AGENT SKILL

create-plugin

Create a Vibe plugin package — the Agent Plugins 1.0 format with optional Vibe extensions for skills, MCP servers, hooks, knowledge, agents, libraries, and connectors. Use when the user wants to create, scaffold, or author a plugin for Mistral Vibe.

License 许可
Apache-2.0
最近更新
2026-11-11
Stars
★ 5.4k
Forks
794

01 前言 Front Matter

name
create-plugin
description
Create a Vibe plugin package — the Agent Plugins 1.0 format with optional Vibe extensions for skills, MCP servers, hooks, knowledge, agents, libraries, and connectors. Use when the user wants to create, scaffold, or author a plugin for Mistral Vibe.

03 SKILL.md 正文

Create a Vibe Plugin

A plugin is a directory containing a plugin.json manifest (Agent Plugins 1.0 schema) plus optional component directories and config files. Vibe discovers plugins under .vibe/plugins/ (project scope, requires trusted folder) and ~/.vibe/plugins/ (user global).

Plugin discovery locations

Scope Path When
Project <project>/.vibe/plugins/<name>/ trusted folder only
User global ~/.vibe/plugins/<name>/ always

Each subdirectory under a plugins root is one plugin package. The first match wins: project plugins take precedence over user plugins with the same name.

Plugin directory structure

my-plugin/
  plugin.json                 # required — Agent Plugins 1.0 manifest
  mcp.json                    # optional — MCP server definitions
  libraries.json              # optional — Node/Python library dependencies
  connectors.json             # optional — managed connector requirements
  skills/                     # optional — one subdirectory per skill
    my-skill/
      SKILL.md
  ai.mistral.vibe/            # optional — Vibe-specific extensions
    hooks.toml                # Vibe extension only
    knowledge/                # one subdirectory per knowledge folder
      topic-name/
        KNOWLEDGE.md
    agents/                   # one .toml per subagent
      researcher.toml
    INSTRUCTIONS.md           # optional plugin-wide instructions (not loaded)

Step 1 — plugin.json (required)

The manifest is a JSON file with $schema and name as required fields.

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "my-plugin",
  "version": "1.0.0",
  "description": "What this plugin provides.",
  "author": {
    "name": "Author Name",
    "email": "author@example.com",
    "url": "https://example.com"
  },
  "homepage": "https://example.com/my-plugin",
  "repository": "https://github.com/example/my-plugin",
  "license": "MIT",
  "keywords": ["productivity", "finance"],
  "extensions": {}
}

Rules:

  • name must match ^[a-z0-9](?:[a-z0-9.-]*[a-z0-9])?$ (1-64 chars), no -- or ...
  • $schema must be exactly https://agent-plugins.org/schemas/1.0.0/plugin.schema.json. Any other schema URL produces a fatal plugin.schema.version_unsupported error.
  • Agent Plugins 1.0 defines no other top-level fields. Vibe ignores one with a plugin.manifest.unknown_field warning and loads the rest; put Vibe settings in the ai.mistral.vibe extension instead.

Step 2 — Vibe extension (optional)

To use Vibe-specific components (hooks, knowledge, agents, libraries, connectors), add an ai.mistral.vibe extension to plugin.json:

{
  "extensions": {
    "ai.mistral.vibe": {
      "schemaVersion": 1,
      "displayName": "My Plugin",
      "toolNamespace": "myPlugin",
      "toolOverrides": {
        "lookup": {
          "name": "search",
          "exposure": "direct_and_programmatic"
        }
      }
    }
  }
}

Without this extension, Vibe only loads plugin.json, skills/, and mcp.json. The toolNamespace defaults to a TypeScript identifier derived from the plugin name. It must not be one of the reserved namespaces: file_system, self, process, skill, subagent, vibe.

vibe is reserved for built-in plugins shipped with the CLI; user plugins must choose a different namespace.

displayName is an optional title of 1 to 100 characters; the plugin name stays its identity.

toolOverrides rename or restrict tool exposure. Each key matches a discovered tool name from MCP servers or connectors. exposure is one of programmatic, direct, or direct_and_programmatic.

Vibe ignores a key the extension's schema does not define, with a plugin.extension.unknown_field warning, and keeps the rest of the extension. An invalid value for a defined key drops the whole extension, and with it the plugin's hooks, knowledge, agents, libraries, and connectors.

Step 3 — Skills (optional)

Place skills under skills/ — one subdirectory per skill, each containing a SKILL.md. The skill name in SKILL.md frontmatter must match the directory name.

skills/
  format-reports/
    SKILL.md

Skills follow the standard SKILL.md format with YAML frontmatter (name, description, user-invocable, allowed-tools). Plugin skills are named <plugin-name>:<skill-name> at runtime, with the plugin's name as written (for example my-plugin:review). Only tools use the toolNamespace.

Step 4 — MCP servers (optional)

Define MCP servers in mcp.json. Three transport types are supported:

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/mcp.schema.json",
  "mcpServers": {
    "lookup": {
      "type": "stdio",
      "command": "node",
      "args": ["${PLUGIN_ROOT}/server/index.js"],
      "env": {
        "DEBUG": "true"
      },
      "cwd": "."
    },
    "remote-api": {
      "type": "streamable-http",
      "url": "https://api.example.com/mcp",
      "headers": {
        "X-Custom-Header": "value"
      }
    }
  }
}

Rules:

  • command, args, and env values support ${PLUGIN_ROOT} and ${PLUGIN_DATA} variable expansion.
  • PLUGIN_ROOT and PLUGIN_DATA are also injected automatically into env; do not set them yourself (rejected with an error).
  • env variable names PLUGIN_ROOT and PLUGIN_DATA are reserved.
  • cwd of . means the plugin root.
  • streamable-http servers support static headers only; use OAuth at the Vibe config level for authenticated remote servers.
  • sse transport is detected but not supported — it produces a diagnostic.
  • The command for stdio servers is resolved relative to the plugin root when it is a bare filename; absolute paths are used as-is.

Step 5 — Hooks (optional, Vibe extension only)

Place hooks in ai.mistral.vibe/hooks.toml. Format is identical to the Vibe hooks.toml format, but hooks run in the plugin root directory with PLUGIN_ROOT and PLUGIN_DATA in the environment. The session folder is the cwd field of the JSON the hook reads on stdin: read it to act on the user's project.

As for MCP servers, the plugin root is the session's read-only copy of the plugin, so editing the installed plugin does not change a running session, and PLUGIN_DATA is the session's data directory, shared with the plugin's MCP servers. Write state to PLUGIN_DATA, never next to the hook script.

[[hooks]]
name = "guard-lookup"
type = "pre_tool"
match = "myPlugin.lookup"
command = "python guard.py"
strict = true

[[hooks]]
name = "post-agent-lint"
type = "post_agent"
command = "python lint_session.py"   # reads `cwd` from stdin, then lints it

Plugin hooks run after the project's own hooks.toml and before the user's ~/.vibe/hooks.toml (project plugins), or after it (user plugins).

Limits:

  • Maximum 128 hooks per plugin.
  • File size limit: 64 KB.
  • Hook names must be unique within the plugin. Runtime names are prefixed as <plugin-name>:<hook-name>.

Step 6 — Knowledge (optional, Vibe extension only)

Place knowledge folders under ai.mistral.vibe/knowledge/. Each folder contains a KNOWLEDGE.md with YAML frontmatter:

ai.mistral.vibe/
  knowledge/
    accounting-policy/
      KNOWLEDGE.md
      reference.md

KNOWLEDGE.md frontmatter:

---
name: accounting-policy
description: How revenue is recognized and reported.
display_name: Accounting Policy
icon: book
---

Rules:

  • name must match the directory name (^[a-z0-9](?:[a-z0-9-]*[a-z0-9])?$).
  • description is 5-300 characters.
  • Maximum 100 knowledge folders per plugin.
  • KNOWLEDGE.md entrypoint limit: 256 KB.
  • No symbolic links inside knowledge directories.
  • Knowledge is named <plugin-name>:<name>.

Step 7 — Agents (optional, Vibe extension only)

Place subagent definitions under ai.mistral.vibe/agents/. Each .toml file defines one subagent:

ai.mistral.vibe/
  agents/
    researcher.toml

researcher.toml:

schemaVersion = 1
agentType = "subagent"
displayName = "Researcher"
description = "Searches the knowledge base and summarizes findings."
safety = "safe"
activeModel = "mistral-medium-3.5"
instructions = "You are a research assistant."

enabledTools = ["read_file", "grep"]
disabledTools = ["bash"]

[tools.bash]
permission = "never"
allowlist = ["ls"]
denylist = ["rm"]

Rules:

  • schemaVersion must be 1.
  • agentType must be "subagent" (the only supported type).
  • Filename must be lowercase kebab-case (^[a-z0-9](?:[a-z0-9-]*[a-z0-9])?$).
  • safety is one of "safe", "neutral", "destructive", "yolo".
  • Maximum 128 agents per plugin, 64 KB per file.
  • Tool lists cannot contain duplicates or empty names.
  • Agent names are <plugin-name>:<filename-stem>.

Step 8 — Libraries (optional, Vibe extension only)

Declare Node and Python library dependencies in libraries.json:

{
  "schemaVersion": 1,
  "node": {
    "@productivity/finance": "./node/finance",
    "lodash": "./node/lodash"
  },
  "python": {
    "vibe_finance": "./python/vibe_finance",
    "utils": "./python/utils.py"
  }
}

Rules:

  • Node aliases must match ^(?:@[a-z0-9][a-z0-9._-]*/)?[a-z0-9][a-z0-9._-]*$ (max 214 chars).
  • Python aliases must match ^[A-Za-z_][A-Za-z0-9_]*$ (max 128 chars).
  • Source paths are relative to the plugin root, using forward slashes only. No .. or absolute paths.
  • Python library sources can be a directory (package) or a single .py file.
  • Node library sources must be directories.
  • No symbolic links inside library directories.
  • Library aliases that collide across plugins are all dropped.
  • Maximum 128 entries per language.

Step 9 — Connectors (optional, Vibe extension only)

Declare required managed connectors in connectors.json:

{
  "schemaVersion": 1,
  "connectors": [
    {
      "id": "github",
      "tools": ["create_issue", "list_issues"]
    },
    {
      "id": "linear",
      "tools": ["create_issue"]
    }
  ]
}

Rules:

  • Connector IDs must be unique within the plugin.
  • Tool names per connector must be unique, 1-256 characters.
  • Maximum 128 connector requirements.
  • A connector requirement is a declaration: it appears whether or not the account has the connector. Availability is reported as a diagnostic.

Diagnostic codes

When a plugin fails to load or has issues, Vibe emits typed diagnostics. The most common ones:

Code Fatal Meaning
plugin.manifest.invalid yes plugin.json missing, unreadable, or invalid
plugin.schema.version_unsupported yes unsupported schema version
plugin.compatibility.format_unrecognized yes no supported manifest found
plugin.namespace.reserved yes namespace is reserved
plugin.namespace.collision yes two plugins claim one namespace
plugin.name.collision yes two plugins share a name at the same scope
plugin.path.outside_root yes a declared path escapes the plugin root
plugin.manifest.unknown_field no plugin.json has a field Agent Plugins 1.0 does not define; it is ignored
plugin.extension.unknown_field no the ai.mistral.vibe extension has a key its schema does not define; it is ignored
plugin.skill.invalid no one SKILL.md failed to parse
plugin.hooks.invalid no one hook entry failed validation
plugin.knowledge.invalid no one knowledge folder failed validation
plugin.agent.invalid no one agent document failed validation
plugin.libraries.invalid no libraries.json failed to load
plugin.library.invalid no one library path is invalid
plugin.library.alias_collision no library alias shared across plugins
plugin.connectors.invalid no connectors.json failed to load
plugin.mcp.connection_failed no an MCP server did not answer
plugin.mcp.authorization_required no an MCP server needs authorization
plugin.tool_override.unused no a toolOverrides key matched no tool

Fatal diagnostics drop the entire plugin. Non-fatal diagnostics drop only the offending component; the rest of the plugin survives.

Compatibility with other plugin formats

Vibe also detects and adapts plugins authored for other agents:

Format Marker Adapted as
Claude Code .claude-plugin/plugin.json skills, MCP servers, hooks
Codex .codex-plugin/plugin.json skills, MCP servers
Kimi Code kimi.plugin.json or .kimi-plugin/plugin.json skills, MCP servers, hooks
OpenCode .opencode/ directory skills only (executable modules unsupported)

If markers for more than one format are present, the plugin is rejected as ambiguous. Native Agent Plugins 1.0 (a valid plugin.json with the correct $schema) always takes precedence.

When creating a new plugin, always use the native Agent Plugins 1.0 format.

Workflow

  1. Determine the plugin name, scope (project vs user global), and which components the user needs.
  2. Create the directory at .vibe/plugins/<name>/ (project) or ~/.vibe/plugins/<name>/ (user).
  3. Write plugin.json with the required $schema and name fields.
  4. Add the ai.mistral.vibe extension to plugin.json if any Vibe-specific components (hooks, knowledge, agents, libraries, connectors) are needed.
  5. Add component directories and files per the instructions above.
  6. Tell the user to run /reload to pick up the new plugin without restarting.

04 同仓库更多 Skills

mistralai
mistralai/mistral-vibe

import

Import the user's setup from other coding CLIs (Claude Code, Codex, OpenCode) into Vibe — instructions, skills, subagents, commands, hooks, MCP servers, permissions, and plugins. Load when the user wants to switch to Vibe, import or reuse another CLI's configuration, or asks whether a file from another CLI works in Vibe.

Skill与智能体开发
★ 5.4k
mistralai
mistralai/mistral-vibe

worktree

Create, reuse, and clean up git worktrees under $VIBE_HOME/worktrees. Use when the user wants to work in an isolated git worktree, create a worktree for a task, or list/remove existing worktrees.

★ 5.4k
mistralai
mistralai/mistral-vibe

vibe

Authoritative reference for Mistral Vibe — the CLI agent you (the model) are running inside. LOAD when the user: - asks anything about Vibe itself, even by indirect name ("this CLI", "this tool", "you"); - wants to change, inspect, or reset their setup; - asks why the agent did or did not act; - asks how to make the CLI do X, where X lives, or what a flag/command/setting does; - asks any meta question about your own behavior; - is unsure whether a command, flag, env var, or file is in scope — this skill is the source of truth. SCOPE: config under `~/.vibe/` and project-local `.vibe/`; `VIBE_*` and `LOG_*` env vars; models and providers; agents and subagents; skills; tools and their permission model; every slash command and CLI flag; hooks; MCP servers; connectors; plugins; trusted folders; `@`-file mentions; logs; themes; voice.

Skill与智能体开发
★ 5.4k
mistralai
mistralai/mistral-vibe

skill-creator

Load this skill before creating, updating, or deleting a Vibe skill, or whenever you plan to add or modify SKILL.md files under a skills directory. It explains Vibe's SKILL.md frontmatter, discovery order, support files, and the permission flow.

Skill与智能体开发
★ 5.4k
mistralai
mistralai/mistral-vibe

prompts

暂无简介

★ 5.4k
mistralai
mistralai/mistral-vibe

write-vibe-tests

Write or refactor Mistral Vibe tests with proper decoupling. Use when adding behavior coverage, testing ports/adapters, replacing brittle mocks, creating fakes, adding characterization tests before refactors, or changing tests under tests/ for vibe/core, vibe/app_server, vibe/acp, tools, config, sessions, skills, hooks, MCP, or setup.

代码审查测试与质量
★ 5.4k