agentic-swe

Product focus

agentic-swe is an open-source foundation for teams that use coding agents. It separates persistent repository rules from task-specific procedures, so agents can make reliable workflow decisions before they edit code.

System model

• Agentic protocol: a concise session-start policy that classifies work and enforces planning for non-trivial changes.
• Repository contract: AGENTS.md carries the versioned project rules, while a machine-readable manifest records selected skills and local workflow paths.
• Skill collection: reusable workflows, templates, and references load only when the task requires them.

Engineering approach

The package is designed to work across OpenAI Codex and Claude Code without tying project governance to a single agent. It ships non-destructive installers, portable validation, and a public MIT-licensed skill collection.