AI Dev Kit

Production-ready AI dev workflow scaffold with 59 skills, 33 agents, hooks, and rules.

---

What Is This?

AI Dev Kit is a complete plugin-oriented developer workspace for AI coding assistants. It ships everything needed to run professional-grade engineering workflows — TDD, code review, security audit, CI/CD, ML pipelines, infrastructure-as-code, code quality analysis, and debt tracking — across Claude Code, Codex, OpenCode, and Gemini CLI.

Think of it as a system prompt that scales to 59 specialized skills and 33 domain agents, with lifecycle hooks and automated validation.

---

Quick Start

Claude Code

/plugin marketplace add noah-sheldon/ai-dev-kit
/plugin install ai-dev-kit@ai-dev-kit

Or install directly via HTTPS:

/plugin marketplace add https://github.com/noah-sheldon/ai-dev-kit
/plugin install ai-dev-kit@ai-dev-kit

Codex

codex extensions add https://github.com/noah-sheldon/ai-dev-kit

Or clone and discover automatically:

git clone https://github.com/noah-sheldon/ai-dev-kit

The marketplace at .agents/plugins/marketplace.json will be discovered automatically.

OpenCode

cd .opencode/plugins && npm install && npm run build

Or install via HTTPS:

npx opencode plugin install https://github.com/noah-sheldon/ai-dev-kit

Gemini CLI

gemini extensions link https://github.com/noah-sheldon/ai-dev-kit

Or install via HTTPS directly:

gemini extensions install https://github.com/noah-sheldon/ai-dev-kit

GitHub Copilot CLI

copilot plugin marketplace add https://github.com/noah-sheldon/ai-dev-kit
copilot plugin install ai-dev-kit@ai-dev-kit

Or install via HTTPS shorthand:

copilot plugin install https://github.com/noah-sheldon/ai-dev-kit

Qwen Code

qwen extensions install https://github.com/noah-sheldon/ai-dev-kit

Or install locally:

qwen extensions install /path/to/ai-dev-kit

Manual Install (any harness)

./install.sh

---

What's Inside

ai-dev-kit/
├── agents/              33 specialized agents (planner, architect, code-reviewer, debt analyzers, ...)
├── skills/              59 skill playbooks (TDD, security, ML, infra, web, data, ...)
├── commands/            41 workflow commands and shims
├── hooks/               lifecycle automation (pre-tool, post-tool, session events)
├── rules/               language-specific guidance (common, python, typescript, web)
├── manifests/           install manifests for deterministic setup
├── schemas/             JSON schemas for validation
├── docs/                architecture, design decisions, troubleshooting
├── examples/            reusable templates
├── scripts/             install, validate, sync, and template helpers
├── tests/               smoke tests and surface validation
├── .claude-plugin/      Claude Code plugin manifest + marketplace
├── .codex-plugin/       Codex plugin manifest
├── .agents/             Codex marketplace catalog
├── .gemini/             Gemini CLI extension
└── .opencode/           OpenCode project config + TypeScript plugin

---

Agents (33)

Agent	When to Use
planner	Complex feature work — breaks requirements into phased, mergeable plans
architect	System design, component boundaries, API contracts
tdd-guide	Writing tests first — RED/GREEN/REFACTOR loop
code-reviewer	Reviewing diffs for correctness, regressions, quality
security-reviewer	Auth, secrets, input validation, OWASP review
ai-judge	Rubric-based validation of plans and outputs
build-error-resolver	Fixing TypeScript, Python, and build pipeline errors
e2e-runner	End-to-end test authoring and execution
refactor-cleaner	Cleanup, modernization, tech debt paydown
doc-updater	Syncing docs with code changes
docs-lookup	Finding and referencing documentation
python-reviewer	Python-specific code review (Pandas, FastAPI, SQLAlchemy)
database-reviewer	Schema, migration, and query review
git-agent-coordinator	Branch coordination, merges, PR orchestration
ml-engineer	ML/LLMOps: RAG, evals, model training, deployment
chrome-ext-developer	WXT and Chrome extension development
data-engineer	ETL, data quality, pipeline architecture
infra-as-code-specialist	IaC, CI/CD, deployment pipelines
observability-telemetry	Logs, metrics, traces, dashboard setup
multi-agent-project-manager	Multi-workflow orchestration, backlog, priority queue (never stops)
workflow-auditor	Health checks, stuck detection, quality gate trending, anomaly reporting
reddit-researcher	Reddit sentiment, production war stories, community consensus
codebase-analyzer	Codebase structure, dependency, and complexity analysis
codebase-learner	Learning and understanding unfamiliar codebases
code-quality-analyzer	Static analysis, code quality metrics, and standards enforcement
test-debt-analyzer	Identifying and tracking test coverage gaps
security-debt-analyzer	Security vulnerabilities and debt tracking
performance-debt-analyzer	Performance bottlenecks and optimization opportunities
dependency-debt-analyzer	Outdated dependencies and upgrade paths
architecture-debt-analyzer	Architectural issues and design debt
process-debt-analyzer	Workflow inefficiencies and process bottlenecks
documentation-debt-analyzer	Documentation gaps and staleness detection
technical-debt-analyzer	Overall technical debt assessment and prioritization

---

Skills (59)

Core Engineering

agentic-engineering api-design api-integrations backend-patterns frontend-patterns frontend-design hexagonal-architecture coding-standards codebase-onboarding

Testing & Quality

tdd-workflow code-review security-review security-scan e2e-testing python-testing eval-harness verification-loop

AI / ML

claude-api openai-api langchain-llamaindex mlops-workflow mlops-rag pytorch-patterns deep-research exa-search search-first iterative-retrieval autonomous-agent-harness autonomous-loops continuous-agent-loop context-prune token-budget-advisor prompt-optimizer mcp-server-patterns

Data

data-pipelines data-pipelines-ai database-migrations postgres-patterns document-processing

Infrastructure

aws-devops aws-deployment docker-patterns deployment-patterns ci-pipeline github-ops observability-telemetry multi-agent-git-workflow

Workflow & Operations

architecture-decision-records dmux-workflows documentation-lookup git-workflow skill-authoring backlog-management workflow-status

Code Quality & Analysis

codebase-report technical-debt-report

Specialized

wxt-chrome-extension

---

Commands (41)

Build & quality: build-fix code-review doctor eval ml-review quality-gate review review-pr test-coverage validate verify

Workflow: checkpoint feature-dev plan project-template promote resume-session save-session sessions skill-create skill-health

Git & multi-agent: git-agent multi-agent-status

DevEx: context-budget context-prune install uninstall update-codemaps update-docs

Hook automation: hookify hookify-configure hookify-list hookify-help

Loop & continuous: loop-start loop-status

ML: e2e launch

Learning: learn learn-eval

---

Cross-Platform Support

Platform	Manifest	Marketplace	Install
Claude Code	.claude-plugin/plugin.json	.claude-plugin/marketplace.json	/plugin marketplace add noah-sheldon/ai-dev-kit
Codex	.codex-plugin/plugin.json	.agents/plugins/marketplace.json	Plugin Directory (after clone)
Gemini CLI	.gemini/gemini-extension.json	.gemini/GEMINI.md	gemini extensions link .
OpenCode	.opencode/opencode.json	.opencode/plugins/ (npm)	npm install opencode-ai-dev-kit
Copilot CLI	.github-copilot/plugin.json	.github-copilot/marketplace.json	copilot plugin install noah-sheldon/ai-dev-kit:.github-copilot
Qwen Code	qwen-extension.json	.qwen/marketplace.json	qwen extensions install noah-sheldon/ai-dev-kit
Cursor	.cursor/	—	Manual context pack

---

Core Principles

1. Agent-first — delegate domain work to the right specialist.
2. Test-driven — write tests before implementation when behavior changes.
3. Security-first — validate inputs, avoid unsafe defaults, never hardcode secrets.
4. Plan-before-execute — break complex work into phases with the planner.
5. Model fallback — if a requested model is unavailable, use the default and continue.

---

Workflow

Request → Planner → Architect → Domain Agents → AI-Judge → Implementation → Code Review → Security Review → Merge

1. Plan complex work with the planner agent before touching code.
2. Write tests first — use tdd-guide for RED/GREEN/REFACTOR discipline.
3. Implement with domain specialists (python-reviewer, ml-engineer, etc.).
4. Validate with ai-judge — rubric covers completeness, correctness, security, feasibility, testability.
5. Review with code-reviewer and security-reviewer before merge.
6. Ship with git-agent-coordinator for clean branch management.

---

MCP Servers (Optional)

MCP servers are not bundled with this kit to avoid corporate network/proxy issues. Add your own MCP servers as needed:

Server	Purpose
github	GitHub API access, PR management
context7	Live documentation lookup
exa	Neural web search
memory	Persistent memory across sessions
playwright	Browser automation & E2E testing
sequential-thinking	Step-by-step reasoning

Configure MCP servers in your own .mcp.json at the project root.

---

Requirements

• Claude Code v2.1+ (hooks auto-load by convention)
• Node.js 24+ LTS (for scripts and validation)
• Git (for agent coordination and version control)

---

FAQ

Q: How do I adapt this for my project? Start from docs/examples/project-guidelines-template.md and customize the skills you need.

Q: What if a model is unavailable? Every agent has a fallback_model setting. If the specified model is down, it falls back to the workspace default.

Q: How do I validate everything is correct?

npm test                    # smoke tests
node scripts/validate-surface.js  # production surface check

Q: Can I install only certain skills? Yes — the kit supports selective install. Each skill is self-contained. Copy only what you need.

Q: How do I contribute? See CONTRIBUTING.md. Run npm test before submitting.

---

Docs

Document	Purpose
Catalog	Full inventory of all skills, agents, commands
Command-Agent Map	Which agent handles each command
Skill Development Guide	How to write new skills
Troubleshooting	Common issues and workarounds
Architecture Improvements	Design recommendations
Selective Install Design	How partial installs work
Session Adapter Contract	Session state specification

---

Security

• Never hardcode secrets — use environment variables or secret managers.
• Validate all external input at boundaries.
• Prefer least privilege and explicit allowlists.
• Review any change touching auth, secrets, or shell execution with security-reviewer.

See SECURITY.md for the full policy.

---

License

MIT — see LICENSE.

Author

Noah Sheldon — noahsheldon.dev