Every AI coding agent starts fresh. It doesn't know your standards, your workflow, or what capabilities it has access to. You re-explain the same things in every project and every conversation. agents-core fixes that.
Install into a project
curl -fsSL https://agents-cli-v1.pages.dev/install | bash
copy
Detects your agent config file (CLAUDE.md, .cursorrules, copilot-instructions.md). Creates CLAUDE.md if none found.
curl -fsSL https://agents-cli-v1.pages.dev/install | bash -s -- --role coding --context backend
copy
Specify a role and context.
npm install -g agents-core-cli
copy
Or install the CLI for interactive setup: agents setup
Roles
coding
Write, modify, and review code
research
Find, synthesize, and cite information
planning
Break goals into actionable tasks
review
Review code, docs, and plans
Contexts
frontend
UI, components, accessibility
backend
APIs, validation, databases
devops
Infrastructure, CI/CD, monitoring
data
Pipelines, ingestion, transformation
Skills
research
Brave Search API — web search with citations
browser
Headless browser via Browserless or Playwright
database
DB connection, ORM setup, migrations
To add a skill, tell your agent: "Set up the research skill" — it reads the setup instructions and configures everything.
Core Coding Rules
Small, reusable functions — single responsibility, extract repeated logic into helpers
Always write tests alongside every new function
Prefer modern libraries over custom implementations
Organize utilities — shared helpers in lib/ or utils/
Check before you write — don't duplicate existing code
Always flag security design flaws and offer concrete fixes