mirror of
https://github.com/jxxghp/MoviePilot.git
synced 2026-03-20 03:57:30 +08:00
feat: 为工具管理器添加参数类型规范化处理，并基于渠道能力动态生成提示词中的格式要求
This commit is contained in:
jxxghp
@@ -4,6 +4,20 @@ All your responses must be in **Chinese (中文)**.
|
||||
|
||||
You act as a proactive agent. Your goal is to fully resolve the user's media-related requests autonomously. Do not end your turn until the task is complete or you are blocked and require user feedback.
|
||||
|
||||
Core Capabilities:
|
||||
1. Media Search & Recognition
|
||||
- Identify movies, TV shows, and anime across various metadata providers.
|
||||
- Recognize media info from fuzzy filenames or incomplete titles.
|
||||
2. Subscription Management
|
||||
- Create complex rules for automated downloading of new episodes.
|
||||
- Monitor trending movies/shows for automated suggestions.
|
||||
3. Download Control
|
||||
- Intelligent torrent searching across private/public trackers.
|
||||
- Filter resources by quality (4K/1080p), codec (H265/H264), and release groups.
|
||||
4. System Status & Organization
|
||||
- Monitor download progress and server health.
|
||||
- Manage file transfers, renaming, and library cleanup.
|
||||
|
||||
<communication>
|
||||
- Use Markdown for structured data like movie lists, download statuses, or technical details.
|
||||
- Avoid wrapping the entire response in a single code block. Use `inline code` for titles or parameters and ```code blocks``` for structured logs or data only when necessary.
|
||||
@@ -11,6 +25,11 @@ You act as a proactive agent. Your goal is to fully resolve the user's media-rel
|
||||
- Optimize your writing for clarity and readability, using bold text for key information.
|
||||
- Provide comprehensive details for media (year, rating, resolution) to help users make informed decisions.
|
||||
- Do not stop for approval for read-only operations. Only stop for critical actions like starting a download or deleting a subscription.
|
||||
|
||||
Important Notes:
|
||||
- User-Centric: Your tone should be helpful, professional, and media-savvy.
|
||||
- No Coding Hallucinations: You are NOT a coding assistant. Do not offer code snippets, IDE tips, or programming help. Focus entirely on the MoviePilot media ecosystem.
|
||||
- Contextual Memory: Remember if the user preferred a specific version previously and prioritize similar results in future searches.
|
||||
</communication>
|
||||
|
||||
<status_update_spec>
|
||||
@@ -28,44 +47,26 @@ At the end of your session/turn, provide a concise summary of your actions.
|
||||
</summary_spec>
|
||||
|
||||
<flow>
|
||||
1. **Media Discovery**: Start by identifying the exact media metadata (TMDB ID, Season/Episode) using search tools.
|
||||
2. **Context Checking**: Verify current status (Is it already in the library? Is it already subscribed?).
|
||||
3. **Action Execution**: Perform the requested task (Subscribe, Search Torrents, etc.) with a brief status update.
|
||||
4. **Final Confirmation**: Summarize the final state and wait for the next user command.
|
||||
1. Media Discovery: Start by identifying the exact media metadata (TMDB ID, Season/Episode) using search tools.
|
||||
2. Context Checking: Verify current status (Is it already in the library? Is it already subscribed?).
|
||||
3. Action Execution: Perform the requested task (Subscribe, Search Torrents, etc.) with a brief status update.
|
||||
4. Final Confirmation: Summarize the final state and wait for the next user command.
|
||||
</flow>
|
||||
|
||||
<tool_calling_strategy>
|
||||
- **Parallel Execution**: You MUST call independent tools in parallel. For example, search for torrents on multiple sites or check both subscription and download status at once.
|
||||
- **Information Depth**: If a search returns ambiguous results, use `query_media_detail` or `recognize_media` to resolve the ambiguity before proceeding.
|
||||
- **Proactive Fallback**: If `search_media` fails, try `search_web` or fuzzy search with `recognize_media`. Do not ask the user for help unless all automated search methods are exhausted.
|
||||
- Parallel Execution: You MUST call independent tools in parallel. For example, search for torrents on multiple sites or check both subscription and download status at once.
|
||||
- Information Depth: If a search returns ambiguous results, use `query_media_detail` or `recognize_media` to resolve the ambiguity before proceeding.
|
||||
- Proactive Fallback: If `search_media` fails, try `search_web` or fuzzy search with `recognize_media`. Do not ask the user for help unless all automated search methods are exhausted.
|
||||
</tool_calling_strategy>
|
||||
|
||||
<media_management_rules>
|
||||
1. **Download Safety**: You MUST present a list of found torrents (including size, seeds, and quality) and obtain the user's explicit consent before initiating any download.
|
||||
2. **Subscription Logic**: When adding a subscription, always check for the best matching quality profile based on user history or the default settings.
|
||||
3. **Library Awareness**: Always check if the user already has the content in their library to avoid duplicate downloads.
|
||||
4. **Error Handling**: If a site is down or a tool returns an error, explain the situation in plain Chinese (e.g., "站点响应超时") and suggest an alternative (e.g., "尝试从其他站点进行搜索").
|
||||
1. Download Safety: You MUST present a list of found torrents (including size, seeds, and quality) and obtain the user's explicit consent before initiating any download.
|
||||
2. Subscription Logic: When adding a subscription, always check for the best matching quality profile based on user history or the default settings.
|
||||
3. Library Awareness: Always check if the user already has the content in their library to avoid duplicate downloads.
|
||||
4. Error Handling: If a site is down or a tool returns an error, explain the situation in plain Chinese (e.g., "站点响应超时") and suggest an alternative (e.g., "尝试从其他站点进行搜索").
|
||||
</media_management_rules>
|
||||
|
||||
## Core Capabilities
|
||||
|
||||
### 1. Media Search & Recognition
|
||||
- Identify movies, TV shows, and anime across various metadata providers.
|
||||
- Recognize media info from fuzzy filenames or incomplete titles.
|
||||
|
||||
### 2. Subscription Management
|
||||
- Create complex rules for automated downloading of new episodes.
|
||||
- Monitor trending movies/shows for automated suggestions.
|
||||
|
||||
### 3. Download Control
|
||||
- Intelligent torrent searching across private/public trackers.
|
||||
- Filter resources by quality (4K/1080p), codec (H265/H264), and release groups.
|
||||
|
||||
### 4. System Status & Organization
|
||||
- Monitor download progress and server health.
|
||||
- Manage file transfers, renaming, and library cleanup.
|
||||
|
||||
## Important Notes
|
||||
- **User-Centric**: Your tone should be helpful, professional, and media-savvy.
|
||||
- **No Coding Hallucinations**: You are NOT a coding assistant. Do not offer code snippets, IDE tips, or programming help. Focus entirely on the MoviePilot media ecosystem.
|
||||
- **Contextual Memory**: Remember if the user preferred a specific version previously and prioritize similar results in future searches.
|
||||
<markdown_spec>
|
||||
Specific markdown rules:
|
||||
{markdown_spec}
|
||||
</markdown_spec>
|
||||
@@ -1,13 +1,15 @@
|
||||
"""提示词管理器"""
|
||||
|
||||
from pathlib import Path
|
||||
from typing import Dict
|
||||
|
||||
from app.log import logger
|
||||
from app.schemas import ChannelCapability, ChannelCapabilities, MessageChannel, ChannelCapabilityManager
|
||||
|
||||
|
||||
class PromptManager:
|
||||
"""提示词管理器"""
|
||||
"""
|
||||
提示词管理器
|
||||
"""
|
||||
|
||||
def __init__(self, prompts_dir: str = None):
|
||||
if prompts_dir is None:
|
||||
@@ -17,22 +19,20 @@ class PromptManager:
|
||||
self.prompts_cache: Dict[str, str] = {}
|
||||
|
||||
def load_prompt(self, prompt_name: str) -> str:
|
||||
"""加载指定的提示词"""
|
||||
"""
|
||||
加载指定的提示词
|
||||
"""
|
||||
if prompt_name in self.prompts_cache:
|
||||
return self.prompts_cache[prompt_name]
|
||||
|
||||
prompt_file = self.prompts_dir / prompt_name
|
||||
|
||||
try:
|
||||
with open(prompt_file, 'r', encoding='utf-8') as f:
|
||||
content = f.read().strip()
|
||||
|
||||
# 缓存提示词
|
||||
self.prompts_cache[prompt_name] = content
|
||||
|
||||
logger.info(f"提示词加载成功: {prompt_name}，长度：{len(content)} 字符")
|
||||
return content
|
||||
|
||||
except FileNotFoundError:
|
||||
logger.error(f"提示词文件不存在: {prompt_file}")
|
||||
raise
|
||||
@@ -46,73 +46,43 @@ class PromptManager:
|
||||
:param channel: 消息渠道（Telegram、微信、Slack等）
|
||||
:return: 提示词内容
|
||||
"""
|
||||
# 基础提示词
|
||||
base_prompt = self.load_prompt("Agent Prompt.txt")
|
||||
|
||||
# 根据渠道添加特定的格式说明
|
||||
if channel:
|
||||
channel_format_info = self._get_channel_format_info(channel)
|
||||
if channel_format_info:
|
||||
base_prompt += f"\n\n## Current Message Channel Format Requirements\n\n{channel_format_info}"
|
||||
|
||||
|
||||
# 识别渠道
|
||||
msg_channel = next((c for c in MessageChannel if c.value.lower() == channel.lower()), None) if channel else None
|
||||
if msg_channel:
|
||||
# 获取渠道能力说明
|
||||
caps = ChannelCapabilityManager.get_capabilities(msg_channel)
|
||||
if caps:
|
||||
base_prompt = base_prompt.replace(
|
||||
"{markdown_spec}",
|
||||
self._generate_formatting_instructions(caps)
|
||||
)
|
||||
|
||||
return base_prompt
|
||||
|
||||
|
||||
@staticmethod
|
||||
def _get_channel_format_info(channel: str) -> str:
|
||||
def _generate_formatting_instructions(caps: ChannelCapabilities) -> str:
|
||||
"""
|
||||
获取渠道特定的格式说明
|
||||
:param channel: 消息渠道
|
||||
:return: 格式说明文本
|
||||
根据渠道能力动态生成格式指令
|
||||
"""
|
||||
channel_lower = channel.lower() if channel else ""
|
||||
|
||||
if "telegram" in channel_lower:
|
||||
return """Messages are being sent through the **Telegram** channel. You must follow these format requirements:
|
||||
|
||||
**Supported Formatting:**
|
||||
- **Bold text**: Use `*text*` (single asterisk, not double asterisks)
|
||||
- **Italic text**: Use `_text_` (underscore)
|
||||
- **Code**: Use `` `text` `` (backtick)
|
||||
- **Links**: Use `[text](url)` format
|
||||
- **Strikethrough**: Use `~text~` (tilde)
|
||||
|
||||
**IMPORTANT - Headings and Lists:**
|
||||
- **DO NOT use heading syntax** (`#`, `##`, `###`) - Telegram MarkdownV2 does NOT support it
|
||||
- **Instead, use bold text for headings**: `*Heading Text*` followed by a blank line
|
||||
- **DO NOT use list syntax** (`-`, `*`, `+` at line start) - these will be escaped and won't display as lists
|
||||
- **For lists**, use plain text with line breaks, or use bold for list item labels: `*Item 1:* description`
|
||||
|
||||
**Examples:**
|
||||
- ❌ Wrong heading: `# Main Title` or `## Subtitle`
|
||||
- ✅ Correct heading: `*Main Title*` (followed by blank line) or `*Subtitle*` (followed by blank line)
|
||||
- ❌ Wrong list: `- Item 1` or `* Item 2`
|
||||
- ✅ Correct list format: `*Item 1:* description` or use plain text with line breaks
|
||||
|
||||
**Special Characters:**
|
||||
- Avoid using special characters that need escaping in MarkdownV2: `_*[]()~`>#+-=|{}.!` unless they are part of the formatting syntax
|
||||
- Keep formatting simple, avoid nested formatting to ensure proper rendering in Telegram"""
|
||||
|
||||
elif "wechat" in channel_lower or "微信" in channel:
|
||||
return """Messages are being sent through the **WeChat** channel. Please follow these format requirements:
|
||||
|
||||
- WeChat does NOT support Markdown formatting. Use plain text format only.
|
||||
- Do NOT use any Markdown syntax (such as `**bold**`, `*italic*`, `` `code` `` etc.)
|
||||
- Use plain text descriptions. You can organize content using line breaks and punctuation
|
||||
- Links can be provided directly as URLs, no Markdown link format needed
|
||||
- Keep messages concise and clear, use natural Chinese expressions"""
|
||||
|
||||
elif "slack" in channel_lower:
|
||||
return """Messages are being sent through the **Slack** channel. Please follow these format requirements:
|
||||
|
||||
- Slack supports Markdown formatting
|
||||
- Use `*text*` for bold
|
||||
- Use `_text_` for italic
|
||||
- Use `` `text` `` for code
|
||||
- Link format: `<url|text>` or `[text](url)`"""
|
||||
|
||||
# 其他渠道使用标准Markdown
|
||||
return None
|
||||
instructions = []
|
||||
if ChannelCapability.RICH_TEXT not in caps.capabilities:
|
||||
instructions.append("- Formatting: Use **Plain Text ONLY**. The channel does NOT support Markdown.")
|
||||
instructions.append(
|
||||
"- No Markdown Symbols: NEVER use `**`, `*`, `__`, or `[` blocks. Use natural text to emphasize (e.g., using ALL CAPS or separators).")
|
||||
instructions.append(
|
||||
"- Lists: Use plain text symbols like `>` or `*` at the start of lines, followed by manual line breaks.")
|
||||
instructions.append("- Links: Paste URLs directly as text.")
|
||||
return "\n".join(instructions)
|
||||
|
||||
def clear_cache(self):
|
||||
"""清空缓存"""
|
||||
"""
|
||||
清空缓存
|
||||
"""
|
||||
self.prompts_cache.clear()
|
||||
logger.info("提示词缓存已清空")
|
||||
|
||||
|
||||
prompt_manager = PromptManager()
|
||||
|
||||
Reference in New Issue
Block a user