Marketplace for trusted AI agent skills

Run architecture-first coding work with Spec-Driven Develop

Skill Detail

Run architecture-first coding work with Spec-Driven Develop

Guides Markdown-capable coding agents through large changes with deep analysis, intent refinement, task decomposition, GitHub issue tracking, batch PRs, adaptive control, and review workflows.

Templates & WorkflowsMulti-Framework
Templates & Workflows Multi-Framework Security Reviewed
⭐ 953 GitHub stars
INSTALL WITH ANY AGENT
npx skills add agentskillexchange/skills --skill run-architecture-first-coding-work-with-spec-driven-develop Copy
Works best when you want a reusable capability, not another fragile one-off prompt.
View source
At a glance
Tools required
Spec-Driven Develop Markdown skills, filesystem access, optional gh CLI for GitHub Issues, Milestones, Projects, worktrees, and PR tracking
Install & setup
Install the repository's Markdown skills into a supported agent skill directory. For shared Agent Skills-compatible installs, use the bundled scripts/install-agents.sh where appropriate, or copy the skills into Claude Code, Codex, OpenCode, Cursor, or another Markdown-capable agent.
Author
zhu1090093659
Publisher
Independent
Last updated
Jul 27, 2026
Quick brief

## What the agent does
The agent invokes Spec-Driven Develop for complex codebase changes. It analyzes architecture, refines intent, decomposes work into phases, creates local or GitHub-backed trackers, groups implementation into reviewable delivery batches, executes with drift checks, and archives the resulting artifacts.

How it works

What this skill actually does

## When to use it
Use this when a coding request is too large or risky for a direct edit, such as migrations, rewrites, architecture refactors, or multi-phase feature delivery. It is a better fit than ordinary chat prompting when the user needs traceable specs, issue-backed execution, and explicit quality gates.

## Scope boundary
This is not a general project-management framework or a generic spec template. The workflow is bounded to AI coding-agent operation through Markdown skills, optional helper scripts, GitHub Issues and PRs when available, and its S.U.P.E.R/adaptive-control execution loop.

## Source evidence
The upstream repository documents the seven-phase pipeline, Spec-Driven Develop, Deep Discuss, Review SPD, GitHub-native issue and batch-PR modes, local-only fallback, helper scripts, and compatibility with Claude Code, Codex, OpenCode, Cursor, and other Markdown-capable agents.