Agent Skills

Development standards, workflows, and AI agent configurations for consistent, high-quality software development.

New to APM? Start with the Getting Started Tutorial to build a todo app with OpenCode.

Want to understand the architecture? See docs/FRAMEWORK.md for how primitives work together.

Architecture

This package uses a lightweight agents + modular skills architecture:

┌─────────────────────────────────────────────────────────────┐
│  Agents (Personas)           Skills (Expertise)             │
│  ─────────────────           ─────────────────              │
│  @code-reviewer       →      review/code, review/security   │
│  @architect           →      architecture/patterns          │
│  @test-engineer       →      testing/strategy, testing/e2e  │
│  @devops-engineer     →      devops/cicd, devops/iac, ...   │
│  @fullstack-developer →      stack/sst, stack/astro, ...    │
└─────────────────────────────────────────────────────────────┘

• Agents are lightweight personas (~80-100 lines) with tool permissions
• Skills are reusable expertise modules loaded dynamically
• Context bloat reduced by ~37% compared to monolithic agents

Installation

Install APM

macOS (Homebrew)

brew install danielmeppiel/tap/apm

Other platforms

curl -sSL https://raw.githubusercontent.com/danielmeppiel/apm/main/install.sh | sh

Install this package

# Install the skills package
apm install chandima/agent-skills

# Compile for your AI tools
apm compile

Compile for Specific AI Tools

APM compiles skills into the format each AI tool expects:

Target	Output Files	Best For
vscode	AGENTS.md, .github/prompts/, .github/agents/, .github/skills/	GitHub Copilot, Cursor, Codex, Gemini
claude	CLAUDE.md, .claude/commands/, SKILL.md	Claude Code, Claude Desktop
all	All of the above	Universal compatibility

Auto-detection behavior:

• .github/ exists only → compiles for vscode target
• .claude/ exists only → compiles for claude target
• Both folders exist → compiles for all targets
• Neither exists → minimal mode (AGENTS.md only)

# Auto-detect and compile (recommended)
apm compile

# Or specify targets explicitly
apm compile --target vscode
apm compile --target claude
apm compile --target all

What's Included

Skills (Expertise Modules)

Fine-grained knowledge packages loaded by agents. Skills focus on detection methodology (how to find issues) rather than coding patterns (which belong in instructions):

Domain	Skills	Purpose
review/	code, security	Review methodology, vulnerability detection
testing/	strategy, e2e	Testing pyramid, Playwright/agent-browser
devops/	cicd, containers, iac, security	Pipeline audit, Dockerfile review, IaC review
architecture/	patterns	System design, trade-off analysis
stack/	sst, astro, alpine, basecoat	Full-stack technology expertise

Agents (AI Personas)

Agent	Skills	Focus
@code-reviewer	review/code, review/security	Code quality (read-only)
@architect	architecture/patterns, review/code	System design
@test-engineer	testing/strategy, testing/e2e	Test coverage
@devops-engineer	devops/cicd, containers, iac, security	CI/CD, infrastructure
@fullstack-developer	stack/sst, astro, alpine, basecoat	SST + Astro stack

Instructions (Coding Standards)

Auto-applied guardrails that define how to write code (vs. skills which define how to review code):

File	Applies To	Purpose
typescript.instructions.md	**/*.{ts,tsx}	Strict types, no any
astro.instructions.md	**/*.astro	Basecoat UI, Islands
alpinejs.instructions.md	**/*.{astro,html}	Directives, state
sst.instructions.md	**/sst.config.ts	AWS components
github-actions.instructions.md	.github/workflows/**	OIDC, permissions
terraform.instructions.md	**/*.tf	HCL patterns, modules
docker.instructions.md	**/Dockerfile*	Multi-stage, security
jenkins.instructions.md	**/Jenkinsfile*	Declarative pipelines
git.instructions.md	All files	Conventional commits
security.instructions.md	All code	No secrets, validation
json-schema.instructions.md	**/*.schema.json	Draft-04, $ref

Prompts (Workflows)

Command	Purpose
/code-review	Systematic review for bugs, security, performance
/pr-description	Generate PR description from staged changes
/debug-issue	Structured debugging workflow
/refactor	Safe refactoring with test preservation
/test-plan	Generate comprehensive test cases

Usage Examples

Using Agents

@code-reviewer Please review the changes in src/auth/
@architect How should I structure a caching layer for this API?
@test-engineer What test cases should I write for the payment flow?
@devops-engineer Set up a CI/CD pipeline for this Node.js project
@fullstack-developer Build a file upload feature with S3 and progress indicator

Using Prompts

/code-review
/pr-description
/debug-issue --param error="TypeError: undefined is not a function"

Package Structure

.apm/
├── instructions/     # Coding standards (11 files)
├── prompts/          # Workflow commands (5 files)
├── agents/           # Lightweight personas (5 files, ~80-100 lines each)
└── skills/           # Expertise modules (13 skills)
    ├── review/       # code, security
    ├── testing/      # strategy, e2e
    ├── devops/       # cicd, containers, iac, security
    ├── architecture/ # patterns
    └── stack/        # sst, astro, alpine, basecoat

Installing Individual Components

# Install just the code-review prompt
apm install chandima/agent-skills/.apm/prompts/code-review.prompt.md

# Install just the architect agent
apm install chandima/agent-skills/.apm/agents/architect.agent.md

# Install just the TypeScript standards
apm install chandima/agent-skills/.apm/instructions/typescript.instructions.md

# Install a specific skill
apm install chandima/agent-skills/.apm/skills/devops/cicd/SKILL.md

Dependencies

This package includes:

• vercel-labs/agent-browser - Browser automation CLI for E2E testing
• vercel-labs/agent-skills#web-design-guidelines - 100+ UI/UX/accessibility audit rules

Optional Dependencies

# n8n workflow automation (requires n8n-mcp MCP server)
apm install czlonkowski/n8n-skills

MCP Server Compatibility

This package works alongside MCP servers for enhanced capabilities:

Server	Use Case	Benefits
GitHub MCP	Code review, PR workflows	PR context, code search
Terraform MCP	Infrastructure	State inspection, plan interpretation
Context7	Library documentation	Up-to-date framework docs
n8n-mcp	Workflow automation	Node docs, validation

GitHub Enterprise Cloud (ghe.com)

{
  "github": {
    "type": "http", 
    "url": "https://copilot-api.<your-subdomain>.ghe.com/mcp"
  }
}

Customization

After installing, customize any file in your project's .apm/ directory. Local changes take precedence over package defaults.

Primitive Architecture

This package follows the AI-Native Development framework. Key distinction:

Primitive	Purpose	Content Type
Instructions	HOW TO WRITE code	Coding rules, patterns, examples
Skills	HOW TO FIND issues	Detection methodology, grep patterns
Prompts	Workflow execution	Orchestration steps
Agents	AI personas	Identity, tool permissions, skill refs

For a complete explanation, see docs/FRAMEWORK.md.

Contributing

Found an issue or have a suggestion? Open an issue or PR on GitHub.

License

MIT