feat: Complete v1.2 system enhancements and agent coordination improvements

- Update action-planning-agent to use gemini-wrapper for improved CLI analysis
- Enhance task execution with simplified control structure and context requirements
- Improve plan-deep command with input validation and clarity requirements
- Add intelligent context acquisition rules to CLAUDE.md with required analysis patterns
- Strengthen agent workflow coordination with TodoWrite management and context rules
- Remove deprecated execution controls and streamline task execution flow

System now enforces proper context gathering before implementation and provides
better coordination between agents through structured TODO management.

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

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

.claude/agents/action-planning-agent.md

@@ -32,7 +32,7 @@ You are a pure execution agent specialized in creating actionable implementation
 1. Parse input requirements and extract control flags
 2. IF DEEP_ANALYSIS_REQUIRED flag present:
      → Check for analysis method markers:
-       - [GEMINI_CLI_REQUIRED] → Execute comprehensive Gemini CLI analysis
+       - [GEMINI_CLI_REQUIRED] → Execute comprehensive Gemini CLI analysis using gemini-wrapper
        - [CODEX_CLI_REQUIRED] → Execute autonomous Codex CLI analysis
      → Use analysis results for planning context
 3. Assess task complexity (simple/medium/complex)  

.claude/commands/task/execute.md

@@ -126,12 +126,6 @@ FUNCTION on_execution_failure(checkpoint):
 END FUNCTION
 ```
 
-### ✨ **Advanced Execution Controls**
-
--   **Dry Run (`--dry-run`)**: Simulates execution, showing the agent, estimated time, and files affected without making changes.
--   **Custom Checkpoints (`--checkpoints="..."`)**: Overrides the default checkpoints with a custom, comma-separated list (e.g., `"design,implement,deploy"`).
--   **Conditional Execution (`--if="..."`)**: Proceeds with execution only if a specified condition (e.g., `"tests-pass"`) is met.
--   **Rollback (`--rollback`)**: Reverts file modifications and restores the previous task state.
 
 ### 📄 **Simplified Context Structure (JSON)**
 

.claude/commands/workflow/plan-deep.md

@@ -47,12 +47,18 @@ Creates comprehensive implementation plans through deep codebase analysis using
 ### 1. Input Processing
 ```
 Input Analysis:
+├── Validate input clarity (reject vague descriptions)
 ├── Parse task description or file
 ├── Extract key technical terms
 ├── Identify potential affected domains
 └── Prepare context for agent
 ```
 
+**Clarity Requirements**:
+- **Minimum specificity**: Must include clear technical goal and affected components
+- **Auto-rejection**: Vague inputs like "optimize system", "refactor code", "improve performance" without context
+- **Response**: `❌ Input too vague. Deep planning requires specific technical objectives and component scope.`
+
 ### 2. Agent Invocation with Deep Analysis Flag
 The command invokes action-planning-agent with special parameters that **enforce** Gemini CLI analysis.
 
@@ -86,7 +92,7 @@ Task(action-planning-agent):
     - Skip session inheritance (standalone planning)
     - Force GEMINI_CLI_REQUIRED flag = true
     - Set analysis_source = "gemini" (深度分析固定值)
-    - Generate hierarchical task decomposition
+    - Generate hierarchical task decomposition (max 2 levels: impl-N.M)
     - Create detailed IMPL_PLAN.md with subtasks
     - Generate TODO_LIST.md for tracking
     
@@ -102,7 +108,7 @@ Task(action-planning-agent):
 ### 4. Output Generation (by Agent)
 The action-planning-agent generates in `.workflow/WFS-[session-id]/`:
 - **IMPL_PLAN.md** - Hierarchical implementation plan with stages
-- **TODO_LIST.md** - Task tracking checklist (if complexity > simple)
+- **TODO_LIST.md** - Unified hierarchical task tracking with ▸ container tasks and indented subtasks
 - **.task/*.json** - Task definitions for complex projects
 - **workflow-session.json** - Session tracking
 - **gemini-analysis.md** - Consolidated Gemini analysis results
@@ -145,6 +151,11 @@ def process_plan_deep_command(input):
 
 ### Common Issues and Solutions
 
+**Input Processing Errors**
+- **Vague text input**: Auto-reject without guidance
+  - Rejected examples: "optimize system", "refactor code", "make it faster", "improve architecture"
+  - Response: Direct rejection message, no further assistance
+
 **Agent Execution Errors**
 - Verify action-planning-agent availability
 - Check for context size limits

.claude/output-styles/agent-workflow-coordination.md

@@ -8,9 +8,11 @@ description: Core coordination principles for multi-agent development workflows
 
 **Purpose**: Thorough upfront planning reduces risk, improves quality, and prevents costly rework.
 
+
 ### TodoWrite Coordination Rules
 
 1.  **TodoWrite FIRST**: Always create TodoWrite entries *before* complex task execution begins.
+2.  **Context Before Implementation**: Context gathering must complete before implementation tasks begin.
 3.  **Agent Coordination**: Each agent is responsible for updating the status of its assigned todo.
 4.  **Progress Visibility**: Provides clear workflow state visibility to stakeholders.