Replace plan mode with TodoList in co ai workflow

[wu-changxing]

Problem

The current co ai agent workflow forces plan mode (enter_plan_mode → write_plan → exit_plan_and_implement) every time a user wants to create an agent. This is over-engineered for simple agents where the tools are obvious.

Plan mode has dedicated tools, READ-ONLY restrictions, a YAML spec format, and a rigid enter/exit workflow — all for something that a TodoList can handle more naturally.

Proposal

Replace plan mode with the existing TodoList tool for agent creation workflow:

• Simple agents: co create → edit agent.py directly. No planning needed.
• Complex agents: Use todo to break down the work into steps, then execute them. The user can see progress and provide feedback naturally.

TodoList is more flexible — it doesn't lock the agent into a read-only mode, doesn't require a rigid YAML format, and lets the agent work incrementally.

Changes needed

• Remove enter_plan_mode, exit_plan_and_implement, write_plan tools from co ai
• Remove tools/enter_plan_mode.md prompt
• Remove system-reminders/plan_mode.md reminder
• Update workflow.md to use TodoList-based approach
• Update main.md references to plan mode
• Remove plan mode plugin from agent.py tool list

Context

Part of the system prompt simplification effort — the base prompt was already reduced from ~25,800 to ~12,186 chars by making workflow/index/examples load on-demand via intent detection.

[wu-changxing]

🔍 Summary

Replace the rigid plan mode workflow (enter_plan_mode → write_plan → exit_plan_and_implement) with the existing TodoList tool. Plan mode forces READ-ONLY restrictions, a YAML spec format, and a dedicated enter/exit ceremony — all of which TodoList handles more naturally and flexibly. This removes 3 tools, 2 prompt files, 1 system reminder, and simplifies the agent creation workflow.

📊 Workflow: Current vs Proposed

CURRENT (Over-engineered):
                                                       
User: "build me an inbox agent"                        
       │                                               
       ▼                                               
┌──────────────────┐                                   
│  detect_intent() │ → is_build: true                  
└────────┬─────────┘                                   
         │ loads workflow.md + agent.md reminder        
         ▼                                             
┌──────────────────┐   ┌─────────────────────────┐     
│ enter_plan_mode()│──▶│ session.mode = 'plan'   │     
└────────┬─────────┘   │ .co/PLAN.md created     │     
         │             │ WebSocket: mode_changed  │     
         │             │ ❌ READ-ONLY restriction │     
         │             └─────────────────────────┘     
         ▼                                             
┌──────────────────┐                                   
│  write_plan()    │ ← YAML spec format required       
└────────┬─────────┘                                   
         ▼                                             
┌──────────────────────────┐                           
│ exit_plan_and_implement()│ ← io.send(plan_review)    
│                          │   io.receive() blocks     
│                          │   mode restored           
└────────┬─────────────────┘                           
         ▼                                             
┌──────────────────┐                                   
│ co create + edit │ ← Finally can write code          
└──────────────────┘                                   
                                                       
   3 tools, YAML format, mode switching, io blocking   
   ────────────────────────────────────────────────── 


PROPOSED (Simple):
                                                       
User: "build me an inbox agent"                        
       │                                               
       ▼                                               
┌──────────────────┐                                   
│  detect_intent() │ → is_build: true                  
└────────┬─────────┘                                   
         │ loads workflow.md (updated)                  
         ▼                                             
┌──────────────────┐                                   
│ todo.add() tasks │ ← Break work into steps           
│  1. Design tools │   (if complex enough)             
│  2. co create    │                                   
│  3. Edit agent   │                                   
│  4. Test         │                                   
└────────┬─────────┘                                   
         │                                             
         ▼                                             
┌──────────────────┐                                   
│ ask_user() if    │ ← Only when design choices exist  
│ design unclear   │                                   
└────────┬─────────┘                                   
         ▼                                             
┌──────────────────┐                                   
│ co create + edit │ ← Start building immediately      
└──────────────────┘                                   
                                                       
   0 new tools, no mode switching, natural flow        

📁 Files to Change

connectonion/cli/co_ai/
├── agent.py                                    ← MODIFY: Remove plan mode imports + tools
├── tools/
│   ├── __init__.py                             ← MODIFY: Remove plan mode exports
│   └── plan_mode.py                            ← DELETE: Entire file
├── prompts/
│   ├── main.md                                 ← (no change — no plan mode refs)
│   ├── workflow.md                             ← MODIFY: Replace plan mode section with TodoList approach
│   ├── tools/
│   │   └── enter_plan_mode.md                  ← DELETE: Plan mode tool doc
│   └── system-reminders/
│       ├── agent.md                            ← MODIFY: Remove "enter plan mode" instruction
│       └── plan_mode.md                        ← DELETE: Plan mode reminder
├── commands/
│   └── help.py                                 ← MODIFY: Remove plan mode from help text

🔄 Data Flow: What Changes Where

agent.py (3 deletions):
  ┌─────────────────────────────────────────────────────────┐
  │ REMOVE from import:                                     │
  │   enter_plan_mode, exit_plan_and_implement, write_plan  │
  │                                                         │
  │ REMOVE from tools list:                                 │
  │   enter_plan_mode,                                      │
  │   exit_plan_and_implement,                              │
  │   write_plan,                                           │
  │                                                         │
  │ KEEP:  todo (TodoList)  ← already in tools list         │
  └─────────────────────────────────────────────────────────┘

tools/__init__.py (3 deletions):
  ┌─────────────────────────────────────────────────────────┐
  │ REMOVE import line:                                     │
  │   from ...plan_mode import enter_plan_mode, ...         │
  │                                                         │
  │ REMOVE from __all__:                                    │
  │   "enter_plan_mode", "exit_plan_and_implement",         │
  │   "write_plan"                                          │
  │                                                         │
  │ REMOVE docstring references to planning tools           │
  └─────────────────────────────────────────────────────────┘

workflow.md (rewrite §Plan Mode → §Complex Tasks):
  ┌─────────────────────────────────────────────────────────┐
  │ BEFORE:                                                 │
  │   "Always use workflow (plan mode + co create) for..."  │
  │   "Building new agents → always start with plan mode"   │
  │                                                         │
  │ AFTER:                                                  │
  │   "For complex agents: use todo to break down steps"    │
  │   "Use ask_user for design choices"                     │
  │   "For simple agents: co create → edit directly"        │
  └─────────────────────────────────────────────────────────┘

agent.md reminder (update instruction):
  ┌─────────────────────────────────────────────────────────┐
  │ BEFORE:                                                 │
  │   "enter plan mode → design spec → get approval →      │
  │    co create → edit agent.py"                           │
  │   "Start with enter_plan_mode() to design the spec"    │
  │                                                         │
  │ AFTER:                                                  │
  │   "Break work into steps with todo → ask_user for      │
  │    design choices → co create → edit agent.py"          │
  └─────────────────────────────────────────────────────────┘

help.py (1-line update):
  ┌─────────────────────────────────────────────────────────┐
  │ REMOVE line 70:                                         │
  │   **Plan Mode:** `enter_plan_mode`, ...                 │
  └─────────────────────────────────────────────────────────┘

📝 Implementation Steps

1. Delete plan mode files (3 files)

   • Delete connectonion/cli/co_ai/tools/plan_mode.py
   • Delete connectonion/cli/co_ai/prompts/tools/enter_plan_mode.md
   • Delete connectonion/cli/co_ai/prompts/system-reminders/plan_mode.md

2. Remove plan mode from agent.py

   • Remove enter_plan_mode, exit_plan_and_implement, write_plan from import at line 40
   • Remove lines 72-74 (the 3 plan mode tools) from the tools list
   • Update docstring header (lines 10, 2) to remove planning references

3. Remove plan mode from tools/init.py

   • Remove the import line (line 64): from ...plan_mode import ...
   • Remove from __all__ (lines 79-81)
   • Remove docstring references to planning tools (lines 8, 14, 38-41)

4. Update workflow.md

   • Replace lines 54-56 ("Always use workflow (plan mode + co create)") with TodoList-based approach
   • Replace lines 98-108 (§Plan Mode section) with guidance on using todo + ask_user for complex agent creation
   • Keep co create as the scaffolding command (that doesn't change)

5. Update agent.md system reminder

   • Line 7: Change "enter plan mode → design spec → get approval" to "use todo for steps → ask_user for design choices"
   • Line 22: Change "Start with enter_plan_mode()" to "Break the work into steps with todo, ask design questions with ask_user"

6. Update help.py

   • Line 70: Remove **Plan Mode:** \enter_plan_mode`, `exit_plan_and_implement`` line

⚠️ Risks & Considerations

• Frontend plan_review handling: The frontend (oo-chat) may have UI for plan_review WebSocket messages and mode_changed events. After this change, those messages will never be sent. The frontend code can be cleaned up separately — it won't break, just becomes dead code.
• Session mode state: The agent.current_session['mode'] field ('safe', 'plan', 'accept_edits') is used by plan mode only. After removal, mode will always be 'safe' (or whatever the default is). If tool_approval plugin checks mode, verify it doesn't depend on 'plan' state. (Checked: no plugin references plan mode.)
• No behavioral regression: The todo tool already exists in the tools list and already has a prompt doc. No new code needed.

🧪 Testing Strategy

• Verify prompt assembly: Run python tests/cli/show_co_ai_prompt.py and confirm no plan mode references in assembled prompt
• Verify agent creation: Run oo "create an agent that lists files" and confirm it uses todo + ask_user flow instead of plan mode
• Verify tools list: Check agent.tools doesn't include plan mode tools
• Search for orphan references: grep -r "plan_mode\|enter_plan\|write_plan\|exit_plan" connectonion/ should return nothing after changes