multi-agent-patterns

Part of the Atomic agent framework, this skill enables deep codebase research by dispatching specialized sub-agents — a codebase-locator finds relevant files, a codebase-analyzer reads implementations, and an online-researcher queries external docs.

Part of the Atomic agent framework, this skill explains code functionality in detail using DeepWiki to provide comprehensive code understanding and documentation.

Part of the Atomic agent framework, this skill helps create, improve, and optimize prompts using best practices, within Atomic's multi-agent architecture that dispatches specialized sub-agents for focused task execution.

An Atomic SDK skill for summarizing transcripts at session boundaries while preserving actionable information to manage long-running agent sessions.

A meta skill from the Atomic multi-agent harness that enables creating, modifying, evaluating, and benchmarking custom agent skills. Auto-invoked when building or iterating on SKILL.md files.

An Atomic SDK skill for optimizing LLM context usage through KV-cache optimization, observation masking, and context budgeting techniques.

A built-in Atomic skill that discovers and installs agent skills from the community. Part of the Atomic multi-agent harness that orchestrates Claude Code, OpenCode, and GitHub Copilot CLI.

An Atomic SDK skill for designing clear tool contracts and reducing agent-tool friction in AI coding agent workflows.

An Atomic SDK skill for validating task-model fit before building and estimating costs for AI agent development projects.

An Atomic SDK skill covering how LLM context windows work, including attention mechanics and progressive disclosure techniques for effective context management.