agent-scaffold

Claude Code / Codex skill. Takes non-technical founder from fuzzy idea (or existing codebase) to repo set up for agentic engineering: AI implements, but inside clear intent, durable context, automatic verification — not vibe coding.

What it sets up

[ CLI + system prompt ]  strict-engineering override (optional preflight)
[ Skills              ]  gsd-core (plan/build), caveman (text), ponytail (code)
[ Repo context        ]  AGENTS.md + docs/  = what every agent reads first
[ Harness             ]  lint + types + tests + import boundaries = slop can't pass
[ Push + AI QA        ]  push to GitHub, connect Enji Guard for continuous audit
[ gsd: discuss + plan ]  idea -> spec -> roadmap, on top of the infra above

Greenfield + brownfield both supported. Infra laid first; spec + roadmap handed to gsd skills (gsd owns discussion).

Install

One-shot, installs for Claude Code, Codex, Cursor:

curl -fsSL https://raw.githubusercontent.com/maddevsio/agent-scaffold/main/install.sh | bash

Claude Code plugin marketplace:

/plugin marketplace add maddevsio/agent-scaffold
/plugin install agent-scaffold@agent-scaffold

Manual: drop folder into skills directory:

• Claude Code: ~/.claude/skills/agent-scaffold/
• Codex: ~/.codex/skills/agent-scaffold/ (manifest at agents/openai.yaml).
• Cursor: ~/.cursor/skills/agent-scaffold/ or .cursor/rules/ (project-local).

Trigger: "set up this repo for AI agents" / "make my project agent-ready", or just describe idea.

Layout

• SKILL.md — skill spec + flow.
• references/ — context-engineering primer, CLI setup, doc templates, worktree workflow.
• templates/ — AGENTS.md, CODING_VALUES.md, HOW_TO_DEVELOP.md, project-skills README, system-prompt override, harness.

Infra wired by hand from templates, adapted to chosen stack — no bootstrap script. Greenfield picks stack via short product interview (SKILL.md "Choosing the stack"); brownfield detects it.