Claude Code plugins and pi packages
Load Claude Code plugins and pi packages in venom as they are, and see which parts of their formats venom supports.
venom reads Claude Code's files and pi's packages without changes. It never writes under .claude/ or ~/.claude.
pi packages
A pi package (a package.json with a pi key, or pi's conventional extensions/ layout) is a venom plugin with only pi's resource kinds. Install it like any plugin:
venom plugin add npm:some-pi-packageA venom key in the same package.json overrides the pi key key by key.
Claude Code plugins
A directory with .claude-plugin/plugin.json loads as a plugin. Plugins you installed with Claude Code are read from ~/.claude/plugins/installed_plugins.json (honouring enabledPlugins) with no extra step. For any other directory:
venom plugin add ~/src/my-claude-plugin| Claude Code | venom |
|---|---|
plugin.json name, version, description |
read |
commands/ (or commands as paths) |
loaded as /<plugin>:<command> |
agents/ (or agents as directories) |
loaded for the task tool; inline mcpServers in a plugin's agent are ignored, as in Claude Code |
skills/ (or skills paths) |
loaded |
hooks/hooks.json, hooks as a path or inline object |
loaded (command and prompt hooks, for the events venom fires) |
.mcp.json, mcpServers as a path or inline map |
loaded |
${CLAUDE_PLUGIN_ROOT} |
expanded in hooks and MCP config |
commands as an object map, agents as .md file paths, hooks/mcpServers arrays |
not yet; use the default directories |
no .claude-plugin/plugin.json |
not recognised; add the manifest |
userConfig, ${user_config.*}, CLAUDE_PLUGIN_OPTION_*, ${CLAUDE_PLUGIN_DATA} |
not supported; use venom's configSchema |
lspServers, .lsp.json, outputStyles, workflows, channels, bin/ on PATH, settings.json, dependencies, defaultEnabled |
not supported |
marketplace.json, /plugin install name@marketplace |
not supported; clone the plugin, then venom plugin add <dir> |
The rest of a Claude Code project
Outside plugins, venom also reads a project's CLAUDE.md, .claude/skills, .claude/commands, .claude/agents, .claude/settings.json (permission rules, hooks, env, model) and .mcp.json. Project hooks, MCP servers and allow rules take effect only once you trust the project. Details per part: Hooks, Skills and commands, MCP servers.