Skip to content
Docs menu

Plugins and extensions

Extend venom with plugins, pi packages, Claude Code plugins, hooks, sounds, skills, MCP servers, themes, templates and roles.

View .md

venom is built on pi, and its extension points are pi's. A venom plugin is a pi extension loaded by pi's own loader, so a pi package works as a venom plugin without changes, and Claude Code plugins load through venom's compatibility layer.

You want to Use Runs code?
Add a tool, slash command, leader key, renderer or HUD field a plugin yes, in-process
Run a shell command or play a sound when the agent finishes, fails or needs you a hook (hooks: in venom.yaml) a shell command
Teach the agent a procedure it loads on demand a skill (SKILL.md) no
Give the agent tools from another process or service an MCP server yes, as a separate process or remote service
Change venom's colours a theme file no
Start a recurring kind of job from a saved setup a template no
Add a standing scheduled agent to venom serve a plugin role no (a schedule entry)

A plugin can bundle all of these: its manifest lists extensions, skills, commands, agents, prompts, themes, hooks, MCP servers and roles.

Trust in one paragraph

A plugin's extension code runs inside venom with your user's rights. venom's permission gate judges the model's tool calls, including a plugin's tools, but it does not sandbox the plugin's own code. Plugins you install yourself load because you installed them. Plugins, hooks and MCP servers that come from a project's files load only after you trust the project (venom trust, or the dialog venom shows the first time), and a change to any of them suspends that trust until you confirm it again. Themes, skills and templates are data, not code.

Built-in plugins

The CLI ships some of its own features as plugins: the terminal charts (charts), the pull requests pane (github-prs) and /rewind. venom plugin list shows them as [builtin], and venom plugin disable <name> turns one off.

In this section

  • Install and manage plugins: Add plugins from a directory, npm, git or your org's marketplace, turn them on and off, configure them and see why one did not load.
  • Write a plugin: Scaffold, write, test and publish a venom plugin with tools, commands, leader keys, block renderers, HUD fields and panel sections.
  • 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.
  • Hooks and venom's events: Run a command or play a sound when the agent finishes, fails, needs approval or completes a step, and use Claude Code's hooks as they are.
  • Sounds: Hear when venom is done, failed or needs approval, pick or replace the sounds, set quiet hours and choose the player.
  • Skills and slash commands: Add skills the agent loads on demand and slash commands from markdown files, list what is loaded, and find skills the agent asked for but did not have.
  • MCP servers: Give the agent tools from MCP servers in Claude Code's .mcp.json format, and use your organisation's approved servers with sign-in and tool pins.
  • Themes: Pick one of venom's sixteen colour themes, tweak one, write your own theme file or ship themes in a plugin.
  • Session templates and learned templates: Start a session from a saved setup (prompt, gear, mode, tools, checks) and save jobs you repeat as templates.
  • Team library: Use your organisation's shared prompts as /org commands and its skills in every session, offline from a cache, and hide what you do not want.
  • Plugin roles: Install a standing agent a plugin ships (cron, prompt, mode, budget) as a reviewed schedule entry that venom serve runs.
  • Terminal charts: Draw bar, line, area, scatter, sparkline, histogram, heatmap, pie, gauge and timeline charts in the terminal from a small YAML or JSON spec.
  • Pull requests pane: Browse the repository's open pull requests beside the chat and approve, comment, request changes or merge through the GitHub CLI.
  • venom-board, GitHub issues as an agent control plane: Run a GitHub issue board where agents file ranked proposals, an owner approves with labels, and agents build, gate and merge what was approved.