Plugins Overview

Plugins receive compiled agent profiles and transform or extend them for specific platforms. The compiler calls each enabled plugin after writing the pure profile.

How plugins work

After compiling a role, Praxis:

1. Writes the pure profile to {agentProfilesOutputDir}/{alias}.md
2. Passes the profile content and role metadata to each enabled plugin
3. Each plugin writes its own output to its configured output directory

Plugins are stateless — they receive a profile string and metadata, and they write files. They do not modify the compilation process or the pure profile output.

Enabling a plugin

Add the plugin name to the plugins array in .praxis/config.json:

{
  "plugins": ["claude-code"]
}

Or use the object form with per-plugin options:

{
  "plugins": [{
    "name": "claude-code",
    "outputDir": "./plugins/my-agents",
    "claudeCodePluginName": "my-org"
  }]
}

Available plugins

Plugin	Description
claude-code	Generates Claude Code agent files with YAML frontmatter, manages plugin.json, and creates a /validate slash command

Plugin output directories

By default, the Claude Code plugin writes to ./plugins/praxis/. Each plugin can be configured with a custom outputDir.

Plugin output directories are separate from agentProfilesOutputDir. You can have both:

agent-profiles/
└── reviewer.md          ← pure profile

plugins/praxis/
└── agents/
    └── reviewer.md      ← Claude Code agent file

Or disable pure profiles and only generate plugin output:

{ "agentProfilesOutputDir": false }

See also

• Claude Code Plugin
• Configuration