astrbot-plugin-dev-skill（astrbot 插件开发技能）

AstrBot Plugin Development

Use this skill to write AstrBot plugins in a reviewer-first way: align with astr-plugin-reviewer hard checks, then follow the current AstrBot repository APIs and docs.

Start Here

Before writing code, always read these two references first:

• references/reviewer-checklist.md: hard constraints from astr-plugin-reviewer and plugin-market submission checks.
• references/project-structure.md: required files, metadata rules, local dev flow, and publishing expectations.

Then load only the references you need:

• references/core-api.md: imports, decorators, handler signatures, hook constraints, platform compatibility.
• references/advanced-features.md: config schema, session control, LLM tools, direct LLM calls, agents, and T2I.
• references/message-components.md: message-chain composition, passive replies, and proactive messages.
• references/patterns.md: reviewer-friendly implementation patterns, persistence, async networking, and platform access.

Default Workflow

1. Create or verify main.py and metadata.yaml first.
2. Treat metadata.yaml as the source of truth for plugin identity. Prefer desc plus repo, and never keep both desc and description.
3. In main.py, define a class that inherits Star. Prefer AstrBot's auto-discovery; do not introduce the deprecated @register decorator unless you are maintaining old code.
4. Import filter exactly with from astrbot.api.event import filter to avoid reviewer failures and naming confusion.
5. Import the logger exactly with from astrbot.api import logger.
6. Keep network I/O async. Prefer httpx or aiohttp; do not use requests, blocking sleeps, or other blocking network calls.
7. If the plugin needs persistent files, prefer StarTools.get_data_dir(). It returns a Path.
8. If you implement LLM hooks, LLM tools, direct LLM calls, or agents, follow the exact signatures and restrictions in references/advanced-features.md.
9. Before finishing, run a self-check against references/reviewer-checklist.md. If the user wants marketplace publishing, also ensure the publish JSON matches metadata.yaml exactly.

Minimal Template

from pathlib import Path

from astrbot.api import logger
from astrbot.api.event import AstrMessageEvent, filter
from astrbot.api.star import Context, Star, StarTools


class MyPlugin(Star):
    def __init__(self, context: Context):
        super().__init__(context)
        self.data_dir: Path = StarTools.get_data_dir()

    @filter.command("helloworld")
    async def helloworld(self, event: AstrMessageEvent):
        """回复一个简单问候。"""
        logger.info(f"helloworld triggered by {event.get_sender_id()}")
        yield event.plain_result(f"Hello, {event.get_sender_name()}!")

    async def terminate(self):
        """Called when the plugin is unloaded or disabled."""

Note: The @register decorator is deprecated in newer versions of AstrBot. Please use metadata.yaml to define plugin metadata. AstrBot automatically detects the plugin class inheriting from Star.

Core Workflows

1. Project Setup and Metadata

A complete plugin requires metadata.yaml for identification, requirements.txt for dependencies, and optionally logo.png, _conf_schema.json, and a README.md.

• Plugin names should start with astrbot_plugin_, be lowercase, have no spaces, and be short.
• See references/project-structure.md for mandatory files, dev environment setup, and publishing.

2. Registering Commands and Filters

Commands are registered using @filter.command(name). AstrBot auto-parses command parameters by type hints. You can also use command groups, command aliases, and filter by event type, platform, or user permission.

• See references/core-api.md for full list of filters, hooks, the platform compatibility matrix, and event propagation control.

3. Handling Messages and Responses

AstrBot uses a message-chain system. You can respond with plain text, images, or a mix of components. Proactive messages are supported via unified_msg_origin and MessageChain.

• See references/message-components.md for how to build and send messages.

4. Advanced Integrations

• Configuration: Use _conf_schema.json for user settings.
• LLM Tools: Register tools via @filter.llm_tool or FunctionTool.
• LLM Direct Calls: Use self.context.llm_generate() to call LLMs directly.
• Agent / Multi-Agent: Use self.context.tool_loop_agent() for tool-loop agents.
• Stateful Interaction: Use session_waiter for multi-step prompts with custom session filters.
• T2I: Render text or HTML/Jinja2 templates to images.
• Conversation & Persona Managers: Access LLM conversation history and persona settings when needed.

See references/advanced-features.md for examples.

Elegant Design Patterns

Follow these patterns for robust, user-friendly plugins:

• Use unified logging via from astrbot.api import logger.
• Handle errors gracefully to avoid bot crashes.
• Use KV storage or the plugin data directory for persistence.
• Ensure all I/O operations are non-blocking.
• Access platform instances, loaded plugins, and protocol-level APIs only when necessary.

See references/patterns.md for detailed code patterns.