A community-driven registry for Claude, Cursor, Windsurf, Cline & more. Not affiliated with Anthropic.
Are you the author? Sign in to claim
A tool-agnostic AI-assisted development framework template. Structured workflow (Spec → Plan → Implement → Review) with
A tool-agnostic framework for AI-assisted software development. It structures the entire development lifecycle — from idea to production — with canonical protocols that any AI coding assistant can follow: Claude Code, Cursor, Codex, Gemini CLI, or any other tool.
This template provides:
.claude/agents/), Cursor (.cursor/agents/, .cursor/commands/), and Codex (.codex/skills/) that point to those protocolsThe key principle: protocols live in docs/ and are tool-agnostic. Tool-specific configs (.claude/, .cursor/) are thin wrappers that reference those protocols. Add support for a new AI tool by creating a thin wrapper — no protocol duplication needed.
When you first clone this template into a new project, run the Project Setup workflow to generate your project-specific documentation:
Use the project-setup agent: Task the agent with "Initialize this project using the setup protocol"
/setup-project
Ask your AI assistant to:
Follow the project setup protocol at docs/workflow/setup/protocol.md
The setup agent will have a structured conversation with you to understand your project, then generate:
docs/project/1-business-domain.md — Domain model, entities, business rulesdocs/project/2-repo-architecture.md — Monorepo/repository structuredocs/project/3-software-architecture.md — Tech stack, design patterns, architectural decisionsdocs/project/4-database-model.md — Data model (if applicable)docs/best-practices/STACK-SPECIFIC.md — Best practices tailored to your stackAGENTS.md with your project context├── AGENTS.md # Universal AI guidance (all tools read this)
├── CLAUDE.md -> AGENTS.md # Symlink for Claude Code
├── CHANGELOG.md
├── e2e/ # Placeholder e2e/regression test project (Playwright)
│ ├── playwright.config.ts # Minimal Playwright config
│ ├── tests/
│ │ └── baseline.spec.ts # Always-passing placeholder test
│ └── package.json
├── .github/
│ └── workflows/
│ ├── auto-tag-release.yml # Tags release branches after merge to main
│ ├── deploy.yml # Placeholder branch-based deployment workflow
│ └── e2e-regression.yml # Label-gated e2e/regression placeholder
│
├── docs/
│ ├── README.md # Documentation index
│ │
│ ├── project/ # Project-specific docs (fill via setup agent)
│ │ ├── 1-business-domain.md # PLACEHOLDER
│ │ ├── 2-repo-architecture.md # PLACEHOLDER
│ │ ├── 3-software-architecture.md # PLACEHOLDER
│ │ └── 4-database-model.md # PLACEHOLDER (delete if no DB)
│ │
│ ├── best-practices/ # Coding standards and conventions
│ │ ├── 1-general.md # General development standards
│ │ ├── 2-version-control.md # Git conventions
│ │ ├── 3-testing.md # Testing standards
│ │ └── STACK-SPECIFIC.md # PLACEHOLDER (fill via setup agent)
│ │
│ └── workflow/
│ └── development-workflow/
│ ├── README.md # Master workflow document
│ ├── agent-model-config.md # Model tier policy and overrides
│ ├── protocols/ # Canonical stage protocols
│ │ ├── 01-generate-spec-protocol.md
│ │ ├── 01-review-spec-protocol.md
│ │ ├── 02-generate-implementation-plan-protocol.md
│ │ ├── 02-review-implementation-plan-protocol.md
│ │ ├── 03-implement-development-protocol.md
│ │ ├── 03-review-implementation-protocol.md
│ │ ├── 04-smoke-test-protocol.md
│ │ ├── 05-prepare-release-protocol.md
│ │ ├── 90-batch-orchestrate-work-protocol.md
│ │ ├── 91-orchestrate-work-protocol.md
│ │ ├── 92-pr-readiness-signal-protocol.md
│ │ └── 93-automated-reviewer-loop-protocol.md
│ ├── templates/ # Spec, plan, and test templates
│ │ ├── spec-template.md
│ │ ├── implementation-plan-template.md
│ │ └── smoke-test-runbook-template.md
│ └── integrations/ # Optional tool integrations
│ ├── linear.md # Issue tracker integration (Linear)
│ ├── greptile.md # Automated PR review (Greptile)
│ ├── github-projects.md # GitHub Projects board integration
│ ├── ci-cd-deployment.md # CI/CD deployment placeholders and customization
│ └── e2e-regression.md # E2E/regression test integration
│
├── .codex/
│ └── skills/ # Codex skills that wrap the workflow protocols and ship UI metadata
│
├── scripts/
│ ├── development-workflow/ # AI workflow helpers (orchestrator, PR/CI loops, state discovery)
│ │ ├── add-backlog-item.sh # Resolves tracker destination; creates GitHub issues when configured
│ │ ├── discover-workflow-state.sh # Summarizes branches, worktrees, development folders, and open PRs
│ │ ├── check-workflow-branch.sh # Checks whether a workflow branch already exists
│ │ ├── pr-review-loop.sh # Polls Greptile PR review until clean / fix / escalate
│ │ ├── pr-ci-loop.sh # Polls CI checks until green / red / timeout
│ │ ├── workflow-batch-plan.sh # Classifies development folders into batch-planning candidates
│ │ ├── workflow-next-action.sh # Classifies the next deterministic workflow action
│ │ └── install-codex-skills.sh # (Codex only) Installs repo skills into your local Codex config
│ └── README.md # Points to development-workflow; add your own scripts here
│
├── .claude/
│ └── agents/ # Claude Code subagent definitions
│ ├── project-setup.md
│ ├── product-manager.md
│ ├── spec-reviewer.md
│ ├── tech-lead.md
│ ├── implementation-plan-reviewer.md
│ ├── developer.md
│ ├── code-reviewer.md
│ ├── item-orchestrator.md
│ └── orchestrator.md
│
└── .cursor/
├── rules/ # Cursor context rules
├── agents/ # Cursor workflow subagents (orchestrator, item-orchestrator, developer, etc.)
└── commands/ # Cursor slash commands
Backlog
│
▼ /generate-new-feature / product-manager agent
Spec Ready ──────────────────────────────── (Human approves PR)
│
▼ /generate-implementation-plan / tech-lead agent
Plan Ready ──────────────────────────────── (Human approves PR)
│
▼ /implement-development / developer agent
In Development ─────────────────────────── (Human approves PR)
│
▼
Merged
│
▼ /prepare-release
Released
Special paths:
refactor/[slug] from develop): code restructuring or tech-debt cleanup with a plan but no specfix/[slug] from develop): bugs and simple changes that don't need a spec or planhotfix/[slug] from main): critical production bugs that need immediate deploymentSee docs/workflow/development-workflow/README.md for the full workflow specification.
.claude/agents/CLAUDE.md (symlink to AGENTS.md) is auto-loadedExample prompts:
Use the orchestrator agent to inspect this repository's AI development workflow state, determine what work can safely advance next, and keep advancing each eligible item until it reaches a real terminal condition. Follow AGENTS.md and the workflow protocols exactly. If multiple items are eligible, prioritize by the documented rules and explain any blockers or escalations.
Use the orchestrator agent to review the current backlog, specs, plans, branches, and open PRs in this repository. Advance deterministic in-flight work, and when Backlog work is next, propose the largest safe start batch by priority and parallelization feasibility before asking for approval. Minimize human interaction, but stop for human decisions, merge review, blocked dependencies, or escalations.
Use the item-orchestrator agent to start and advance work for [feature or issue name]. Resolve the request to one workflow item, then keep progressing that item through creator, reviewer, PR, automated review, and CI until it is waiting on a human, blocked, or escalated.
.cursor/rules/ provide automatic context.cursor/commands/ are invoked with /command-name.cursor/agents/ (e.g. /developer, /orchestrator, /item-orchestrator) run stage-specific protocols; see docs/workflow/development-workflow/agent-model-config.md for model config.cursor/.mcp.jsonExample commands:
/run-work
/run-work Review the current backlog, specs, plans, branches, and open PRs in this repository. Advance deterministic in-flight work, and when Backlog work is next, propose the largest safe start batch by priority and parallelization feasibility before asking for approval. Minimize human interaction, but stop for human decisions, merge review, blocked dependencies, or escalations.
/run-work Start and advance work for [feature or issue name]. Inspect the current workflow state first, then keep progressing that item through creator, reviewer, PR, automated review, and CI until it is waiting on a human, blocked, or escalated.
/run-item-work Start and advance work for [feature or issue name]. Resolve the request to one workflow item, then keep progressing that item through creator, reviewer, PR, automated review, and CI until it is waiting on a human, blocked, or escalated.
./scripts/development-workflow/install-codex-skills.shworkflow-orchestrator as the default portfolio-wide entrypoint for the workflowworkflow-orchestrator on an economy tier by default; only escalate when the stage-specific skill recommends itworkflow-item-orchestrator when you want to resume or advance one specific development, branch, or PR.codex/skills/ when you want to run a specific stage directlyagents/openai.yaml metadata for cleaner labels and starter prompts in Codex-compatible UIsExample prompts:
Use $workflow-orchestrator to inspect this repository's AI development workflow state, determine what work can safely advance next, and keep advancing each eligible item until it reaches a real terminal condition. Follow AGENTS.md and the workflow protocols exactly. If multiple items are eligible, prioritize by the documented rules and explain any blockers or escalations.
Use $workflow-orchestrator to review the current backlog, specs, plans, branches, and open PRs in this repository. Advance deterministic in-flight work, and when Backlog work is next, propose the largest safe start batch by priority and parallelization feasibility before asking for approval. Minimize human interaction, but stop for human decisions, merge review, blocked dependencies, or escalations.
Use $workflow-item-orchestrator to start and advance work for [feature or issue name]. Resolve the request to one workflow item, then keep progressing that item through creator, reviewer, PR, automated review, and CI until it is waiting on a human, blocked, or escalated.
AGENTS.md for project contextdocs/workflow/development-workflow/protocols/docs/workflow/development-workflow/integrations/linear.mddocs/workflow/development-workflow/integrations/coderabbit.md or docs/workflow/development-workflow/integrations/greptile.md. For projects that want to avoid per-hour rate-limit stalls, claude-code-action is the recommended phase_after_clean value (no review cap; uses your own Anthropic API key). See docs/workflow/development-workflow/integrations/claude-code-action.md for setup instructions (added by sibling item #707).docs/workflow/development-workflow/integrations/github-projects.mddocs/workflow/development-workflow/integrations/ci-cd-deployment.mddocs/best-practices/2-version-control.md#changelog, including parallel-batch handling.--force, reset --hard, rebase on shared branches).~/Git/
├── ai-dev-framework-template/ ← this repo (upstream)
├── project-a/ ← downstream: has its own copy of framework files
├── project-b/ ← downstream: has its own copy of framework files
└── ...
Each project created from this template carries its own copy of the framework files. This keeps projects fully self-contained — every team member, CI/CD system, and AI agent has access to the docs without external dependencies.
When you improve a protocol, agent, or best practice in this template repo, copy the framework-level files into existing projects and commit the changes there.
Framework-level paths to propagate:
docs/workflow/.claude/agents/.claude/skills/.codex/skills/.cursor/rules/.cursor/agents/.cursor/commands/scripts/development-workflow/install-codex-skills.shscripts/development-workflow/add-backlog-item.shscripts/development-workflow/discover-workflow-state.shscripts/development-workflow/check-workflow-branch.shdocs/best-practices/1-general.mddocs/best-practices/2-version-control.mddocs/best-practices/3-testing.mdProject-specific paths to avoid overwriting:
docs/project/AGENTS.mdREADME.mdCHANGELOG.mddocs/best-practices/STACK-SPECIFIC.mdIf your project has the .claude/skills/sync-template.md skill, you can automate this process:
| Tool | Command |
|---|---|
| Claude Code | /sync-template |
| Cursor | /sync-template |
The skill compares your project against the template, shows exactly what changed (categorized by auto-apply vs. manual review vs. skipped), and applies updates only after your explicit approval. It also generates ready-to-use git instructions for branching, committing, and opening a PR.
On first run, the skill will ask you for the template source (local path or remote URL) and save it to .tmp/template-config.json for future runs (.tmp/ is gitignored).
When you improve a framework file while working on a specific project, port the relevant changes back into this template repo via a normal PR (copy only the framework-level paths listed above, and avoid importing project-specific content).
To enable the "Use this template" button on GitHub:
New projects will be created as independent repositories (not forks), with a clean Git history.
Portable skills, agents, and templates that add Spec-Driven Development and TDD workflows to any Claude Code project. De
Visual, example-driven guide with copy-paste CLAUDE.md templates for quick setup
ATLAS: a senior-engineer layer for Claude Code. Explore with wireframes & prototypes, clarify the essentials, capture it
Production-tested CLAUDE.md and rules that make Claude Code write better code and more human text. Anti-AI-slop writing