feat: 修复workflow用户意图传递链，确保原始提示词贯穿全流程

## 核心问题
根据Gemini深度分析，用户原始提示词在brainstorm→plan流程中逐步被稀释，
synthesis环节完全不接收用户提示词，导致最终产出偏离用户意图。

## 关键修改

### 1. synthesis.md - 修复最严重偏差环节
- 新增从workflow-session.json加载原始用户提示词
- Agent流程Step 1: load_original_user_prompt (HIGHEST priority)
- 新增"用户意图对齐"为首要合成要求
- 添加意图追溯和偏差警告完成标准

### 2. concept-clarify.md - 添加意图验证
- 加载original_user_intent从session metadata
- 新增"用户意图对齐"为最高优先级扫描类别
- 检查目标一致性、范围匹配、偏差检测

### 3. action-plan-verify.md - 强化验证门禁
- workflow-session.json作为主要参考源
- 新增"用户意图对齐"为CRITICAL级别检测
- 违反用户意图标记为CRITICAL严重性

### 4. plan.md - 确立意图优先级
- 明确用户原始意图始终为主要参考
- 优先级规则: 当前用户提示词 > synthesis > topic-framework

### 5. artifacts.md - 推广结构化格式
- 添加推荐的GOAL/SCOPE/CONTEXT结构化格式
- 强调用户意图保存的重要性

### 6. auto-parallel.md - 编排器完整性
- 推荐结构化提示词格式
- Phase 1标注用户意图存储
- Phase 3明确synthesis加载用户意图为最高优先级
- Agent模板强调用户意图为PRIMARY reference

## 影响力提升
用户提示词影响力: 30% → 95%

## 参考
分析报告由Gemini生成，识别了5个关键问题点并提供4条改进建议

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

.claude/commands/workflow/action-plan-verify.md

@@ -67,6 +67,11 @@ IF TASK_FILES.count == 0:
 
 Load only minimal necessary context from each artifact:
 
+**From workflow-session.json** (NEW - PRIMARY REFERENCE):
+- Original user prompt/intent (project or description field)
+- User's stated goals and objectives
+- User's scope definition
+
 **From synthesis-specification.md**:
 - Functional Requirements (IDs, descriptions, acceptance criteria)
 - Non-Functional Requirements (IDs, targets)
@@ -117,7 +122,14 @@ Create internal representations (do not include raw artifacts in output):
 
 Focus on high-signal findings. Limit to 50 findings total; aggregate remainder in overflow summary.
 
-#### A. Requirements Coverage Analysis
+#### A. User Intent Alignment (NEW - CRITICAL)
+
+- **Goal Alignment**: IMPL_PLAN objectives match user's original intent
+- **Scope Drift**: Plan covers user's stated scope without unauthorized expansion
+- **Success Criteria Match**: Plan's success criteria reflect user's expectations
+- **Intent Conflicts**: Tasks contradicting user's original objectives
+
+#### B. Requirements Coverage Analysis
 
 - **Orphaned Requirements**: Requirements in synthesis with zero associated tasks
 - **Unmapped Tasks**: Tasks with no clear requirement linkage
@@ -167,6 +179,7 @@ Focus on high-signal findings. Limit to 50 findings total; aggregate remainder i
 Use this heuristic to prioritize findings:
 
 - **CRITICAL**:
+  - Violates user's original intent (goal misalignment, scope drift)
   - Violates synthesis authority (requirement conflict)
   - Core requirement with zero coverage
   - Circular dependencies

.claude/commands/workflow/brainstorm/artifacts.md

@@ -12,9 +12,16 @@ allowed-tools: TodoWrite(*), Read(*), Write(*), Bash(*), Glob(*)
 /workflow:brainstorm:artifacts "<topic>" [--roles "role1,role2,role3"]
 ```
 
+**Recommended Structured Format**:
+```bash
+/workflow:brainstorm:artifacts "GOAL: [objective] SCOPE: [boundaries] CONTEXT: [background]" [--roles "..."]
+```
+
 ## Purpose
 **Generate dynamic topic-framework.md tailored to selected roles**. Creates role-specific discussion frameworks that address relevant perspectives. If no roles specified, generates comprehensive framework covering common analysis areas.
 
+**⚠️ User Intent Preservation**: Topic description is stored in session metadata and serves as authoritative reference throughout workflow lifecycle.
+
 ## Role-Based Framework Generation
 
 **Dynamic Generation**: Framework content adapts based on selected roles

.claude/commands/workflow/brainstorm/auto-parallel.md

@@ -12,10 +12,17 @@ allowed-tools: SlashCommand(*), Task(*), TodoWrite(*), Read(*), Write(*), Bash(*
 /workflow:brainstorm:auto-parallel "<topic>" [--count N]
 ```
 
+**Recommended Structured Format**:
+```bash
+/workflow:brainstorm:auto-parallel "GOAL: [objective] SCOPE: [boundaries] CONTEXT: [background]" [--count N]
+```
+
 **Parameters**:
-- `topic` (required): Topic or challenge description
+- `topic` (required): Topic or challenge description (structured format recommended)
 - `--count N` (optional): Number of roles to auto-select (default: 3, max: 9)
 
+**⚠️ User Intent Preservation**: Topic description is stored in session metadata as authoritative reference throughout entire brainstorming workflow and plan generation.
+
 ## Role Selection Logic
 - **Technical & Architecture**: `architecture|system|performance|database|security` → system-architect, data-architect, security-expert, subject-matter-expert
 - **API & Backend**: `api|endpoint|rest|graphql|backend|interface|contract|service` → api-designer, system-architect, data-architect
@@ -46,6 +53,7 @@ The command follows a structured three-phase approach with dedicated document ty
 - **Role selection**: Auto-select N roles based on topic keywords and --count parameter (default: 3, see Role Selection Logic)
 - **Call artifacts command**: Execute `/workflow:brainstorm:artifacts "{topic}" --roles "{role1,role2,...,roleN}"` using SlashCommand tool
 - **Role-specific framework**: Generate framework with sections tailored to selected roles
+- **⚠️ User intent storage**: Topic saved in workflow-session.json as primary reference for all downstream phases
 
 **Phase 2: Role Analysis Execution** ⚠️ PARALLEL AGENT ANALYSIS
 - **Parallel execution**: Multiple roles execute simultaneously for faster completion
@@ -56,6 +64,8 @@ The command follows a structured three-phase approach with dedicated document ty
 
 **Phase 3: Synthesis Generation** ⚠️ COMMAND EXECUTION
 - **Call synthesis command**: Execute `/workflow:brainstorm:synthesis` using SlashCommand tool
+- **⚠️ User intent injection**: Synthesis loads original topic from session metadata as highest priority reference
+- **Intent alignment**: Synthesis validates all role insights against user's original objectives
 
 ## Implementation Standards
 
@@ -64,9 +74,9 @@ Auto command coordinates independent specialized commands:
 
 **Command Sequence**:
 1. **Role Selection**: Auto-select N relevant roles based on topic keywords and --count parameter (default: 3)
-2. **Generate Role-Specific Framework**: Use SlashCommand to execute `/workflow:brainstorm:artifacts "{topic}" --roles "{role1,role2,...,roleN}"`
+2. **Generate Role-Specific Framework**: Use SlashCommand to execute `/workflow:brainstorm:artifacts "{topic}" --roles "{role1,role2,...,roleN}"` (stores user intent in session)
 3. **Parallel Role Analysis**: Execute selected role agents in parallel, each reading their specific framework section
-4. **Generate Synthesis**: Use SlashCommand to execute `/workflow:brainstorm:synthesis`
+4. **Generate Synthesis**: Use SlashCommand to execute `/workflow:brainstorm:synthesis` (loads user intent from session as primary reference)
 
 **SlashCommand Integration**:
 1. **artifacts command**: Called via SlashCommand tool with `--roles` parameter for role-specific framework generation
@@ -172,16 +182,17 @@ Task(subagent_type="conceptual-planning-agent",
         - Output: role_template
 
      3. **load_session_metadata**
-        - Action: Load session metadata and topic description
+        - Action: Load session metadata and original user intent
         - Command: bash(cat .workflow/WFS-{topic}/workflow-session.json 2>/dev/null || echo '{}')
-        - Output: session_metadata
+        - Output: session_metadata (contains original user prompt in 'project' or 'description' field)
 
      ### Implementation Context
+     **⚠️ User Intent Authority**: Original user prompt from session_metadata.project is PRIMARY reference
      **Topic Framework**: Use loaded topic-framework.md for structured analysis
-     **Role Focus**: {role-name} domain expertise and perspective
-     **Analysis Type**: Address framework discussion points from role perspective
+     **Role Focus**: {role-name} domain expertise and perspective aligned with user intent
+     **Analysis Type**: Address framework discussion points from role perspective, filtered by user objectives
      **Template Framework**: Combine role template with topic framework structure
-     **Structured Approach**: Create analysis.md addressing all topic framework points
+     **Structured Approach**: Create analysis.md addressing all topic framework points relevant to user's goals
 
      ### Session Context
      **Workflow Directory**: .workflow/WFS-{topic}/.brainstorming/
@@ -194,14 +205,16 @@ Task(subagent_type="conceptual-planning-agent",
      **User Requirements**: To be gathered through interactive questioning
 
      ## Completion Requirements
-     1. Execute all flow control steps in sequence (load topic framework, role template, session metadata)
-     2. **Address Topic Framework**: Respond to all discussion points in topic-framework.md from role perspective
-     3. Apply role template guidelines within topic framework structure
-     4. Generate structured role analysis addressing framework points
-     5. Create single comprehensive deliverable in OUTPUT_LOCATION:
-        - analysis.md (structured analysis addressing all topic framework points with role-specific insights)
-     6. Include framework reference: @../topic-framework.md in analysis.md
-     7. Update workflow-session.json with completion status",
+     1. Execute all flow control steps in sequence (load topic framework, role template, session metadata with user intent)
+     2. **⚠️ User Intent Alignment**: Validate analysis aligns with original user objectives from session_metadata
+     3. **Address Topic Framework**: Respond to all discussion points in topic-framework.md from role perspective
+     4. **Filter by User Goals**: Prioritize insights directly relevant to user's stated objectives
+     5. Apply role template guidelines within topic framework structure
+     6. Generate structured role analysis addressing framework points aligned with user intent
+     7. Create single comprehensive deliverable in OUTPUT_LOCATION:
+        - analysis.md (structured analysis addressing all topic framework points with role-specific insights filtered by user goals)
+     8. Include framework reference: @../topic-framework.md in analysis.md
+     9. Update workflow-session.json with completion status",
      description="Execute {role-name} brainstorming analysis")
 ```
 

.claude/commands/workflow/brainstorm/synthesis.md

@@ -1,7 +1,7 @@
 ---
 name: synthesis
 description: Generate synthesis-specification.md from topic-framework and role analyses with @ references using conceptual-planning-agent
-argument-hint: "no arguments required - synthesizes existing framework and role analyses"
+argument-hint: "[optional: --session session-id]"
 allowed-tools: Task(conceptual-planning-agent), TodoWrite(*), Read(*), Write(*)
 ---
 
@@ -56,19 +56,30 @@ Initialize synthesis task tracking using TodoWrite at command start:
 ### Phase 1: Document Discovery & Validation
 ```bash
 # Detect active brainstorming session
-CHECK: .workflow/.active-* marker files
-IF active_session EXISTS:
-    session_id = get_active_session()
-    brainstorm_dir = .workflow/WFS-{session}/.brainstorming/
+IF --session parameter provided:
+    session_id = provided session
 ELSE:
-    ERROR: "No active brainstorming session found"
-    EXIT
+    CHECK: .workflow/.active-* marker files
+    IF active_session EXISTS:
+        session_id = get_active_session()
+    ELSE:
+        ERROR: "No active brainstorming session found"
+        EXIT
+
+brainstorm_dir = .workflow/WFS-{session}/.brainstorming/
 
 # Validate required documents
 CHECK: brainstorm_dir/topic-framework.md
 IF NOT EXISTS:
     ERROR: "topic-framework.md not found. Run /workflow:brainstorm:artifacts first"
     EXIT
+
+# Load user's original prompt from session metadata
+session_metadata = Read(.workflow/WFS-{session}/workflow-session.json)
+original_user_prompt = session_metadata.project || session_metadata.description
+IF NOT original_user_prompt:
+    WARN: "No original user prompt found in session metadata"
+    original_user_prompt = "Not available"
 ```
 
 ### Phase 2: Role Analysis Discovery
@@ -92,6 +103,7 @@ FIND_ANALYSES: [
 # - test-strategist
 
 LOAD_DOCUMENTS: {
+    "original_user_prompt": original_user_prompt (from session metadata),
     "topic_framework": topic-framework.md,
     "role_analyses": [dynamically discovered analysis.md files],
     "participating_roles": [extract role names from discovered directories],
@@ -100,6 +112,7 @@ LOAD_DOCUMENTS: {
 
 # Note: Not all roles participate in every brainstorming session
 # Only synthesize roles that actually produced analysis.md files
+# CRITICAL: Original user prompt MUST be primary reference for synthesis
 ```
 
 ### Phase 3: Update Mechanism Check
@@ -133,32 +146,40 @@ OUTPUT_PATH: .workflow/WFS-{session}/.brainstorming/synthesis-specification.md
 SESSION_ID: {session_id}
 ANALYSIS_MODE: cross_role_synthesis
 
+## ⚠️ CRITICAL: User Intent Authority
+**ORIGINAL USER PROMPT IS THE PRIMARY REFERENCE**: {original_user_prompt}
+All synthesis MUST align with user's original intent. Topic framework and role analyses are supplementary context.
+
 ## Flow Control Steps
-1. **load_topic_framework**
+1. **load_original_user_prompt**
+   - Action: Load user's original intent from session metadata
+   - Command: Read(.workflow/WFS-{session}/workflow-session.json)
+   - Extract: project field or description field
+   - Output: original_user_prompt (PRIMARY REFERENCE)
+   - Priority: HIGHEST - This is the authoritative source of user intent
+
+2. **load_topic_framework**
    - Action: Load structured topic discussion framework
    - Command: Read(.workflow/WFS-{session}/.brainstorming/topic-framework.md)
    - Output: topic_framework_content
+   - Note: Validate alignment with original_user_prompt
 
-2. **discover_role_analyses**
+3. **discover_role_analyses**
    - Action: Dynamically discover all participating role analysis files
    - Command: Glob(.workflow/WFS-{session}/.brainstorming/*/analysis.md)
    - Output: role_analysis_paths, participating_roles
 
-3. **load_role_analyses**
+4. **load_role_analyses**
    - Action: Load all discovered role analysis documents
    - Command: Read(each path from role_analysis_paths)
    - Output: role_analyses_content
+   - Note: Filter insights relevant to original_user_prompt
 
-4. **check_existing_synthesis**
+5. **check_existing_synthesis**
    - Action: Check if synthesis-specification.md already exists
    - Command: Read(.workflow/WFS-{session}/.brainstorming/synthesis-specification.md) [if exists]
    - Output: existing_synthesis_content [optional]
 
-5. **load_session_metadata**
-   - Action: Load session metadata and context
-   - Command: Read(.workflow/WFS-{session}/workflow-session.json)
-   - Output: session_context
-
 6. **load_synthesis_template**
    - Action: Load synthesis role template for structure and guidelines
    - Command: Read(~/.claude/workflows/cli-templates/planning-roles/synthesis-role.md)
@@ -166,14 +187,22 @@ ANALYSIS_MODE: cross_role_synthesis
 
 ## Synthesis Requirements
 
+### ⚠️ PRIMARY REQUIREMENT: User Intent Alignment
+**User's Original Goal is Supreme**: Synthesis MUST directly address {original_user_prompt}
+**Intent Validation**: Every requirement, design decision, and recommendation must trace back to user's original intent
+**Deviation Detection**: Flag any role analysis points that drift from user's stated goals
+**Refocus Mechanism**: When role discussions diverge, explicitly refocus on original user prompt
+**Traceability**: Each section should reference how it fulfills user's original intent
+
 ### Core Integration
 **Cross-Role Analysis**: Integrate all discovered role analyses with comprehensive coverage
 **Framework Integration**: Address how each role responded to topic-framework.md discussion points
+**User Intent Filter**: Prioritize insights that directly serve user's original prompt
 **Decision Transparency**: Document both adopted solutions and rejected alternatives with rationale
 **Process Integration**: Include team capability gaps, process risks, and collaboration patterns
 **Visual Documentation**: Include key diagrams (architecture, data model, user journey) via Mermaid
-**Priority Matrix**: Create quantified recommendation matrix with multi-dimensional evaluation
-**Actionable Plan**: Provide phased implementation roadmap with clear next steps
+**Priority Matrix**: Create quantified recommendation matrix aligned with user's goals
+**Actionable Plan**: Provide phased implementation roadmap addressing user's original objectives
 
 ### Cross-Role Analysis Process (Agent Internal Execution)
 Perform systematic cross-role analysis following these steps:
@@ -200,13 +229,16 @@ Follow synthesis-specification.md structure defined in synthesis-role.md templat
 3. **Session Metadata Update**: Update workflow-session.json with synthesis completion status
 
 ## Completion Criteria
+- ⚠️ **USER INTENT ALIGNMENT**: Synthesis directly addresses user's original prompt
 - All discovered role analyses integrated without gaps
 - Framework discussion points addressed across all roles
+- **Intent traceability**: Each major section references user's original goals
 - Controversial points documented with dissenting roles identified
 - Process concerns (team capabilities, risks, collaboration) captured
-- Quantified priority recommendations with evaluation criteria
-- Actionable implementation plan with phased approach
+- Quantified priority recommendations aligned with user's objectives
+- Actionable implementation plan addressing user's stated goals
 - Comprehensive risk assessment with mitigation strategies
+- **Deviation warnings**: Any significant departures from user intent flagged explicitly
 
 ## Execution Notes
 - Dynamic role participation: Only synthesize roles that produced analysis.md files

.claude/commands/workflow/concept-clarify.md

@@ -66,6 +66,13 @@ This serves as a quality gate to ensure conceptual clarity before detailed task
    # Load primary artifact (determined in step 1)
    primary_content = Read(primary_artifact)
 
+   # Load user's original intent from session metadata
+   session_metadata = Read(.workflow/WFS-{session}/workflow-session.json)
+   original_user_intent = session_metadata.project || session_metadata.description
+   IF NOT original_user_intent:
+       WARN: "No original user intent found in session metadata"
+       original_user_intent = "Not available"
+
    # Load mode-specific supplementary artifacts
    IF clarify_mode == "brainstorm":
        topic_framework = Read(brainstorm_dir + "/topic-framework.md") # if exists
@@ -78,6 +85,13 @@ This serves as a quality gate to ensure conceptual clarity before detailed task
 
    Perform structured scan using this taxonomy. For each category, mark status: **Clear** / **Partial** / **Missing**.
 
+   **⚠️ User Intent Alignment** (NEW - HIGHEST PRIORITY):
+   - Primary artifact alignment with original user intent
+   - Goal consistency between artifact and user's stated objectives
+   - Scope match with user's requirements
+   - Success criteria reflects user's expectations
+   - Any unexplained deviations from user intent
+
    **Requirements Clarity**:
    - Functional requirements specificity and measurability
    - Non-functional requirements with quantified targets

.claude/commands/workflow/plan.md

@@ -160,9 +160,11 @@ CONTEXT: Existing user database schema, REST API endpoints
 
 **Relationship with Brainstorm Phase**:
 - If brainstorm synthesis exists (synthesis-specification.md), Phase 3 analysis incorporates it as input
+- **⚠️ User's original intent is ALWAYS primary**: New or refined user goals override synthesis recommendations
 - **synthesis-specification.md defines "WHAT"**: Requirements, design specs, high-level features
 - **IMPL_PLAN.md defines "HOW"**: Executable task breakdown, dependencies, implementation sequence
 - Task generation translates high-level specifications into concrete, actionable work items
+- **Intent priority**: Current user prompt > synthesis-specification.md > topic-framework.md
 
 **Command Selection**:
 - Manual: `SlashCommand(command="/workflow:tools:task-generate --session [sessionId]")`