feat: Generate and inject coding style guide from CLAUDE.md

CLAUDE.md

@@ -109,6 +109,7 @@ uv tool install claude-task-master --force --reinstall
 ├── state.json            # Machine state
 ├── progress.md           # Progress summary
 ├── context.md            # Accumulated learnings
+├── coding-style.md       # Coding style guide (generated from CLAUDE.md)
 ├── mailbox.json          # Pending messages for plan updates
 ├── logs/
 │   └── run-*.txt         # Last 10 logs kept
@@ -119,7 +120,7 @@ uv tool install claude-task-master --force --reinstall
 
 ## Exit Codes
 
-- **0 (Success)**: Tasks done, cleanup all except logs/, keep last 10 logs
+- **0 (Success)**: Tasks done, cleanup all except logs/ and coding-style.md, keep last 10 logs
 - **1 (Blocked)**: Need intervention, keep everything for resume
 - **2 (Interrupted)**: Ctrl+C, keep everything for resume
 
@@ -177,11 +178,22 @@ All commands check `state_manager.exists()` first:
 - Uses `PlanUpdater` to integrate change request into existing plan
 - Preserves completed tasks, modifies pending tasks as needed
 
+### Coding Style Generation
+- Before planning, generates `coding-style.md` if it doesn't exist
+- Analyzes `CLAUDE.md` and convention files to extract:
+  - Development workflow (TDD, test-first patterns)
+  - Code style conventions (naming, formatting)
+  - Project-specific requirements
+- Concise guide (~600 words) injected into planning and work prompts
+- Preserved across runs (not deleted on success) to save tokens
+- Uses Opus for high-quality extraction
+
 ### Planning Prompt
 - Instructs Claude to add `.claude-task-master/` to .gitignore
 - Use Read, Glob, Grep to explore codebase
 - Create task list with checkboxes
 - Define success criteria
+- Includes coding style guide for task planning
 
 ## Testing
 

src/claude_task_master/core/agent.py

@@ -170,15 +170,35 @@ def _import_sdk(self) -> None:
                 ValueError("query must be callable"),
             )
 
-    def run_planning_phase(self, goal: str, context: str = "") -> dict[str, Any]:
+    def run_planning_phase(
+        self, goal: str, context: str = "", coding_style: str | None = None
+    ) -> dict[str, Any]:
         """Run planning phase with read-only tools.
 
         Always uses Opus (smartest model) for planning to ensure
         high-quality task breakdown and complexity classification.
 
+        Args:
+            goal: The goal to plan for.
+            context: Additional context for planning.
+            coding_style: Optional coding style guide to inject into prompt.
+
+        Delegates to AgentPhaseExecutor for implementation.
+        """
+        return self._phase_executor.run_planning_phase(goal, context, coding_style)
+
+    def generate_coding_style(self) -> dict[str, Any]:
+        """Generate a coding style guide by analyzing the codebase.
+
+        Analyzes CLAUDE.md, convention files, and codebase to create a
+        concise coding style guide with workflow and conventions.
+
+        Returns:
+            Dict with 'coding_style' and 'raw_output' keys.
+
         Delegates to AgentPhaseExecutor for implementation.
         """
-        return self._phase_executor.run_planning_phase(goal, context)
+        return self._phase_executor.generate_coding_style()
 
     def run_work_session(
         self,
@@ -190,6 +210,7 @@ def run_work_session(
         create_pr: bool = True,
         pr_group_info: dict | None = None,
         target_branch: str = "main",
+        coding_style: str | None = None,
     ) -> dict[str, Any]:
         """Run a work session with full tools.
 
@@ -203,6 +224,7 @@ def run_work_session(
             create_pr: If True, instruct agent to create PR. If False, commit only.
             pr_group_info: Optional dict with PR group context (name, completed_tasks, etc).
             target_branch: The target branch for rebasing (default: "main").
+            coding_style: Optional coding style guide to inject into prompt.
 
         Returns:
             Dict with 'output', 'success', and 'model_used' keys.
@@ -218,6 +240,7 @@ def run_work_session(
             create_pr=create_pr,
             pr_group_info=pr_group_info,
             target_branch=target_branch,
+            coding_style=coding_style,
         )
 
     def verify_success_criteria(self, criteria: str, context: str = "") -> dict[str, Any]:

src/claude_task_master/core/agent_phases.py

@@ -13,7 +13,13 @@
 
 from . import console
 from .agent_models import ModelType, get_tools_for_phase
-from .prompts import build_planning_prompt, build_verification_prompt, build_work_prompt
+from .prompts import (
+    build_coding_style_prompt,
+    build_planning_prompt,
+    build_verification_prompt,
+    build_work_prompt,
+    extract_coding_style,
+)
 
 if TYPE_CHECKING:
     from .agent_query import AgentQueryExecutor
@@ -104,7 +110,9 @@ def __init__(
         self.get_agents_func = get_agents_func
         self.process_message_func = process_message_func
 
-    def run_planning_phase(self, goal: str, context: str = "") -> dict[str, Any]:
+    def run_planning_phase(
+        self, goal: str, context: str = "", coding_style: str | None = None
+    ) -> dict[str, Any]:
         """Run planning phase with read-only tools.
 
         Always uses Opus (smartest model) for planning to ensure
@@ -113,12 +121,17 @@ def run_planning_phase(self, goal: str, context: str = "") -> dict[str, Any]:
         Args:
             goal: The goal to plan for.
             context: Additional context for planning.
+            coding_style: Optional coding style guide to inject into prompt.
 
         Returns:
             Dict with 'plan', 'criteria', and 'raw_output' keys.
         """
         # Build prompt for planning
-        prompt = build_planning_prompt(goal=goal, context=context if context else None)
+        prompt = build_planning_prompt(
+            goal=goal,
+            context=context if context else None,
+            coding_style=coding_style,
+        )
 
         # Always use Opus for planning (smartest model)
         console.info("Planning with Opus (smartest model)...")
@@ -152,6 +165,7 @@ def run_work_session(
         create_pr: bool = True,
         pr_group_info: dict | None = None,
         target_branch: str = "main",
+        coding_style: str | None = None,
     ) -> dict[str, Any]:
         """Run a work session with full tools.
 
@@ -165,6 +179,7 @@ def run_work_session(
             create_pr: If True, instruct agent to create PR. If False, commit only.
             pr_group_info: Optional dict with PR group context (name, completed_tasks, etc).
             target_branch: The target branch for rebasing (default: "main").
+            coding_style: Optional coding style guide to inject into prompt.
 
         Returns:
             Dict with 'output', 'success', and 'model_used' keys.
@@ -178,6 +193,7 @@ def run_work_session(
             create_pr=create_pr,
             pr_group_info=pr_group_info,
             target_branch=target_branch,
+            coding_style=coding_style,
         )
 
         # Run async query with optional model override
@@ -328,3 +344,37 @@ def _extract_criteria(self, result: str) -> str:
 
         # Default criteria if none specified
         return "All tasks in the task list are completed successfully."
+
+    def generate_coding_style(self) -> dict[str, Any]:
+        """Generate a coding style guide by analyzing the codebase.
+
+        Analyzes CLAUDE.md, convention files, and sample source files
+        to create a concise coding style guide.
+
+        Returns:
+            Dict with 'coding_style' and 'raw_output' keys.
+        """
+        # Build prompt for coding style generation
+        prompt = build_coding_style_prompt()
+
+        console.info("Generating coding style guide with Opus...")
+
+        # Run with planning tools (read-only) and Opus for quality
+        result = run_async_with_cleanup(
+            self.query_executor.run_query(
+                prompt=prompt,
+                tools=self.get_tools_for_phase("planning"),
+                model_override=ModelType.OPUS,  # Use Opus for quality
+                get_model_name_func=self.get_model_name_func,
+                get_agents_func=self.get_agents_func,
+                process_message_func=self.process_message_func,
+            )
+        )
+
+        # Extract the coding style content
+        coding_style = extract_coding_style(result)
+
+        return {
+            "coding_style": coding_style,
+            "raw_output": result,
+        }

src/claude_task_master/core/orchestrator.py

@@ -1337,11 +1337,15 @@ def _run_verification_fix(self, verification_details: str, state: TaskState) ->
 After completing your fixes, end with: TASK COMPLETE"""
 
         try:
+            # Load coding style for consistent style in fix session
+            coding_style = self.state_manager.load_coding_style()
+
             self.agent.run_work_session(
                 task_description=task_description,
                 context=context,
                 model_override=ModelType.OPUS,
                 create_pr=True,
+                coding_style=coding_style,
             )
             state.session_count += 1
             self.state_manager.save_state(state)

src/claude_task_master/core/planner.py

@@ -2,6 +2,7 @@
 
 from typing import Any
 
+from . import console
 from .agent import AgentWrapper
 from .state import StateManager
 
@@ -14,13 +15,50 @@ def __init__(self, agent: AgentWrapper, state_manager: StateManager):
         self.agent = agent
         self.state_manager = state_manager
 
+    def ensure_coding_style(self) -> str | None:
+        """Ensure coding style guide exists, generating it if needed.
+
+        Checks if coding-style.md exists. If not, generates it by analyzing
+        CLAUDE.md and convention files in the codebase.
+
+        Returns:
+            The coding style guide content, or None if generation failed.
+        """
+        # Check if coding style already exists
+        coding_style = self.state_manager.load_coding_style()
+        if coding_style:
+            console.info("Using existing coding style guide")
+            return coding_style
+
+        # Generate coding style by analyzing codebase
+        console.info("Generating coding style guide from codebase...")
+        result = self.agent.generate_coding_style()
+
+        coding_style_content: str = result.get("coding_style", "")
+        if coding_style_content:
+            self.state_manager.save_coding_style(coding_style_content)
+            console.success("Coding style guide generated and saved")
+            return coding_style_content
+
+        console.warning("Could not generate coding style guide")
+        return None
+
     def create_plan(self, goal: str) -> dict[str, Any]:
-        """Create initial task plan using read-only tools."""
+        """Create initial task plan using read-only tools.
+
+        First generates coding style guide if it doesn't exist, then
+        runs planning phase with the coding style injected.
+        """
+        # Ensure coding style exists (generate if needed)
+        coding_style = self.ensure_coding_style()
+
         # Load any existing context
         context = self.state_manager.load_context()
 
-        # Run planning phase with Claude
-        result = self.agent.run_planning_phase(goal=goal, context=context)
+        # Run planning phase with Claude (with coding style)
+        result = self.agent.run_planning_phase(
+            goal=goal, context=context, coding_style=coding_style
+        )
 
         # Extract plan and criteria from result
         plan = result.get("plan", "")

src/claude_task_master/core/prompts.py

@@ -1,6 +1,7 @@
 """Prompt Templates - Centralized, maintainable prompt generation.
 
 This module provides structured prompt templates for different agent phases:
+- Coding Style: Generate concise coding style guide from codebase analysis
 - Planning: Initial codebase analysis and task creation
 - Plan Update: Modifying existing plans based on change requests
 - Working: Task execution with verification
@@ -12,6 +13,7 @@
 This module re-exports all prompt functions for backward compatibility.
 The actual implementations are in:
 - prompts_base.py: PromptSection, PromptBuilder
+- prompts_coding_style.py: build_coding_style_prompt, extract_coding_style
 - prompts_planning.py: build_planning_prompt
 - prompts_plan_update.py: build_plan_update_prompt
 - prompts_working.py: build_work_prompt
@@ -24,6 +26,9 @@
 # Re-export base classes
 from .prompts_base import PromptBuilder, PromptSection
 
+# Re-export coding style prompts
+from .prompts_coding_style import build_coding_style_prompt, extract_coding_style
+
 # Re-export planning prompts
 from .prompts_plan_update import build_plan_update_prompt
 from .prompts_planning import build_planning_prompt
@@ -43,6 +48,9 @@
     # Base classes
     "PromptSection",
     "PromptBuilder",
+    # Coding Style
+    "build_coding_style_prompt",
+    "extract_coding_style",
     # Planning
     "build_planning_prompt",
     "build_plan_update_prompt",