AgileFlow

AgileFlow

Next

Portable workflows for coding agents. Install a skill once and use it with Codex, Claude Code, Cursor, OpenCode, and Gemini CLI.

The same skills also work in tools built on top of these agents. Small, versioned workflows. No agent runtime. No repository takeover.

What AgileFlow is

AgileFlow is a skill manager, compatibility layer, and evaluation system for coding agents.

  • It installs standard Agent Skills (SKILL.md directories) into .agents/skills, the location Codex, Cursor, OpenCode, and Gemini CLI read natively.
  • It makes the same skills visible to Claude Code through per-skill links in .claude/skills.
  • It records exact versions and hashes in agileflow.lock, so updates are deterministic and your edits are never silently overwritten.
  • It lets you fork any skill and own it permanently.
  • It ships evals so skill descriptions and instructions improve because measured behavior improved.
  • Optionally, Agile Work keeps product, roadmap, epics, stories, and decisions as Markdown in one directory, with a CLI that derives boards and validates relationships.

What AgileFlow is not

AgileFlow is not an agent framework, a multi-agent runtime, a hook runtime, a Jira clone, or a repository scaffolder. Agile Work is opt-in and stays small: five artifact types, no sprints, no points, no second state store. Your coding agent keeps doing the reasoning, planning, delegation, tool use, and permissions.

Once skills are installed, nothing from AgileFlow runs while your agent works: no hook, no background process, no AgileFlow agent. If you uninstall the CLI, your skills keep working because they are standard SKILL.md files in provider-native locations.

Where to start

Looking for AgileFlow v4? v4 (plugins, agents, hooks, slash commands, docs scaffolding) is superseded by v5. Its documentation is archived with the source on the v4 branch on GitHub. Nothing in that archive describes current behavior.