planning-from-research

name	planning-from-research
description	Generate execution-ready implementation plans from research documents. STRICTLY planning only—NO execution, NO coding. Validate research completeness first; if vague, resolve via subagents or user confirmation. Output is a standalone plan that a junior developer can execute without additional documentation.
allowed-tools	Read Write Bash Agent

Planning From Research

Transform a research document into a concrete, executable implementation plan.

CRITICAL BOUNDARIES

Activity	Status
Planning	✅ This skill
Research	❌ Use codebase-research skill first
Execution/Coding	❌ NEVER in this skill

This skill is STRICTLY for planning. Do not write code, do not implement, do not execute tasks. Produce a document that someone else (or a future session) can execute from.

Purpose

Convert the structural understanding from codebase-research into an actionable work plan. This skill bridges the gap between "what exists" (research) and "what to build" (execution).

When to Use

Use this skill when:

• A research document exists in memory-bank/research/
• The goal is to create an implementation plan for development work
• The output must be clear enough for a junior developer to execute independently

Workflow

Step 1: Load Research Document

Locate and read the full research document:

• If a path is provided: Read from memory-bank/research/<path>
• If a topic is provided: Find the latest matching file in memory-bank/research/

Extract:

• Scope and constraints
• Key files and their purposes
• Identified patterns and dependencies
• Any questions or TODOs noted during research

Step 2: Validate Completeness

CRITICAL: Verify the research document contains no blocking vagueness before planning.

Check for these red flags:

Flag	Example	Action Required
Open questions	"Should we use X or Y?"	Must be resolved
Uncertainty markers	"maybe", "perhaps", "consider"	Must be converted to decisions
Incomplete items	"TODO", "FIXME", "investigate"	Must be completed or deferred
Competing approaches	Multiple options without selection	Must choose ONE
Missing scope boundaries	Unclear what is in/out of scope	Must be defined
Absent acceptance criteria	No definition of "done"	Must be specified

If Blocking Issues Found

Option A: Deploy Subagents (if available)

Deploy up to 3 subagents in parallel to resolve open questions:

1. context-synthesis: Analyze open questions; determine which block planning
2. codebase-analyzer: Investigate technical unknowns; provide concrete recommendations
3. codebase-locator: Find pattern examples; show how similar problems were solved

Option B: Confirm with User (if subagents unavailable)

Present open questions to the user and request explicit answers before proceeding.

Option C: Follow Workflow in Order

If neither subagents nor user confirmation is viable, document the open questions as assumptions in the plan with clear risk flags, then proceed. The plan must note: "These items were unresolved at planning time and may require re-planning."

DO NOT proceed to Step 3 until blocking questions are resolved, documented as assumptions, or explicitly deferred.

Step 3: Check for Drift

Capture current repository state:

git rev-parse --short HEAD 2>/dev/null || echo "not-a-repo"
git status --porcelain 2>/dev/null || echo "clean"

If code changed significantly since research:

• Add a "Drift Note" to the plan
• Mark affected assumptions/tasks for re-verification

Step 4: Generate Plan

Create memory-bank/plan/YYYY-MM-DD_HH-MM-SS_<topic>.md with this structure:

---
title: "<topic> – Plan"
phase: Plan
date: "YYYY-MM-DD HH:MM:SS"
owner: "<agent_or_user>"
parent_research: "memory-bank/research/<file>.md"
git_commit_at_plan: "<sha_or_not-a-repo>"
tags: [plan, <topic>]
---

## Goal

**Singular Focus**: One crisp sentence describing the exact outcome.

Non-goals (explicitly excluded):
- [Non-goal 1]
- [Non-goal 2]

## Scope & Assumptions

### In Scope
- [Specific bounded item 1]
- [Specific bounded item 2]

### Out of Scope
- [Item that appears related but is excluded]

### Assumptions
- [ ] []
 [ ] []


 [  ]
 [  ]



 

   [ ]  [  ]  [] 
   [ ]  [  ]  [] 





 [ ] [ ]
 [ ] [  ]
 [ ] [ ]
 [ ] [   ]



         

       []   
      []   
       []   
       []   
       []   



   



          

   [   ,  ]  []     
   [ ]    
   [ ]  []     
   [ ]    



          

   [ ]  []      
   [ ]    



          

   [ ]  []      
   [ ]    



          

   [ ]  []      
   [ ]    



          

   [ ]  []      
   [ ]    



          

 []      []  [ ] 



           

       

   []   
   []   
   []   





        

 []  []  [ ]  [ ] 

      



  
  
  
  



     
  
  

Step 5: Final Summary

Output completion summary:

✅ Plan Created: memory-bank/plan/YYYY-MM-DD_HH-MM-SS_<topic>.md

Summary:
- Milestones: [N]
- Total Tasks: [N]
- Critical Gates: [list]
- Estimated Duration: [total]

Next: Begin execution with M1 tasks

Constraints

Constraint	Rule
NO EXECUTION	Never write, modify, or execute code
NO CODING	Never implement, prototype, or test solutions
NO NEW RESEARCH	Only use existing research document; do not investigate unless resolving blocking questions
Singular focus	One primary plan; maximum one alternative
Granular tasks	Each task: 30 minutes to 4 hours of work
Ordered execution	Tasks must be completable in sequence
Junior-ready	Clear enough to execute without questions
Test limit	Maximum one test file per milestone

Subagent Usage

If additional analysis is needed during planning:

With subagents available: Deploy maximum 3 in parallel:

Subagent	When to Deploy
context-synthesis	Identify hidden dependencies or integration points
codebase-analyzer	Understand implementation patterns at specific locations
codebase-locator	Find examples of how similar problems were solved

Without subagents: Follow the workflow in order manually:

1. Read research document completely
2. Identify and list all open questions
3. If critical: ask user for clarification
4. If deferrable: document as assumptions with risk flags
5. Proceed to generate the plan