all-agents

open source

Making AI-assisted coding more predictable. Tools and patterns for developers working with AI.

all-agents is an AI-first development framework designed to supercharge Claude Code workflows. It provides a structured approach to AI-assisted development through atomic documentation, reusable skill commands, and orchestrated multi-agent systems.

Atomic documentation

The framework is built around atomic documentation principles - small, focused, self-contained documents that AI agents can easily consume and act upon. This approach enables:

  • Consistent context across AI sessions
  • Modular documentation that can be composed for specific tasks
  • Clear boundaries between different concerns

Skills system

Reusable commands that encapsulate common development workflows:

  • Brainwriting - structured brainstorming sessions for feature planning
  • Code review - multi-dimensional code analysis covering technical quality and intent alignment
  • Research commands - deep research with parallel web searches that synthesize findings into structured summaries

Multi-agent workflows

Orchestrated agent systems for complex tasks:

  • Research agent - runs 5-7 parallel web searches on a topic, synthesizes findings into executive summaries with categorized findings and recommendations
  • Planning agents - high-level and low-level planning for breaking down features into implementable chunks
  • Review agent - reviews code on technical quality (bugs, security, tests, architecture, style) and intent alignment
  • ESLint fixers - the spawn-eslint-fixers command spawns multiple eslint-fix subagents in parallel, each handling specific linting issues - this works particularly well because linting is a clear pass/fail situation

Development workflow

The framework supports a complete AI coding workflow:

  1. Plan feature - brainwriting and brainstorming
  2. Plan code - deep research (online and repository), high-level and low-level planning
  3. Write code - consistent implementation following best practices
  4. Check code - linting, security checks via orchestrated fixers
  5. Review code - structured code review with prioritized feedback
  6. Document code - automated documentation maintenance

Integration

all-agents is designed to work seamlessly with Claude Code, providing a structured layer on top of the AI assistant that makes development more predictable and efficient.