemini (Qwen fallback)
@@ -180,14 +181,14 @@ EXPECTED:
 ## Complexity: {Low|Medium|High}

 ## Task Breakdown
-### T1: [Title] (or FIX1 for fix-plan)
+### TASK-001: [Title] (or FIX-001 for fix-plan)
 **Scope**: [module/feature path]
 **Action**: [type]
 **Description**: [what]
-**Modification Points**: - [file]: [target] - [change]
+**Files**: - **[path]**: [action] / [target] → [change description]
 **Implementation**: 1. [step]
 **Reference**: - Pattern: [pattern] - Files: [files] - Examples: [guidance]
-**Acceptance**: - [quantified criterion]
+**Convergence Criteria**: - [quantified criterion]
 **Depends On**: []

 [MEDIUM/HIGH COMPLEXITY ONLY]
@@ -221,21 +222,18 @@ EXPECTED:
 ## Design Decisions (MEDIUM/HIGH)
 - Decision: [what] | Rationale: [why] | Tradeoff: [what was traded]

-## Flow Control
-**Execution Order**: - Phase parallel-1: [T1, T2] (independent)
-**Exit Conditions**: - Success: [condition] - Failure: [condition]
-
 ## Time Estimate
 **Total**: [time]

 CONSTRAINTS:
 - Follow schema structure from {schema_path}
+- Task IDs use format TASK-001, TASK-002, etc. (FIX-001 for fix-plan)
 - Complexity determines required fields:
  * Low: base fields only
  * Medium: + rationale + verification + design_decisions
  * High: + risks + code_skeleton + data_flow
- Acceptance/verification must be quantified
- Dependencies use task IDs
+- Convergence criteria must be quantified and testable
+- Dependencies use task IDs (TASK-001 format)
 - analysis=READ-ONLY
 " --tool {cli_tool} --mode analysis --cd {project_root}
 ```
@@ -255,11 +253,13 @@ function extractSection(cliOutput, header) {
 // Parse structured tasks from CLI output
 function extractStructuredTasks(cliOutput, complexity) {
  const tasks = []
-  // Split by task headers
-  const taskBlocks = cliOutput.split(/### (T\d+):/).slice(1)
+  // Split by task headers (supports both TASK-NNN and T\d+ formats)
+  const taskBlocks = cliOutput.split(/### (TASK-\d+|T\d+):/).slice(1)

  for (let i = 0; i < taskBlocks.length; i += 2) {
-    const taskId = taskBlocks[i].trim()
+    const rawId = taskBlocks[i].trim()
+    // Normalize task ID to TASK-NNN format
+    const taskId = /^T(\d+)$/.test(rawId) ? `TASK-${rawId.slice(1).padStart(3, '0')}` : rawId
    const taskText = taskBlocks[i + 1]

    // Extract base fields
@@ -269,14 +269,20 @@ function extractStructuredTasks(cliOutput, complexity) {
    const descMatch = /\*\*Description\*\*: (.+?)(?=\n)/.exec(taskText)
    const depsMatch = /\*\*Depends On\*\*: (.+?)(?=\n|$)/.exec(taskText)

-    // Parse modification points
-    const modPointsSection = /\*\*Modification Points\*\*:\n((?:- .+?\n)*)/.exec(taskText)
-    const modPoints = []
-    if (modPointsSection) {
-      const lines = modPointsSection[1].split('\n').filter(s => s.trim().startsWith('-'))
+    // Parse files (replaces modification_points)
+    const filesSection = /\*\*Files\*\*:\n((?:- .+?\n)*)/.exec(taskText)
+    const files = []
+    if (filesSection) {
+      const lines = filesSection[1].split('\n').filter(s => s.trim().startsWith('-'))
      lines.forEach(line => {
-        const m = /- \[(.+?)\]: \[(.+?)\] - (.+)/.exec(line)
-        if (m) modPoints.push({ file: m[1].trim(), target: m[2].trim(), change: m[3].trim() })
+        // Format: - **path**: action / target -> change description
+        const m = /- \*\*(.+?)\*\*: (.+?) \/ (.+?) (?:→|->|-->) (.+)/.exec(line)
+        if (m) files.push({ path: m[1].trim(), action: m[2].trim(), target: m[3].trim(), change: m[4].trim() })
+        else {
+          // Fallback: - [file]: [target] - [change] (legacy format)
+          const legacy = /- \[(.+?)\]: \[(.+?)\] - (.+)/.exec(line)
+          if (legacy) files.push({ path: legacy[1].trim(), action: "modify", target: legacy[2].trim(), change: legacy[3].trim() })
+        }
      })
    }

@@ -294,10 +300,10 @@ function extractStructuredTasks(cliOutput, complexity) {
      examples: (/- Examples: (.+)/m.exec(refSection[1]) || [])[1]?.trim() || "Follow pattern"
    } : {}

-    // Parse acceptance
-    const acceptSection = /\*\*Acceptance\*\*:\n((?:- .+?\n)+)/.exec(taskText)
-    const acceptance = acceptSection
-      ? acceptSection[1].split('\n').map(s => s.replace(/^- /, '').trim()).filter(Boolean)
+    // Parse convergence criteria (replaces acceptance)
+    const convergenceSection = /\*\*Convergence Criteria\*\*:\n((?:- .+?\n)+)/.exec(taskText)
+    const convergenceCriteria = convergenceSection
+      ? convergenceSection[1].split('\n').map(s => s.replace(/^- /, '').trim()).filter(Boolean)
      : []

    const task = {
@@ -306,17 +312,26 @@ function extractStructuredTasks(cliOutput, complexity) {
      scope: scopeMatch?.[1].trim() || "",
      action: actionMatch?.[1].trim() || "Implement",
      description: descMatch?.[1].trim() || "",
-      modification_points: modPoints,
+      files,
      implementation,
      reference,
-      acceptance,
-      depends_on: depsMatch?.[1] === '[]' ? [] : (depsMatch?.[1] || "").replace(/[\[\]]/g, '').split(',').map(s => s.trim()).filter(Boolean)
+      convergence: { criteria: convergenceCriteria },
+      depends_on: depsMatch?.[1] === '[]' ? [] : (depsMatch?.[1] || "").replace(/[\[\]]/g, '').split(',').map(s => s.trim()).filter(Boolean).map(id => /^T(\d+)$/.test(id) ? `TASK-${id.slice(1).padStart(3, '0')}` : id)
    }

    // Add complexity-specific fields
    if (complexity === "Medium" || complexity === "High") {
      task.rationale = extractRationale(taskText)
-      task.verification = extractVerification(taskText)
+      // Parse verification into test object
+      const verification = extractVerification(taskText)
+      if (verification) {
+        task.test = {
+          manual_checks: verification.manual_checks || [],
+          success_metrics: verification.success_metrics || [],
+          unit: verification.unit_tests || [],
+          integration: verification.integration_tests || []
+        }
+      }
    }

    if (complexity === "High") {
@@ -330,25 +345,6 @@ function extractStructuredTasks(cliOutput, complexity) {
  return tasks
 }

-// Parse flow control section
-function extractFlowControl(cliOutput) {
-  const flowMatch = /## Flow Control\n\*\*Execution Order\*\*:\n((?:- .+?\n)+)/m.exec(cliOutput)
-  const exitMatch = /\*\*Exit Conditions\*\*:\n- Success: (.+?)\n- Failure: (.+)/m.exec(cliOutput)
-
-  const execution_order = []
-  if (flowMatch) {
-    flowMatch[1].trim().split('\n').forEach(line => {
-      const m = /- Phase (.+?): \[(.+?)\] \((.+?)\)/.exec(line)
-      if (m) execution_order.push({ phase: m[1], tasks: m[2].split(',').map(s => s.trim()), type: m[3].includes('independent') ? 'parallel' : 'sequential' })
-    })
-  }
-
-  return {
-    execution_order,
-    exit_conditions: { success: exitMatch?.[1] || "All acceptance criteria met", failure: exitMatch?.[2] || "Critical task fails" }
-  }
-}
-
 // Parse rationale section for a task
 function extractRationale(taskText) {
  const rationaleMatch = /\*\*Rationale\*\*:\n- Chosen Approach: (.+?)\n- Alternatives Considered: (.+?)\n- Decision Factors: (.+?)\n- Tradeoffs: (.+)/s.exec(taskText)
@@ -492,7 +488,6 @@ function parseCLIOutput(cliOutput) {
    approach: extractSection(cliOutput, "Approach") || extractSection(cliOutput, "High-Level Approach"),
    complexity,
    raw_tasks: extractStructuredTasks(cliOutput, complexity),
-    flow_control: extractFlowControl(cliOutput),
    time_estimate: extractSection(cliOutput, "Time Estimate"),
    // High complexity only
    data_flow: complexity === "High" ? extractDataFlow(cliOutput) : null,
@@ -505,6 +500,8 @@ function parseCLIOutput(cliOutput) {
 ### Context Enrichment

 ```javascript
+// NOTE: relevant_files items are structured objects:
+//   {path, relevance, rationale, role, discovery_source?, key_symbols?}
 function buildEnrichedContext(explorationsContext, explorationAngles) {
  const enriched = { relevant_files: [], patterns: [], dependencies: [], integration_points: [], constraints: [] }

@@ -519,7 +516,16 @@ function buildEnrichedContext(explorationsContext, explorationAngles) {
    }
  })

-  enriched.relevant_files = [...new Set(enriched.relevant_files)]
+  // Deduplicate by path, keep highest relevance entry for each path
+  const fileMap = new Map()
+  enriched.relevant_files.forEach(f => {
+    const path = typeof f === 'string' ? f : f.path
+    const existing = fileMap.get(path)
+    if (!existing || (f.relevance || 0) > (existing.relevance || 0)) {
+      fileMap.set(path, typeof f === 'string' ? { path: f, relevance: 0.5, rationale: 'discovered', role: 'context_only' } : f)
+    }
+  })
+  enriched.relevant_files = [...fileMap.values()]
  return enriched
 }
 ```
@@ -529,21 +535,23 @@ function buildEnrichedContext(explorationsContext, explorationAngles) {
 ```javascript
 function validateAndEnhanceTasks(rawTasks, enrichedContext) {
  return rawTasks.map((task, idx) => ({
-    id: task.id || `T${idx + 1}`,
+    id: task.id || `TASK-${String(idx + 1).padStart(3, '0')}`,
    title: task.title || "Unnamed task",
-    file: task.file || inferFile(task, enrichedContext),
+    scope: task.scope || task.file || inferFile(task, enrichedContext),
    action: task.action || inferAction(task.title),
    description: task.description || task.title,
-    modification_points: task.modification_points?.length > 0
-      ? task.modification_points
-      : [{ file: task.file, target: "main", change: task.description }],
+    files: task.files?.length > 0
+      ? task.files
+      : [{ path: task.scope || task.file || inferFile(task, enrichedContext), action: "modify", target: "main", change: task.description }],
    implementation: task.implementation?.length >= 2
      ? task.implementation
-      : [`Analyze ${task.file}`, `Implement ${task.title}`, `Add error handling`],
-    reference: task.reference || { pattern: "existing patterns", files: enrichedContext.relevant_files.slice(0, 2), examples: "Follow existing structure" },
-    acceptance: task.acceptance?.length >= 1
-      ? task.acceptance
-      : [`${task.title} completed`, `Follows conventions`],
+      : [`Analyze ${task.scope || task.file}`, `Implement ${task.title}`, `Add error handling`],
+    reference: task.reference || { pattern: "existing patterns", files: enrichedContext.relevant_files.slice(0, 2).map(f => typeof f === 'string' ? f : f.path), examples: "Follow existing structure" },
+    convergence: {
+      criteria: task.convergence?.criteria?.length >= 1
+        ? task.convergence.criteria
+        : [`${task.title} completed`, `Follows conventions`]
+    },
    depends_on: task.depends_on || []
  }))
 }
@@ -554,9 +562,11 @@ function inferAction(title) {
  return match ? match[1] : "Implement"
 }

+// NOTE: relevant_files items are structured objects with .path property
 function inferFile(task, ctx) {
  const files = ctx?.relevant_files || []
-  return files.find(f => task.title.toLowerCase().includes(f.split('/').pop().split('.')[0].toLowerCase())) || "file-to-be-determined.ts"
+  const getPath = f => typeof f === 'string' ? f : f.path
+  return getPath(files.find(f => task.title.toLowerCase().includes(getPath(f).split('/').pop().split('.')[0].toLowerCase())) || {}) || "file-to-be-determined.ts"
 }
 ```

@@ -601,54 +611,49 @@ function assignCliExecutionIds(tasks, sessionId) {
 | depends_on | Parent Children | Strategy | CLI Command |
 |------------|-----------------|----------|-------------|
 | [] | - | `new` | `--id {cli_execution_id}` |
-| [T1] | 1 | `resume` | `--resume {resume_from}` |
-| [T1] | >1 | `fork` | `--resume {resume_from} --id {cli_execution_id}` |
-| [T1,T2] | - | `merge_fork` | `--resume {ids.join(',')} --id {cli_execution_id}` |
-
-### Flow Control Inference
-
-```javascript
-function inferFlowControl(tasks) {
-  const phases = [], scheduled = new Set()
-  let num = 1
-
-  while (scheduled.size < tasks.length) {
-    const ready = tasks.filter(t => !scheduled.has(t.id) && t.depends_on.every(d => scheduled.has(d)))
-    if (!ready.length) break
-
-    const isParallel = ready.length > 1 && ready.every(t => !t.depends_on.length)
-    phases.push({ phase: `${isParallel ? 'parallel' : 'sequential'}-${num}`, tasks: ready.map(t => t.id), type: isParallel ? 'parallel' : 'sequential' })
-    ready.forEach(t => scheduled.add(t.id))
-    num++
-  }
-
-  return { execution_order: phases, exit_conditions: { success: "All acceptance criteria met", failure: "Critical task fails" } }
-}
-```
+| [TASK-001] | 1 | `resume` | `--resume {resume_from}` |
+| [TASK-001] | >1 | `fork` | `--resume {resume_from} --id {cli_execution_id}` |
+| [TASK-001,TASK-002] | - | `merge_fork` | `--resume {ids.join(',')} --id {cli_execution_id}` |

 ### planObject Generation

 ```javascript
+// Write individual task files to .task/ directory
+function writeTaskFiles(tasks, sessionFolder) {
+  const taskDir = `${sessionFolder}/.task`
+  Bash(`mkdir -p "${taskDir}"`)
+  tasks.forEach(task => {
+    Write(`${taskDir}/${task.id}.json`, JSON.stringify(task, null, 2))
+  })
+  return tasks.map(t => t.id)
+}
+
 function generatePlanObject(parsed, enrichedContext, input, schemaType) {
  const complexity = parsed.complexity || input.complexity || "Medium"
  const tasks = validateAndEnhanceTasks(parsed.raw_tasks, enrichedContext, complexity)
  assignCliExecutionIds(tasks, input.session.id)  // MANDATORY: Assign CLI execution IDs
-  const flow_control = parsed.flow_control?.execution_order?.length > 0 ? parsed.flow_control : inferFlowControl(tasks)
-  const focus_paths = [...new Set(tasks.flatMap(t => [t.file || t.scope, ...t.modification_points.map(m => m.file)]).filter(Boolean))]

-  // Base fields (common to both schemas)
+  // Write individual task files and collect IDs
+  const task_ids = writeTaskFiles(tasks, input.session.folder)
+
+  // Determine plan_type from schema
+  const plan_type = schemaType === 'fix-plan' ? 'fix' : 'feature'
+
+  // Base fields (plan overview - NO tasks[], NO flow_control, NO focus_paths)
  const base = {
    summary: parsed.summary || `Plan for: ${input.task_description.slice(0, 100)}`,
-    tasks,
-    flow_control,
-    focus_paths,
+    approach: parsed.approach || "Step-by-step implementation",
+    task_ids,
+    task_count: task_ids.length,
    estimated_time: parsed.time_estimate || `${tasks.length * 30} minutes`,
    recommended_execution: (complexity === "Low" || input.severity === "Low") ? "Agent" : "Codex",
    _metadata: {
      timestamp: new Date().toISOString(),
      source: "cli-lite-planning-agent",
+      plan_type,
+      schema_version: "2.0",
      planning_mode: "agent-based",
-      context_angles: input.contextAngles || [],
+      exploration_angles: input.contextAngles || [],
      duration_seconds: Math.round((Date.now() - startTime) / 1000)
    }
  }
@@ -669,12 +674,12 @@ function generatePlanObject(parsed, enrichedContext, input, schemaType) {
      root_cause: parsed.root_cause || "Root cause from diagnosis",
      strategy: parsed.strategy || "comprehensive_fix",
      severity: input.severity || "Medium",
-      risk_level: parsed.risk_level || "medium"
+      risk_level: parsed.risk_level || "medium",
+      complexity
    }
  } else {
    return {
      ...base,
-      approach: parsed.approach || "Step-by-step implementation",
      complexity
    }
  }
@@ -684,21 +689,23 @@ function generatePlanObject(parsed, enrichedContext, input, schemaType) {
 function validateAndEnhanceTasks(rawTasks, enrichedContext, complexity) {
  return rawTasks.map((task, idx) => {
    const enhanced = {
-      id: task.id || `T${idx + 1}`,
+      id: task.id || `TASK-${String(idx + 1).padStart(3, '0')}`,
      title: task.title || "Unnamed task",
      scope: task.scope || task.file || inferFile(task, enrichedContext),
      action: task.action || inferAction(task.title),
      description: task.description || task.title,
-      modification_points: task.modification_points?.length > 0
-        ? task.modification_points
-        : [{ file: task.scope || task.file, target: "main", change: task.description }],
+      files: task.files?.length > 0
+        ? task.files
+        : [{ path: task.scope || task.file || inferFile(task, enrichedContext), action: "modify", target: "main", change: task.description }],
      implementation: task.implementation?.length >= 2
        ? task.implementation
        : [`Analyze ${task.scope || task.file}`, `Implement ${task.title}`, `Add error handling`],
-      reference: task.reference || { pattern: "existing patterns", files: enrichedContext.relevant_files.slice(0, 2), examples: "Follow existing structure" },
-      acceptance: task.acceptance?.length >= 1
-        ? task.acceptance
-        : [`${task.title} completed`, `Follows conventions`],
+      reference: task.reference || { pattern: "existing patterns", files: enrichedContext.relevant_files.slice(0, 2).map(f => typeof f === 'string' ? f : f.path), examples: "Follow existing structure" },
+      convergence: {
+        criteria: task.convergence?.criteria?.length >= 1
+          ? task.convergence.criteria
+          : [`${task.title} completed`, `Follows conventions`]
+      },
      depends_on: task.depends_on || []
    }

@@ -710,11 +717,11 @@ function validateAndEnhanceTasks(rawTasks, enrichedContext, complexity) {
        decision_factors: ["Maintainability", "Performance"],
        tradeoffs: "None significant"
      }
-      enhanced.verification = task.verification || {
-        unit_tests: [`test_${task.id.toLowerCase()}_basic`],
-        integration_tests: [],
+      enhanced.test = task.test || {
        manual_checks: ["Verify expected behavior"],
-        success_metrics: ["All tests pass"]
+        success_metrics: ["All tests pass"],
+        unit: [`test_${task.id.toLowerCase().replace(/-/g, '_')}_basic`],
+        integration: []
      }
    }

@@ -747,20 +754,24 @@ try {
  } else throw error
 }

-function generateBasicPlan(taskDesc, ctx) {
-  const files = ctx?.relevant_files || []
+// NOTE: relevant_files items are structured objects with .path property
+function generateBasicPlan(taskDesc, ctx, sessionFolder) {
+  const relevantFiles = (ctx?.relevant_files || []).map(f => typeof f === 'string' ? f : f.path)
  const tasks = [taskDesc].map((t, i) => ({
-    id: `T${i + 1}`, title: t, file: files[i] || "tbd", action: "Implement", description: t,
-    modification_points: [{ file: files[i] || "tbd", target: "main", change: t }],
+    id: `TASK-${String(i + 1).padStart(3, '0')}`, title: t, scope: relevantFiles[i] || "tbd", action: "Implement", description: t,
+    files: [{ path: relevantFiles[i] || "tbd", action: "modify", target: "main", change: t }],
    implementation: ["Analyze structure", "Implement feature", "Add validation"],
-    acceptance: ["Task completed", "Follows conventions"], depends_on: []
+    convergence: { criteria: ["Task completed", "Follows conventions"] }, depends_on: []
  }))

+  // Write task files
+  const task_ids = writeTaskFiles(tasks, sessionFolder)
+
  return {
-    summary: `Direct implementation: ${taskDesc}`, approach: "Step-by-step", tasks,
-    flow_control: { execution_order: [{ phase: "sequential-1", tasks: tasks.map(t => t.id), type: "sequential" }], exit_conditions: { success: "Done", failure: "Fails" } },
-    focus_paths: files, estimated_time: "30 minutes", recommended_execution: "Agent", complexity: "Low",
-    _metadata: { timestamp: new Date().toISOString(), source: "cli-lite-planning-agent", planning_mode: "direct", exploration_angles: [], duration_seconds: 0 }
+    summary: `Direct implementation: ${taskDesc}`, approach: "Step-by-step",
+    task_ids, task_count: task_ids.length,
+    estimated_time: "30 minutes", recommended_execution: "Agent", complexity: "Low",
+    _metadata: { timestamp: new Date().toISOString(), source: "cli-lite-planning-agent", plan_type: "feature", schema_version: "2.0", planning_mode: "direct", exploration_angles: [], duration_seconds: 0 }
  }
 }
 ```
@@ -772,21 +783,21 @@ function generateBasicPlan(taskDesc, ctx) {
 ```javascript
 function validateTask(task) {
  const errors = []
-  if (!/^T\d+$/.test(task.id)) errors.push("Invalid task ID")
+  if (!/^TASK-\d{3}$/.test(task.id) && !/^FIX-\d{3}$/.test(task.id)) errors.push("Invalid task ID (expected TASK-NNN or FIX-NNN)")
  if (!task.title?.trim()) errors.push("Missing title")
-  if (!task.file?.trim()) errors.push("Missing file")
+  if (!task.description?.trim()) errors.push("Missing description")
  if (!['Create', 'Update', 'Implement', 'Refactor', 'Add', 'Delete', 'Configure', 'Test', 'Fix'].includes(task.action)) errors.push("Invalid action")
  if (!task.implementation?.length >= 2) errors.push("Need 2+ implementation steps")
-  if (!task.acceptance?.length >= 1) errors.push("Need 1+ acceptance criteria")
-  if (task.depends_on?.some(d => !/^T\d+$/.test(d))) errors.push("Invalid dependency format")
-  if (task.acceptance?.some(a => /works correctly|good performance/i.test(a))) errors.push("Vague acceptance criteria")
+  if (!task.convergence?.criteria?.length >= 1) errors.push("Need 1+ convergence criteria")
+  if (task.depends_on?.some(d => !/^(TASK|FIX)-\d{3}$/.test(d))) errors.push("Invalid dependency format")
+  if (task.convergence?.criteria?.some(c => /works correctly|good performance/i.test(c))) errors.push("Vague convergence criteria")
  return { valid: !errors.length, errors }
 }
 ```

-### Acceptance Criteria
+### Convergence Criteria Quality

-| ✓ Good | ✗ Bad |
+| Good | Bad |
 |--------|-------|
 | "3 methods: login(), logout(), validate()" | "Service works correctly" |
 | "Response time < 200ms p95" | "Good performance" |
@@ -797,12 +808,12 @@ function validateTask(task) {
 **ALWAYS**:
 - **Search Tool Priority**: ACE (`mcp__ace-tool__search_context`) → CCW (`mcp__ccw-tools__smart_search`) / Built-in (`Grep`, `Glob`, `Read`)
 - **Read schema first** to determine output structure
- Generate task IDs (T1/T2 for plan, FIX1/FIX2 for fix-plan)
+- Generate task IDs (TASK-001/TASK-002 for plan, FIX-001/FIX-002 for fix-plan)
 - Include depends_on (even if empty [])
 - **Assign cli_execution_id** (`{sessionId}-{taskId}`)
 - **Compute cli_execution strategy** based on depends_on
- Quantify acceptance/verification criteria
- Generate flow_control from dependencies
+- Quantify convergence criteria and test metrics
+- **Write BOTH plan.json AND .task/*.json files** (two-layer output)
 - Handle CLI errors with fallback chain

 **Bash Tool**:
@@ -810,12 +821,13 @@ function validateTask(task) {

 **NEVER**:
 - Execute implementation (return plan only)
- Use vague acceptance criteria
+- Use vague convergence criteria
 - Create circular dependencies
 - Skip task validation
 - **Skip CLI execution ID assignment**
 - **Ignore schema structure**
 - **Skip Phase 5 Plan Quality Check**
+- **Embed tasks[] in plan.json** (use task_ids[] referencing .task/ files)

 ---

@@ -832,7 +844,7 @@ After generating plan.json, **MUST** execute CLI quality check before returning
 | **Completeness** | All user requirements reflected in tasks | Yes |
 | **Task Granularity** | Each task 15-60 min scope | No |
 | **Dependencies** | No circular deps, correct ordering | Yes |
-| **Acceptance Criteria** | Quantified and testable (not vague) | No |
+| **Convergence Criteria** | Quantified and testable (not vague) | No |
 | **Implementation Steps** | 2+ actionable steps per task | No |
 | **Constraint Compliance** | Follows project-guidelines.json | Yes |

@@ -841,9 +853,9 @@ After generating plan.json, **MUST** execute CLI quality check before returning
 Use `ccw cli` with analysis mode to validate plan against quality dimensions:

 ```bash
-ccw cli -p "Validate plan quality: completeness, granularity, dependencies, acceptance criteria, implementation steps, constraint compliance" \
+ccw cli -p "Validate plan quality: completeness, granularity, dependencies, convergence criteria, implementation steps, constraint compliance" \
  --tool gemini --mode analysis \
-  --context "@{plan_json_path} @.workflow/project-guidelines.json"
+  --context "@{plan_json_path} @{task_dir}/*.json @.workflow/project-guidelines.json"
 ```

 **Expected Output Structure**:
@@ -855,7 +867,7 @@ ccw cli -p "Validate plan quality: completeness, granularity, dependencies, acce
 ### Result Parsing

 Parse CLI output sections using regex to extract:
- **6 Dimension Results**: Each with `passed` boolean and issue lists (missing requirements, oversized/undersized tasks, vague criteria, etc.)
+- **6 Dimension Results**: Each with `passed` boolean and issue lists (missing requirements, oversized/undersized tasks, vague convergence criteria, etc.)
 - **Summary Counts**: Critical issues, minor issues
 - **Recommendation**: `PASS` | `AUTO_FIX` | `REGENERATE`
 - **Fixes**: Optional JSON patches for auto-fixable issues
@@ -866,7 +878,7 @@ Apply automatic fixes for minor issues:

 | Issue Type | Auto-Fix Action | Example |
 |-----------|----------------|---------|
-| **Vague Acceptance** | Replace with quantified criteria | "works correctly" → "All unit tests pass with 100% success rate" |
+| **Vague Convergence** | Replace with quantified criteria | "works correctly" → "All unit tests pass with 100% success rate" |
 | **Insufficient Steps** | Expand to 4-step template | Add: Analyze → Implement → Error handling → Verify |
 | **CLI-Provided Patches** | Apply JSON patches from CLI output | Update task fields per patch specification |

@@ -876,7 +888,7 @@ After fixes, update `_metadata.quality_check` with fix log.

 After Phase 4 planObject generation:

-1. **Write Initial Plan** → `${sessionFolder}/plan.json`
+1. **Write Task Files** → `${sessionFolder}/.task/TASK-*.json` + **Write Plan** → `${sessionFolder}/plan.json`
 2. **Execute CLI Check** → Gemini (Qwen fallback)
 3. **Parse Results** → Extract recommendation and issues
 4. **Handle Recommendation**:
--- a/.claude/agents/context-search-agent.md
+++ b/.claude/agents/context-search-agent.md
@@ -165,12 +165,16 @@ if (file_exists(manifestPath)) {
 }

 // Synthesis helper functions (conceptual)
+// NOTE: relevant_files items are now structured objects:
+//   {path, relevance, rationale, role, discovery_source?, key_symbols?}
 function synthesizeCriticalFiles(allRelevantFiles) {
-  // 1. Group by path
+  // 1. Group by path (files are objects with .path property)
  // 2. Count mentions across angles
  // 3. Average relevance scores
-  // 4. Rank by: (mention_count * 0.6) + (avg_relevance * 0.4)
-  // 5. Return top 10-15 with mentioned_by_angles attribution
+  // 4. Merge rationales from different angles (join with "; ")
+  // 5. Collect unique roles and key_symbols across angles
+  // 6. Rank by: (mention_count * 0.6) + (avg_relevance * 0.4)
+  // 7. Return top 10-15 with: path, relevance, rationale, role, mentioned_by_angles, key_symbols
 }

 function synthesizeConflictIndicators(explorationData) {
@@ -544,7 +548,7 @@ Calculate risk level based on:
      }
    ],
    "aggregated_insights": {
-      "critical_files": [{"path": "src/auth/AuthService.ts", "relevance": 0.95, "mentioned_by_angles": ["architecture"]}],
+      "critical_files": [{"path": "src/auth/AuthService.ts", "relevance": 0.95, "rationale": "Contains login/register/verifyToken - core auth entry points", "role": "modify_target", "mentioned_by_angles": ["architecture"], "key_symbols": ["AuthService", "login", "verifyToken"]}],
      "conflict_indicators": [{"type": "pattern_mismatch", "description": "...", "source_angle": "architecture", "severity": "medium"}],
      "clarification_needs": [{"question": "...", "context": "...", "options": [], "source_angle": "architecture"}],
      "constraints": [{"constraint": "Must follow existing DI pattern", "source_angle": "architecture"}],
--- a/.claude/commands/team/plan.md
+++ b/.claude/commands/team/plan.md
@@ -52,7 +52,7 @@ mcp__ccw-tools__team_msg({ operation: "log", team: teamName, from: "planner", to
 mcp__ccw-tools__team_msg({ operation: "log", team: teamName, from: "planner", to: "coordinator", type: "plan_revision", summary: "已按反馈拆分task-2为两个子任务" })

 // 错误上报
-mcp__ccw-tools__team_msg({ operation: "log", team: teamName, from: "planner", to: "coordinator", type: "error", summary: "plan-json-schema.json 未找到, 使用默认结构" })
+mcp__ccw-tools__team_msg({ operation: "log", team: teamName, from: "planner", to: "coordinator", type: "error", summary: "plan-overview-base-schema.json 未找到, 使用默认结构" })
 ```

 ## Execution Process
@@ -73,10 +73,10 @@ Phase 2: Multi-Angle Exploration
   └─ Write exploration results to session folder

 Phase 3: Plan Generation
-   ├─ Read plan-json-schema.json for structure reference
+   ├─ Read plan-overview-base-schema.json + task-schema.json for structure reference
   ├─ Low complexity → Direct Claude planning
   ├─ Medium/High → cli-lite-planning-agent
-   └─ Output: plan.json
+   └─ Output: plan.json (overview with task_ids[]) + .task/TASK-*.json (independent task files)

 Phase 4: Submit for Approval
   ├─ SendMessage plan summary to coordinator
@@ -208,6 +208,12 @@ Execute **${angle}** exploration for task planning context.
 ## Expected Output
 Write JSON to: ${sessionFolder}/exploration-${angle}.json
 Follow explore-json-schema.json structure with ${angle}-focused findings.
+
+**MANDATORY**: Every file in relevant_files MUST have:
+- **rationale** (required): Specific selection basis tied to ${angle} topic (>10 chars, not generic)
+- **role** (required): modify_target|dependency|pattern_reference|test_target|type_definition|integration_point|config|context_only
+- **discovery_source** (recommended): bash-scan|cli-analysis|ace-search|dependency-trace|manual
+- **key_symbols** (recommended): Key functions/classes/types relevant to task
 `
    })
  })
@@ -232,7 +238,7 @@ Write(`${sessionFolder}/explorations-manifest.json`, JSON.stringify(explorationM

 ```javascript
 // Read schema reference
-const schema = Bash(`cat ~/.ccw/workflows/cli-templates/schemas/plan-json-schema.json`)
+const schema = Bash(`cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-base-schema.json`)

 if (complexity === 'Low') {
  // Direct Claude planning
@@ -242,18 +248,31 @@ if (complexity === 'Low') {
    // Incorporate findings into plan
  })

-  // Generate plan following schema
+  // Generate task files in .task/ directory
+  Bash(`mkdir -p ${sessionFolder}/.task`)
+
+  const tasks = [/* structured tasks with dependencies, files[].change, convergence.criteria */]
+  const taskIds = tasks.map(t => t.id)
+
+  // Write individual task files following task-schema.json
+  tasks.forEach(task => {
+    Write(`${sessionFolder}/.task/${task.id}.json`, JSON.stringify(task, null, 2))
+  })
+
+  // Generate plan overview following plan-overview-base-schema.json
  const plan = {
    summary: "...",
    approach: "...",
-    tasks: [/* structured tasks with dependencies, modification points, acceptance criteria */],
+    task_ids: taskIds,
+    task_count: taskIds.length,
    estimated_time: "...",
    recommended_execution: "Agent",
    complexity: "Low",
    _metadata: {
      timestamp: new Date().toISOString(),
      source: "team-planner",
-      planning_mode: "direct"
+      planning_mode: "direct",
+      plan_type: "feature"
    }
  }
  Write(`${sessionFolder}/plan.json`, JSON.stringify(plan, null, 2))
@@ -264,16 +283,26 @@ if (complexity === 'Low') {
    run_in_background: false,
    description: "Generate detailed implementation plan",
    prompt: `
-Generate implementation plan and write plan.json.
+Generate implementation plan with two-layer output.

 ## Output Location
 **Session Folder**: ${sessionFolder}
 **Output Files**:
 - ${sessionFolder}/planning-context.md
- ${sessionFolder}/plan.json
+- ${sessionFolder}/plan.json (overview with task_ids[])
+- ${sessionFolder}/.task/TASK-*.json (independent task files)

 ## Output Schema Reference
-Execute: cat ~/.ccw/workflows/cli-templates/schemas/plan-json-schema.json
+Execute: cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-base-schema.json
+Execute: cat ~/.ccw/workflows/cli-templates/schemas/task-schema.json
+
+## Output Format: Two-Layer Structure
+- plan.json: Overview with task_ids[] referencing .task/ files (NO tasks[] array)
+- .task/TASK-*.json: Independent task files following task-schema.json
+
+plan.json required: summary, approach, task_ids, task_count, _metadata (with plan_type)
+Task files required: id, title, description, depends_on, convergence (with criteria[])
+Task fields: files[].change (not modification_points), convergence.criteria (not acceptance), test (not verification)

 ## Task Description
 ${task.description}
@@ -286,9 +315,9 @@ Path: ${exp.path}`).join('\n\n')}
 ${complexity}

 ## Requirements
-Generate plan.json following schema. Key constraints:
- tasks: 2-7 structured tasks (group by feature/module, NOT by file)
- Each task: id, title, scope, modification_points, implementation, acceptance, depends_on
+Generate plan.json + .task/*.json following schemas. Key constraints:
+- 2-7 structured tasks (group by feature/module, NOT by file)
+- Each task file: id, title, description, files[].change, convergence.criteria, depends_on
 - Prefer parallel tasks (minimize depends_on)
 `
  })
@@ -301,6 +330,10 @@ Generate plan.json following schema. Key constraints:
 // Read generated plan
 const plan = JSON.parse(Read(`${sessionFolder}/plan.json`))

+// Load tasks from .task/ directory (two-layer format)
+const tasks = plan.task_ids.map(id => JSON.parse(Read(`${sessionFolder}/.task/${id}.json`)))
+const taskCount = plan.task_count || plan.task_ids.length
+
 // Send plan summary to coordinator
 SendMessage({
  type: "message",
@@ -309,19 +342,20 @@ SendMessage({

 **Task**: ${task.subject}
 **Complexity**: ${complexity}
-**Tasks**: ${plan.tasks.length}
+**Tasks**: ${taskCount}

 ### Task Summary
-${plan.tasks.map((t, i) => `${i+1}. ${t.title} (${t.scope || 'N/A'})`).join('\n')}
+${tasks.map((t, i) => `${i+1}. ${t.title} (${t.scope || 'N/A'})`).join('\n')}

 ### Approach
 ${plan.approach}

 ### Plan Location
 ${sessionFolder}/plan.json
+${plan.task_ids ? `Task Files: ${sessionFolder}/.task/` : ''}

 Please review and approve or request revisions.`,
-  summary: `Plan ready: ${plan.tasks.length} tasks`
+  summary: `Plan ready: ${taskCount} tasks`
 })

 // Wait for coordinator response
@@ -359,7 +393,11 @@ if (nextTasks.length > 0) {
 ├── exploration-{angle2}.json
 ├── explorations-manifest.json     # Exploration index
 ├── planning-context.md            # Evidence + understanding (Medium/High)
-└── plan.json                      # Implementation plan
+├── plan.json                      # Plan overview with task_ids[] (NO embedded tasks[])
+└── .task/                         # Independent task files
+    ├── TASK-001.json              # Task file following task-schema.json
+    ├── TASK-002.json
+    └── ...
 ```

 ## Error Handling
--- a/.claude/commands/workflow/lite-fix.md
+++ b/.claude/commands/workflow/lite-fix.md
@@ -44,7 +44,8 @@ Intelligent lightweight bug fixing command with dynamic workflow adaptation base
 | `diagnosis-{angle}.json` | Per-angle diagnosis results (1-4 files based on severity) |
 | `diagnoses-manifest.json` | Index of all diagnosis files |
 | `planning-context.md` | Evidence paths + synthesized understanding |
-| `fix-plan.json` | Structured fix plan (fix-plan-json-schema.json) |
+| `fix-plan.json` | Fix plan overview with `task_ids[]` (plan-overview-fix-schema.json) |
+| `.task/FIX-*.json` | Independent fix task files (one per task) |

 **Output Directory**: `.workflow/.lite-fix/{bug-slug}-{YYYY-MM-DD}/`

@@ -52,7 +53,7 @@ Intelligent lightweight bug fixing command with dynamic workflow adaptation base
 - Low/Medium severity → Direct Claude planning (no agent)
 - High/Critical severity → `cli-lite-planning-agent` generates `fix-plan.json`

-**Schema Reference**: `~/.ccw/workflows/cli-templates/schemas/fix-plan-json-schema.json`
+**Schema Reference**: `~/.ccw/workflows/cli-templates/schemas/plan-overview-fix-schema.json`

 ## Auto Mode Defaults

@@ -91,7 +92,7 @@ Phase 2: Clarification (optional, multi-round)

 Phase 3: Fix Planning (NO CODE EXECUTION - planning only)
   +- Decision (based on Phase 1 severity):
-      |- Low/Medium -> Load schema: cat ~/.ccw/workflows/cli-templates/schemas/fix-plan-json-schema.json -> Direct Claude planning (following schema) -> fix-plan.json -> MUST proceed to Phase 4
+      |- Low/Medium -> Load schema: cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-fix-schema.json -> Direct Claude planning (following schema) -> fix-plan.json -> MUST proceed to Phase 4
      +- High/Critical -> cli-lite-planning-agent -> fix-plan.json -> MUST proceed to Phase 4

 Phase 4: Confirmation & Selection
@@ -254,8 +255,12 @@ Execute **${angle}** diagnosis for bug root cause analysis. Analyze codebase fro
  **IMPORTANT**: Use structured format:
  \`{file: "src/module/file.ts", line_range: "45-60", issue: "Description", confidence: 0.85}\`
 - affected_files: Files involved from ${angle} perspective
-  **IMPORTANT**: Use object format with relevance scores:
-  \`[{path: "src/file.ts", relevance: 0.85, rationale: "Contains ${angle} logic"}]\`
+  **MANDATORY**: Every file MUST use structured object format with ALL required fields:
+  \`[{path: "src/file.ts", relevance: 0.85, rationale: "Contains handleLogin() at line 45 where null check is missing", change_type: "fix_target", discovery_source: "bash-scan", key_symbols: ["handleLogin"]}]\`
+  - **rationale** (required): Specific reason why this file is affected (>10 chars, not generic)
+  - **change_type** (required): fix_target|needs_update|test_coverage|reference_only
+  - **discovery_source** (recommended): bash-scan|cli-analysis|ace-search|dependency-trace|stack-trace|manual
+  - **key_symbols** (recommended): Key functions/classes related to the bug
 - reproduction_steps: Steps to reproduce the bug
 - fix_hints: Suggested fix approaches from ${angle} viewpoint
 - dependencies: Dependencies relevant to ${angle} diagnosis
@@ -268,7 +273,9 @@ Execute **${angle}** diagnosis for bug root cause analysis. Analyze codebase fro
 - [ ] Schema obtained via cat diagnosis-json-schema.json
 - [ ] get_modules_by_depth.sh executed
 - [ ] Root cause identified with confidence score
- [ ] At least 3 affected files identified with ${angle} rationale
+- [ ] At least 3 affected files identified with specific rationale + change_type
+- [ ] Every file has rationale >10 chars (not generic like "Contains ${angle} logic")
+- [ ] Every file has change_type classification (fix_target/needs_update/etc.)
 - [ ] Fix hints are actionable (specific code changes, not generic advice)
 - [ ] Reproduction steps are verifiable
 - [ ] JSON output follows schema exactly
@@ -426,21 +433,74 @@ if (autoYes) {
 **Low/Medium Severity** - Direct planning by Claude:
 ```javascript
 // Step 1: Read schema
-const schema = Bash(`cat ~/.ccw/workflows/cli-templates/schemas/fix-plan-json-schema.json`)
+const schema = Bash(`cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-fix-schema.json`)

-// Step 2: Generate fix-plan following schema (Claude directly, no agent)
-// For Medium complexity: include rationale + verification (optional, but recommended)
+// Step 2: Generate fix tasks with NEW field names (Claude directly, no agent)
+// Field mapping: modification_points -> files, acceptance -> convergence, verification -> test
+const fixTasks = [
+  {
+    id: "FIX-001",
+    title: "...",
+    description: "...",
+    scope: "...",
+    action: "Fix|Update|Refactor|Add|Delete",
+    depends_on: [],
+    convergence: {
+      criteria: ["..."]  // Quantified acceptance criteria
+    },
+    files: [
+      { path: "src/module/file.ts", action: "modify", target: "functionName", change: "Description of change" }
+    ],
+    implementation: ["Step 1: ...", "Step 2: ..."],
+    test: {
+      manual_checks: ["Reproduce issue", "Verify fix"],
+      success_metrics: ["Issue resolved", "No regressions"]
+    },
+    complexity: "Low|Medium",
+
+    // Medium severity fields (optional for Low, recommended for Medium)
+    ...(severity === "Medium" ? {
+      rationale: {
+        chosen_approach: "Direct fix approach",
+        alternatives_considered: ["Workaround", "Refactor"],
+        decision_factors: ["Minimal impact", "Quick turnaround"],
+        tradeoffs: "Doesn't address underlying issue"
+      },
+      test: {
+        unit: ["test_bug_fix_basic"],
+        integration: [],
+        manual_checks: ["Reproduce issue", "Verify fix"],
+        success_metrics: ["Issue resolved", "No regressions"]
+      }
+    } : {})
+  }
+  // ... additional tasks as needed
+]
+
+// Step 3: Write individual task files to .task/ directory
+const taskDir = `${sessionFolder}/.task`
+Bash(`mkdir -p "${taskDir}"`)
+
+fixTasks.forEach(task => {
+  Write(`${taskDir}/${task.id}.json`, JSON.stringify(task, null, 2))
+})
+
+// Step 4: Generate fix-plan overview (NO embedded tasks[])
 const fixPlan = {
  summary: "...",
-  root_cause: "...",
-  strategy: "immediate_patch|comprehensive_fix|refactor",
-  tasks: [...],  // Each task: { id, title, scope, ..., depends_on, complexity }
+  approach: "...",
+  task_ids: fixTasks.map(t => t.id),
+  task_count: fixTasks.length,
+  fix_context: {
+    root_cause: "...",
+    strategy: "immediate_patch|comprehensive_fix|refactor",
+    severity: severity,
+    risk_level: "..."
+  },
  estimated_time: "...",
  recommended_execution: "Agent",
-  severity: severity,
-  risk_level: "...",

-  // Medium complexity fields (optional for direct planning, auto-filled for Low)
+  // Medium complexity fields (optional for Low)
  ...(severity === "Medium" ? {
    design_decisions: [
      {
@@ -448,58 +508,22 @@ const fixPlan = {
        rationale: "Keeps changes localized and quick to review",
        tradeoff: "Defers comprehensive refactoring"
      }
-    ],
-    tasks_with_rationale: {
-      // Each task gets rationale if Medium
-      task_rationale_example: {
-        rationale: {
-          chosen_approach: "Direct fix approach",
-          alternatives_considered: ["Workaround", "Refactor"],
-          decision_factors: ["Minimal impact", "Quick turnaround"],
-          tradeoffs: "Doesn't address underlying issue"
-        },
-        verification: {
-          unit_tests: ["test_bug_fix_basic"],
-          integration_tests: [],
-          manual_checks: ["Reproduce issue", "Verify fix"],
-          success_metrics: ["Issue resolved", "No regressions"]
-        }
-      }
-    }
+    ]
  } : {}),

  _metadata: {
    timestamp: getUtc8ISOString(),
    source: "direct-planning",
    planning_mode: "direct",
+    plan_type: "fix",
    complexity: severity === "Medium" ? "Medium" : "Low"
  }
 }

-// Step 3: Merge task rationale into tasks array
-if (severity === "Medium") {
-  fixPlan.tasks = fixPlan.tasks.map(task => ({
-    ...task,
-    rationale: fixPlan.tasks_with_rationale[task.id]?.rationale || {
-      chosen_approach: "Standard fix",
-      alternatives_considered: [],
-      decision_factors: ["Correctness", "Simplicity"],
-      tradeoffs: "None"
-    },
-    verification: fixPlan.tasks_with_rationale[task.id]?.verification || {
-      unit_tests: [`test_${task.id}_basic`],
-      integration_tests: [],
-      manual_checks: ["Verify fix works"],
-      success_metrics: ["Test pass"]
-    }
-  }))
-  delete fixPlan.tasks_with_rationale  // Clean up temp field
-}
-
-// Step 4: Write fix-plan to session folder
+// Step 5: Write fix-plan overview to session folder
 Write(`${sessionFolder}/fix-plan.json`, JSON.stringify(fixPlan, null, 2))

-// Step 5: MUST continue to Phase 4 (Confirmation) - DO NOT execute code here
+// Step 6: MUST continue to Phase 4 (Confirmation) - DO NOT execute code here
 ```

 **High/Critical Severity** - Invoke cli-lite-planning-agent:
@@ -510,17 +534,18 @@ Task(
  run_in_background=false,
  description="Generate detailed fix plan",
  prompt=`
-Generate fix plan and write fix-plan.json.
+Generate fix plan using two-layer output format.

 ## Output Location

 **Session Folder**: ${sessionFolder}
 **Output Files**:
 - ${sessionFolder}/planning-context.md (evidence + understanding)
- ${sessionFolder}/fix-plan.json (fix plan)
+- ${sessionFolder}/fix-plan.json (fix plan overview -- NO embedded tasks[])
+- ${sessionFolder}/.task/FIX-*.json (independent fix task files, one per task)

 ## Output Schema Reference
-Execute: cat ~/.ccw/workflows/cli-templates/schemas/fix-plan-json-schema.json (get schema reference before generating plan)
+Execute: cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-fix-schema.json (get schema reference before generating plan)

 ## Project Context (MANDATORY - Read Both Files)
 1. Read: .workflow/project-tech.json (technology stack, architecture, key components)
@@ -550,20 +575,49 @@ ${JSON.stringify(clarificationContext) || "None"}
 ${severity}

 ## Requirements
-Generate fix-plan.json with:
+
+**Output Format**: Two-layer structure:
+- fix-plan.json: Overview with task_ids[] referencing .task/ files (NO tasks[] array)
+- .task/FIX-*.json: Independent task files following task-schema.json
+
+**fix-plan.json required fields**:
 - summary: 2-3 sentence overview of the fix
- root_cause: Consolidated root cause from all diagnoses
- strategy: "immediate_patch" | "comprehensive_fix" | "refactor"
- tasks: 1-5 structured fix tasks (**IMPORTANT: group by fix area, NOT by file**)
-  - **Task Granularity Principle**: Each task = one complete fix unit
-  - title: action verb + target (e.g., "Fix token validation edge case")
-  - scope: module path (src/auth/) or feature name
-  - action: "Fix" | "Update" | "Refactor" | "Add" | "Delete"
-  - description
-  - modification_points: ALL files to modify for this fix (group related changes)
-  - implementation (2-5 steps covering all modification_points)
-  - acceptance: Quantified acceptance criteria
-  - depends_on: task IDs this task depends on (use sparingly)
+- approach: Overall fix approach description
+- task_ids: Array of task IDs (e.g., ["FIX-001", "FIX-002"])
+- task_count: Number of tasks
+- fix_context:
+  - root_cause: Consolidated root cause from all diagnoses
+  - strategy: "immediate_patch" | "comprehensive_fix" | "refactor"
+  - severity: ${severity}
+  - risk_level: "Low" | "Medium" | "High"
+- estimated_time, recommended_execution
+- data_flow (High/Critical REQUIRED): How data flows through affected code
+  - diagram: "A -> B -> C" style flow
+  - stages: [{stage, input, output, component}]
+- design_decisions (High/Critical REQUIRED): Global fix decisions
+  - [{decision, rationale, tradeoff}]
+- _metadata:
+  - timestamp, source, planning_mode
+  - plan_type: "fix"
+  - complexity: "High" | "Critical"
+  - diagnosis_angles: ${JSON.stringify(manifest.diagnoses.map(d => d.angle))}
+
+**Each .task/FIX-*.json required fields**:
+- id: "FIX-001" (prefix FIX-, NOT TASK-)
+- title: action verb + target (e.g., "Fix token validation edge case")
+- description
+- scope: module path (src/auth/) or feature name
+- action: "Fix" | "Update" | "Refactor" | "Add" | "Delete"
+- depends_on: task IDs this task depends on (use sparingly)
+- convergence: { criteria: ["Quantified acceptance criterion 1", "..."] }
+- files: ALL files to modify for this fix (group related changes)
+  - [{ path: "src/file.ts", action: "modify|create|delete", target: "component/function", change: "Description of what changes" }]
+- implementation: ["Step 1: ...", "Step 2: ..."] (2-5 steps covering all files)
+- test:
+  - unit: ["test names to add/verify"]
+  - integration: ["integration test names"]
+  - manual_checks: ["manual verification steps"]
+  - success_metrics: ["quantified success criteria"]

  **High/Critical complexity fields per task** (REQUIRED):
  - rationale:
@@ -571,11 +625,6 @@ Generate fix-plan.json with:
    - alternatives_considered: Other approaches evaluated
    - decision_factors: Key factors influencing choice
    - tradeoffs: Known tradeoffs of this approach
-  - verification:
-    - unit_tests: Test names to add/verify
-    - integration_tests: Integration test names
-    - manual_checks: Manual verification steps
-    - success_metrics: Quantified success criteria
  - risks:
    - description: Risk description
    - probability: Low|Medium|High
@@ -586,18 +635,10 @@ Generate fix-plan.json with:
    - interfaces: [{name, definition, purpose}]
    - key_functions: [{signature, purpose, returns}]

-**Top-level High/Critical fields** (REQUIRED):
- data_flow: How data flows through affected code
-  - diagram: "A → B → C" style flow
-  - stages: [{stage, input, output, component}]
- design_decisions: Global fix decisions
-  - [{decision, rationale, tradeoff}]
-
- estimated_time, recommended_execution, severity, risk_level
- _metadata:
-  - timestamp, source, planning_mode
-  - complexity: "High" | "Critical"
-  - diagnosis_angles: ${JSON.stringify(manifest.diagnoses.map(d => d.angle))}
+**Field name rules** (do NOT use old names):
+- files[].change (NOT modification_points)
+- convergence.criteria (NOT acceptance)
+- test (NOT verification at task level)

 ## Task Grouping Rules
 1. **Group by fix area**: All changes for one fix = one task (even if 2-3 files)
@@ -605,25 +646,30 @@ Generate fix-plan.json with:
 3. **Substantial tasks**: Each task should represent 10-45 minutes of work
 4. **True dependencies only**: Only use depends_on when Task B cannot start without Task A's output
 5. **Prefer parallel**: Most tasks should be independent (no depends_on)
+6. **Task IDs**: Use FIX-001, FIX-002 prefix (NOT TASK-)

 ## Execution
 1. Read ALL diagnosis files for comprehensive context
 2. Execute CLI planning using Gemini (Qwen fallback) with --rule planning-fix-strategy template
 3. Synthesize findings from multiple diagnosis angles
-4. Generate fix-plan with:
-   - For High/Critical: REQUIRED new fields (rationale, verification, risks, code_skeleton, data_flow, design_decisions)
-   - Each task MUST have rationale (why this fix), verification (how to verify success), and risks (potential issues)
-5. Parse output and structure fix-plan
+4. Generate fix tasks (1-5 tasks):
+   - Each task file written to \`${sessionFolder}/.task/FIX-NNN.json\`
+   - For High/Critical: REQUIRED fields (rationale, test, risks, code_skeleton)
+5. Generate fix-plan overview:
+   - Written to \`${sessionFolder}/fix-plan.json\`
+   - Contains task_ids[] referencing .task/ files (NO embedded tasks[])
+   - For High/Critical: REQUIRED fields (data_flow, design_decisions)
 6. **Write**: \`${sessionFolder}/planning-context.md\` (evidence paths + understanding)
-7. **Write**: \`${sessionFolder}/fix-plan.json\`
-8. Return brief completion summary
+7. **Write**: \`${sessionFolder}/.task/FIX-*.json\` (individual task files)
+8. **Write**: \`${sessionFolder}/fix-plan.json\` (plan overview with task_ids[])
+9. Return brief completion summary

 ## Output Format for CLI
 Include these sections in your fix-plan output:
- Summary, Root Cause, Strategy (existing)
+- Summary, Root Cause (in fix_context), Strategy (existing)
 - Data Flow: Diagram showing affected code paths
 - Design Decisions: Key architectural choices in the fix
- Tasks: Each with rationale (Medium/High), verification (Medium/High), risks (High), code_skeleton (High)
+- Task files: Each with convergence, files, test, rationale (High), risks (High), code_skeleton (High)
 `
 )
 ```
@@ -638,18 +684,26 @@ Include these sections in your fix-plan output:
 ```javascript
 const fixPlan = JSON.parse(Read(`${sessionFolder}/fix-plan.json`))

+// Load tasks from .task/ directory (two-layer format)
+const tasks = (fixPlan.task_ids || []).map(id => {
+  return JSON.parse(Read(`${sessionFolder}/.task/${id}.json`))
+})
+const taskList = tasks
+
+const fixContext = fixPlan.fix_context || {}
+
 console.log(`
 ## Fix Plan

 **Summary**: ${fixPlan.summary}
-**Root Cause**: ${fixPlan.root_cause}
-**Strategy**: ${fixPlan.strategy}
+**Root Cause**: ${fixContext.root_cause || fixPlan.root_cause}
+**Strategy**: ${fixContext.strategy || fixPlan.strategy}

-**Tasks** (${fixPlan.tasks.length}):
-${fixPlan.tasks.map((t, i) => `${i+1}. ${t.title} (${t.scope})`).join('\n')}
+**Tasks** (${taskList.length}):
+${taskList.map((t, i) => `${i+1}. ${t.title} (${t.scope})`).join('\n')}

-**Severity**: ${fixPlan.severity}
-**Risk Level**: ${fixPlan.risk_level}
+**Severity**: ${fixContext.severity || fixPlan.severity}
+**Risk Level**: ${fixContext.risk_level || fixPlan.risk_level}
 **Estimated Time**: ${fixPlan.estimated_time}
 **Recommended**: ${fixPlan.recommended_execution}
 `)
@@ -679,7 +733,7 @@ if (autoYes) {
  userSelection = AskUserQuestion({
    questions: [
      {
-        question: `Confirm fix plan? (${fixPlan.tasks.length} tasks, ${fixPlan.severity} severity)`,
+        question: `Confirm fix plan? (${taskList.length} tasks, ${fixContext.severity || fixPlan.severity} severity)`,
        header: "Confirm",
        multiSelect: false,
        options: [
@@ -695,7 +749,7 @@ if (autoYes) {
        options: [
          { label: "Agent", description: "@code-developer agent" },
          { label: "Codex", description: "codex CLI tool" },
-          { label: "Auto", description: `Auto: ${fixPlan.severity === 'Low' ? 'Agent' : 'Codex'}` }
+          { label: "Auto", description: `Auto: ${(fixContext.severity || fixPlan.severity) === 'Low' ? 'Agent' : 'Codex'}` }
        ]
      },
      {
@@ -734,14 +788,21 @@ manifest.diagnoses.forEach(diag => {

 const fixPlan = JSON.parse(Read(`${sessionFolder}/fix-plan.json`))

+const fixSeverity = fixPlan.fix_context?.severity || fixPlan.severity
+
 executionContext = {
  mode: "bugfix",
-  severity: fixPlan.severity,
+  severity: fixSeverity,
  planObject: {
    ...fixPlan,
    // Ensure complexity is set based on severity for new field consumption
-    complexity: fixPlan.complexity || (fixPlan.severity === 'Critical' ? 'High' : (fixPlan.severity === 'High' ? 'High' : 'Medium'))
+    complexity: fixPlan.complexity || (fixSeverity === 'Critical' ? 'High' : (fixSeverity === 'High' ? 'High' : 'Medium'))
  },
+  // Task files from .task/ directory (two-layer format)
+  taskFiles: (fixPlan.task_ids || []).map(id => ({
+    id,
+    path: `${sessionFolder}/.task/${id}.json`
+  })),
  diagnosisContext: diagnoses,
  diagnosisAngles: manifest.diagnoses.map(d => d.angle),
  diagnosisManifest: manifest,
@@ -758,7 +819,8 @@ executionContext = {
        path: diag.path
      })),
      diagnoses_manifest: `${sessionFolder}/diagnoses-manifest.json`,
-      fix_plan: `${sessionFolder}/fix-plan.json`
+      fix_plan: `${sessionFolder}/fix-plan.json`,
+      task_dir: `${sessionFolder}/.task`
    }
  }
 }
@@ -780,7 +842,11 @@ Skill(skill="workflow:lite-execute", args="--in-memory --mode bugfix")
 ├── diagnosis-{angle4}.json      # Diagnosis angle 4 (if applicable)
 ├── diagnoses-manifest.json      # Diagnosis index
 ├── planning-context.md          # Evidence + understanding
-└── fix-plan.json                # Fix plan
+├── fix-plan.json                # Fix plan overview (task_ids[], NO embedded tasks[])
+└── .task/                       # Independent fix task files
+    ├── FIX-001.json             # Fix task 1
+    ├── FIX-002.json             # Fix task 2
+    └── ...                      # Additional fix tasks
 ```

 **Example**:
@@ -791,7 +857,10 @@ Skill(skill="workflow:lite-execute", args="--in-memory --mode bugfix")
 ├── diagnosis-validation.json
 ├── diagnoses-manifest.json
 ├── planning-context.md
-└── fix-plan.json
+├── fix-plan.json
+└── .task/
+    ├── FIX-001.json
+    └── FIX-002.json
 ```

 ## Error Handling
--- a/.claude/commands/workflow/lite-plan.md
+++ b/.claude/commands/workflow/lite-plan.md
@@ -44,7 +44,8 @@ Intelligent lightweight planning command with dynamic workflow adaptation based
 | `exploration-{angle}.json` | Per-angle exploration results (1-4 files based on complexity) |
 | `explorations-manifest.json` | Index of all exploration files |
 | `planning-context.md` | Evidence paths + synthesized understanding |
-| `plan.json` | Structured implementation plan (plan-json-schema.json) |
+| `plan.json` | Plan overview with task_ids[] (plan-overview-base-schema.json) |
+| `.task/TASK-*.json` | Independent task files (one per task) |

 **Output Directory**: `.workflow/.lite-plan/{task-slug}-{YYYY-MM-DD}/`

@@ -52,7 +53,7 @@ Intelligent lightweight planning command with dynamic workflow adaptation based
 - Low complexity → Direct Claude planning (no agent)
 - Medium/High complexity → `cli-lite-planning-agent` generates `plan.json`

-**Schema Reference**: `~/.ccw/workflows/cli-templates/schemas/plan-json-schema.json`
+**Schema Reference**: `~/.ccw/workflows/cli-templates/schemas/plan-overview-base-schema.json`

 ## Auto Mode Defaults

@@ -89,7 +90,7 @@ Phase 2: Clarification (optional, multi-round)

 Phase 3: Planning (NO CODE EXECUTION - planning only)
   └─ Decision (based on Phase 1 complexity):
-      ├─ Low → Load schema: cat ~/.ccw/workflows/cli-templates/schemas/plan-json-schema.json → Direct Claude planning (following schema) → plan.json
+      ├─ Low → Load schema: cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-base-schema.json → Direct Claude planning (following schema) → plan.json
      └─ Medium/High → cli-lite-planning-agent → plan.json (agent internally executes quality check)

 Phase 4: Confirmation & Selection
@@ -251,9 +252,13 @@ Execute **${angle}** exploration for task planning context. Analyze codebase fro
 **Required Fields** (all ${angle} focused):
 - project_structure: Modules/architecture relevant to ${angle}
 - relevant_files: Files affected from ${angle} perspective
-  **IMPORTANT**: Use object format with relevance scores for synthesis:
-  \`[{path: "src/file.ts", relevance: 0.85, rationale: "Core ${angle} logic"}]\`
-  Scores: 0.7+ high priority, 0.5-0.7 medium, <0.5 low
+  **MANDATORY**: Every file MUST use structured object format with ALL required fields:
+  \`[{path: "src/file.ts", relevance: 0.85, rationale: "Contains AuthService.login() - entry point for JWT token generation", role: "modify_target", discovery_source: "bash-scan", key_symbols: ["AuthService", "login"]}]\`
+  - **rationale** (required): Specific selection basis tied to ${angle} topic (>10 chars, not generic)
+  - **role** (required): modify_target|dependency|pattern_reference|test_target|type_definition|integration_point|config|context_only
+  - **discovery_source** (recommended): bash-scan|cli-analysis|ace-search|dependency-trace|manual
+  - **key_symbols** (recommended): Key functions/classes/types in the file relevant to the task
+  - Scores: 0.7+ high priority, 0.5-0.7 medium, <0.5 low
 - patterns: ${angle}-related patterns to follow
 - dependencies: Dependencies relevant to ${angle}
 - integration_points: Where to integrate from ${angle} viewpoint (include file:line locations)
@@ -264,7 +269,9 @@ Execute **${angle}** exploration for task planning context. Analyze codebase fro
 ## Success Criteria
 - [ ] Schema obtained via cat explore-json-schema.json
 - [ ] get_modules_by_depth.sh executed
- [ ] At least 3 relevant files identified with ${angle} rationale
+- [ ] At least 3 relevant files identified with specific rationale + role
+- [ ] Every file has rationale >10 chars (not generic like "Related to ${angle}")
+- [ ] Every file has role classification (modify_target/dependency/etc.)
 - [ ] Patterns are actionable (code examples, not generic advice)
 - [ ] Integration points include file:line locations
 - [ ] Constraints are project-specific to ${angle}
@@ -416,7 +423,11 @@ if (autoYes) {
 // 2. 默认 → agent

 const executorAssignments = {}  // { taskId: { executor: 'gemini'|'codex'|'agent', reason: string } }
-plan.tasks.forEach(task => {
+
+// Load tasks from .task/ directory for executor assignment
+const taskFiles = Glob(`${sessionFolder}/.task/TASK-*.json`)
+taskFiles.forEach(taskPath => {
+  const task = JSON.parse(Read(taskPath))
  // Claude 根据上述规则语义分析，为每个 task 分配 executor
  executorAssignments[task.id] = { executor: '...', reason: '...' }
 })
@@ -425,7 +436,7 @@ plan.tasks.forEach(task => {
 **Low Complexity** - Direct planning by Claude:
 ```javascript
 // Step 1: Read schema
-const schema = Bash(`cat ~/.ccw/workflows/cli-templates/schemas/plan-json-schema.json`)
+const schema = Bash(`cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-base-schema.json`)

 // Step 2: ⚠️ MANDATORY - Read and review ALL exploration files
 const manifest = JSON.parse(Read(`${sessionFolder}/explorations-manifest.json`))
@@ -434,22 +445,51 @@ manifest.explorations.forEach(exp => {
  console.log(`\n### Exploration: ${exp.angle}\n${explorationData}`)
 })

-// Step 3: Generate plan following schema (Claude directly, no agent)
-// ⚠️ Plan MUST incorporate insights from exploration files read in Step 2
+// Step 3: Generate task objects (Claude directly, no agent)
+// ⚠️ Tasks MUST incorporate insights from exploration files read in Step 2
+// Task fields use NEW names: convergence.criteria (not acceptance), files[].change (not modification_points), test (not verification)
+const tasks = [
+  {
+    id: "TASK-001",
+    title: "...",
+    description: "...",
+    depends_on: [],
+    convergence: { criteria: ["..."] },
+    files: [{ path: "...", change: "..." }],
+    implementation: ["..."],
+    test: "..."
+  },
+  // ... more tasks
+]
+
+// Step 4: Write task files to .task/ directory
+const taskDir = `${sessionFolder}/.task`
+Bash(`mkdir -p "${taskDir}"`)
+tasks.forEach(task => {
+  Write(`${taskDir}/${task.id}.json`, JSON.stringify(task, null, 2))
+})
+
+// Step 5: Generate plan overview (NO embedded tasks[])
 const plan = {
  summary: "...",
  approach: "...",
-  tasks: [...],  // Each task: { id, title, scope, ..., depends_on, execution_group, complexity }
+  task_ids: tasks.map(t => t.id),
+  task_count: tasks.length,
+  complexity: "Low",
  estimated_time: "...",
  recommended_execution: "Agent",
-  complexity: "Low",
-  _metadata: { timestamp: getUtc8ISOString(), source: "direct-planning", planning_mode: "direct" }
+  _metadata: {
+    timestamp: getUtc8ISOString(),
+    source: "direct-planning",
+    planning_mode: "direct",
+    plan_type: "feature"
+  }
 }

-// Step 4: Write plan to session folder
+// Step 6: Write plan overview to session folder
 Write(`${sessionFolder}/plan.json`, JSON.stringify(plan, null, 2))

-// Step 5: MUST continue to Phase 4 (Confirmation) - DO NOT execute code here
+// Step 7: MUST continue to Phase 4 (Confirmation) - DO NOT execute code here
 ```

 **Medium/High Complexity** - Invoke cli-lite-planning-agent:
@@ -467,10 +507,11 @@ Generate implementation plan and write plan.json.
 **Session Folder**: ${sessionFolder}
 **Output Files**:
 - ${sessionFolder}/planning-context.md (evidence + understanding)
- ${sessionFolder}/plan.json (implementation plan)
+- ${sessionFolder}/plan.json (plan overview -- NO embedded tasks[])
+- ${sessionFolder}/.task/TASK-*.json (independent task files, one per task)

 ## Output Schema Reference
-Execute: cat ~/.ccw/workflows/cli-templates/schemas/plan-json-schema.json (get schema reference before generating plan)
+Execute: cat ~/.ccw/workflows/cli-templates/schemas/plan-overview-base-schema.json (get schema reference before generating plan)

 ## Project Context (MANDATORY - Read Both Files)
 1. Read: .workflow/project-tech.json (technology stack, architecture, key components)
@@ -500,10 +541,17 @@ ${JSON.stringify(clarificationContext) || "None"}
 ${complexity}

 ## Requirements
-Generate plan.json following the schema obtained above. Key constraints:
- tasks: 2-7 structured tasks (**group by feature/module, NOT by file**)
+Generate plan.json and .task/*.json following the schema obtained above. Key constraints:
 - _metadata.exploration_angles: ${JSON.stringify(manifest.explorations.map(e => e.angle))}

+**Output Format**: Two-layer structure:
+- plan.json: Overview with task_ids[] referencing .task/ files (NO tasks[] array)
+- .task/TASK-*.json: Independent task files following task-schema.json
+
+plan.json required fields: summary, approach, task_ids, task_count, _metadata (with plan_type: "feature")
+Each task file required fields: id, title, description, depends_on, convergence (with criteria[])
+Task fields use: files[].change (not modification_points), convergence.criteria (not acceptance), test (not verification)
+
 ## Task Grouping Rules
 1. **Group by feature**: All changes for one feature = one task (even if 3-5 files)
 2. **Group by context**: Tasks with similar context or related functional changes can be grouped together
@@ -517,10 +565,12 @@ Generate plan.json following the schema obtained above. Key constraints:
 1. Read schema file (cat command above)
 2. Execute CLI planning using Gemini (Qwen fallback)
 3. Read ALL exploration files for comprehensive context
-4. Synthesize findings and generate plan following schema
+4. Synthesize findings and generate tasks + plan overview
 5. **Write**: \`${sessionFolder}/planning-context.md\` (evidence paths + understanding)
-6. **Write**: \`${sessionFolder}/plan.json\`
-7. Return brief completion summary
+6. **Create**: \`${sessionFolder}/.task/\` directory (mkdir -p)
+7. **Write**: \`${sessionFolder}/.task/TASK-001.json\`, \`TASK-002.json\`, etc. (one per task)
+8. **Write**: \`${sessionFolder}/plan.json\` (overview with task_ids[], NO tasks[])
+9. Return brief completion summary
 `
 )
 ```
@@ -535,14 +585,21 @@ Generate plan.json following the schema obtained above. Key constraints:
 ```javascript
 const plan = JSON.parse(Read(`${sessionFolder}/plan.json`))

+// Load tasks from .task/ directory
+const tasks = (plan.task_ids || []).map(id => {
+  const taskPath = `${sessionFolder}/.task/${id}.json`
+  return JSON.parse(Read(taskPath))
+})
+const taskList = tasks
+
 console.log(`
 ## Implementation Plan

 **Summary**: ${plan.summary}
 **Approach**: ${plan.approach}

-**Tasks** (${plan.tasks.length}):
-${plan.tasks.map((t, i) => `${i+1}. ${t.title} (${t.file})`).join('\n')}
+**Tasks** (${taskList.length}):
+${taskList.map((t, i) => `${i+1}. ${t.title} (${t.scope || t.files?.[0]?.path || ''})`).join('\n')}

 **Complexity**: ${plan.complexity}
 **Estimated Time**: ${plan.estimated_time}
@@ -575,7 +632,7 @@ if (autoYes) {
  userSelection = AskUserQuestion({
    questions: [
      {
-        question: `Confirm plan? (${plan.tasks.length} tasks, ${plan.complexity})`,
+        question: `Confirm plan? (${taskList.length} tasks, ${plan.complexity})`,
        header: "Confirm",
        multiSelect: false,
        options: [
@@ -632,7 +689,11 @@ manifest.explorations.forEach(exp => {
 const plan = JSON.parse(Read(`${sessionFolder}/plan.json`))

 executionContext = {
-  planObject: plan,
+  planObject: plan,  // plan overview (no tasks[])
+  taskFiles: (plan.task_ids || []).map(id => ({
+    id,
+    path: `${sessionFolder}/.task/${id}.json`
+  })),
  explorationsContext: explorations,
  explorationAngles: manifest.explorations.map(e => e.angle),
  explorationManifest: manifest,
@@ -653,7 +714,8 @@ executionContext = {
        path: exp.path
      })),
      explorations_manifest: `${sessionFolder}/explorations-manifest.json`,
-      plan: `${sessionFolder}/plan.json`
+      plan: `${sessionFolder}/plan.json`,
+      task_dir: `${sessionFolder}/.task`
    }
  }
 }
@@ -674,7 +736,12 @@ Skill(skill="workflow:lite-execute", args="--in-memory")
 ├── exploration-{angle3}.json      # Exploration angle 3 (if applicable)
 ├── exploration-{angle4}.json      # Exploration angle 4 (if applicable)
 ├── explorations-manifest.json     # Exploration index
-└── plan.json                      # Implementation plan
+├── planning-context.md            # Evidence paths + understanding
+├── plan.json                      # Plan overview (task_ids[])
+└── .task/                         # Task files directory
+    ├── TASK-001.json
+    ├── TASK-002.json
+    └── ...
 ```

 **Example**:
@@ -684,7 +751,12 @@ Skill(skill="workflow:lite-execute", args="--in-memory")
 ├── exploration-auth-patterns.json
 ├── exploration-security.json
 ├── explorations-manifest.json
-└── plan.json
+├── planning-context.md
+├── plan.json
+└── .task/
+    ├── TASK-001.json
+    ├── TASK-002.json
+    └── TASK-003.json
 ```

 ## Error Handling
--- a/.claude/commands/workflow/tools/context-gather.md
+++ b/.claude/commands/workflow/tools/context-gather.md
@@ -164,9 +164,13 @@ Execute **${angle}** exploration for task planning context. Analyze codebase fro
 **Required Fields** (all ${angle} focused):
 - project_structure: Modules/architecture relevant to ${angle}
 - relevant_files: Files affected from ${angle} perspective
-  **IMPORTANT**: Use object format with relevance scores for synthesis:
-  \`[{path: "src/file.ts", relevance: 0.85, rationale: "Core ${angle} logic"}]\`
-  Scores: 0.7+ high priority, 0.5-0.7 medium, <0.5 low
+  **MANDATORY**: Every file MUST use structured object format with ALL required fields:
+  \`[{path: "src/file.ts", relevance: 0.85, rationale: "Contains AuthService.login() - entry point for JWT token generation", role: "modify_target", discovery_source: "bash-scan", key_symbols: ["AuthService", "login"]}]\`
+  - **rationale** (required): Specific selection basis tied to ${angle} topic (>10 chars, not generic)
+  - **role** (required): modify_target|dependency|pattern_reference|test_target|type_definition|integration_point|config|context_only
+  - **discovery_source** (recommended): bash-scan|cli-analysis|ace-search|dependency-trace|manual
+  - **key_symbols** (recommended): Key functions/classes/types in the file relevant to the task
+  - Scores: 0.7+ high priority, 0.5-0.7 medium, <0.5 low
 - patterns: ${angle}-related patterns to follow
 - dependencies: Dependencies relevant to ${angle}
 - integration_points: Where to integrate from ${angle} viewpoint (include file:line locations)