spec-coding-skill

A spec-driven skill for LLM agents to plan and implement large-scale coding tasks with structure, consistency, and quality. Uses round-based iterative execution — problems found during implementation are logged and resolved in subsequent rounds rather than silently deviating from the plan. Includes a project-level Blueprint for tracking all requirements, their phases, dependencies, and status at a glance.

What It Does

Guides an agent through a structured, document-first workflow with round-based execution. Each round is a full pass through the spec-driven flow:

Task Sizing → Phase 01-05 → Phase 06 → Phase 07 → Review → (done or next round)
                                                                      |
                                                                      v
                                                              Phase 01* (Round N+1)
                                                              → back to Phase 02-07

If problems arise during execution, they are captured in issues.md and resolved in the next round — the agent never silently deviates from the plan.

Usage

Activate this skill when you need to:

• Start a new software project or feature from scratch
• Refactor or extend existing code in a structured way
• Implement a complex feature that spans multiple files or modules
• Ensure coding work follows OOP principles and language-specific standards

Getting Started

Tell your agent:

Use the spec-coding-skill to implement [your requirement].

The agent will:

1. Size the task (XS/S skip ceremony, M+ enter full flow)
2. Create .dev/[NNN]-[req-name]/init.md with requirement definition
3. Register the requirement in .dev/blueprint.md
4. Produce planning documents under .dev/[NNN]-[req-name]/generated/
5. Execute tasks in rounds, each round having its own plan.md and tasks.md
6. If issues arise, log them to issues.md and resolve in the next round

Project Structure

.dev/
├── blueprint.md                     # Project-wide req overview (status, deps, milestones)
├── TODO.md                          # Cross-requirement backlog
└── 001-your-requirement/
    ├── init.md                      # Requirement definition
    ├── issues.md                    # Cross-round issue log
    └── generated/
        ├── start-and-resume.md      # Shared execution rules + Round History
        └── rounds/
            ├── round-001/
            │   ├── plan.md          # Round 1 technical approach
            │   └── tasks.md         # Round 1 task list
            └── round-002/           # Current round
                ├── plan.md
                └── tasks.md

Reference Files

File	Purpose
SKILL.md	Entry point — phase overview and core rules
references/README.md	Reader's guide — file navigation by category
references/constitution/00-agent-execution.md	Agent behavior rules (global, all phases)
references/constitution/01-oop-principles.md	OOP & SOLID principles
references/constitution/02-coding-standards.md	Coding standards: universal + language-specific
references/constitution/03-git-workflow.md	Git: branch, commit, PR/merge, tagging
references/execution/00-start-and-resume.md	Execution loop + Deviation Protocol + Round History
references/execution/01-round-mechanism.md	Round state machine + decision matrix + issues.md format
references/phases/00-initialization.md	Phase 01: creating init.md
references/phases/01-prerequisite-tasks.md	Phase 02: inspect / research / profiling / diagnosis
references/phases/02-algorithm-design.md	Phase 03: algorithm design
references/phases/03-implementation-plan.md	Phase 04: creating plan.md
references/phases/04-task-planning.md	Phase 05: creating tasks.md
references/phases/05-blueprint-management.md	Blueprint layer: project-level req roadmap

Installation

Install using skills:

npx skills add https://github.com/Hunter-Leo/llm-agent-oop-coding-skill --skill spec-coding-skill -a kiro-cli -g -y

Requirements

• A LLM agent with file read/write and shell execution capabilities (e.g. Kiro, Claude Code)
• Git initialized in the project root