execute-from-plan

name	execute-from-plan
description	Execute implementation plans from memory-bank/plan/. STRICTLY execution only—NO research, NO planning, NO deviation from plan without confirmation. Trust that planning-from-research has already prepared the work. Output is working code with execution logs.
allowed-tools	Read Write Edit Bash Agent

Execute From Plan

Execute tasks from a prepared implementation plan. This skill walks the plan milestone by milestone, task by task, producing working code.

CRITICAL BOUNDARIES

Activity	Status
Execution	✅ This skill
Planning	❌ Use planning-from-research first
Research	❌ Use codebase-research first

This skill is STRICTLY for execution. Do not plan, do not research, do not question the plan. Trust that preparation work is complete. Execute what is written.

When to Use

Use this skill when:

• A plan document exists in memory-bank/plan/
• The task is to implement/build what the plan specifies
• Planning and research phases are already complete

Workflow

Step 1: Load Plan

Locate and read the plan document:

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

Extract:

• Goal and non-goals
• Milestones (M1-M5) and their tasks
• Acceptance criteria for each task
• Files and interfaces to modify
• Assumptions and constraints

Step 2: Check for Drift

Compare current git state to the plan's recorded state:

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

If current HEAD differs from git_commit_at_plan:

• Warn: "DRIFT DETECTED — code changed since plan was created"
• Review the plan's "Drift Notes" section
• If significant drift: Ask user before proceeding
• If minor: Proceed with caution, document deviations

Step 3: Execute Tasks in Order

Execute strictly in milestone order: M1 → M2 → M3 → M4 → M5

Within each milestone, execute tasks by ID order (T1, T2, T3...).

For each task:

1. Read task details

   • Task description (what to implement)
   • Files/interfaces to touch
   • Dependencies (must be complete first)
   • Acceptance criteria

2. Verify dependencies

   • Confirm all dependent tasks are marked complete
   • If not: Stop, report blocking dependency

3. Execute the task

   • Implement exactly what the task specifies
   • Touch only the files/interfaces listed
   • Do not add features beyond the task scope

4. Run acceptance verification

   • Verify against task acceptance criteria
   • Run any tests specified in the plan
   • If criteria fail: Fix or stop and report

5. Update plan document

   • Mark task as complete
   • Add execution notes (what was done, any issues)
   • Record actual time taken (optional)

6. Commit (optional)

   • Consider committing after each milestone
   • Use milestone ID in commit message: "M1: Architecture & skeleton"

Step 4: Handle Blocks

If blocked during execution:

1. Document the block

   • What was being attempted
   • What went wrong
   • What was tried

2. Update plan document

   • Mark task status as "BLOCKED"
   • Add notes describing the block

3. Ask user for guidance

   • Present the documented block
   • Request decision: fix, skip, or replan

Never deviate from the plan without explicit user confirmation.

Step 5: Complete or Report

If all tasks complete:

• Mark plan phase as "EXECUTED"
• Create execution log: memory-bank/execution/YYYY-MM-DD_HH-MM-SS_<topic>.md
• Output summary: tasks completed, files modified, tests passing

If blocked and cannot proceed:

• Report incomplete tasks
• Describe blocking issue
• Suggest next steps (replan, research, user intervention)

Task Execution Rules

Rule	Description
Trust the plan	Do not second-guess or research alternatives
Stay in scope	Implement exactly what the task specifies
Verify after each	Run acceptance criteria before marking complete
Update the plan	Document progress and notes in the plan file
Stop on failure	Do not proceed past failed acceptance criteria
Ask on deviation	Get user confirmation before changing the plan

Subagent Usage

If execution assistance is needed:

With subagents available: Deploy maximum 3:

Subagent	When to Deploy
lint-issue-resolver	Fix lint/type errors introduced during implementation
antipattern-sniffer	Review code for anti-patterns after significant changes
codebase-analyzer	Understand existing code patterns before modifying

Without subagents: Execute manually following the plan's task descriptions.

Execution Log Format

Create memory-bank/execution/YYYY-MM-DD_HH-MM-SS_<topic>.md:

---
title: "<topic> – Execution"
phase: Execution
date: "YYYY-MM-DD HH:MM:SS"
owner: "<agent_or_user>"
parent_plan: "memory-bank/plan/<file>.md"
git_commit_at_start: "<sha>"
git_commit_at_end: "<sha>"
tags: [execution, <topic>]
---

## Summary

| Metric | Value |
|--------|-------|
| Milestones completed | N/N |
| Tasks completed | N/N |
| Tasks blocked | N |
| Files modified | N |
| Tests added | N |
| Execution time   





      

      [] 
      [] 
      [ ] 



      

      [] 

[]



   [ ]
   [ ]



      

   []  [ ] 



      
  []
  []
  []

Constraints

Constraint	Rule
NO PLANNING	Never create new plans or modify existing plan structure
NO RESEARCH	Never investigate beyond what the plan specifies
Strict ordering	M1 before M2, T1 before T2
Verify before proceeding	Acceptance criteria must pass
Document everything	Update plan and create execution log
Stop on block	Do not guess or work around blocks