Skill Memory

Routes tasks to preferred skills with preset parameters for images, videos, screenshots, TTS, news aggregation, and trending topics.

Skill Memory / 技能记忆

Self-learning task-to-skill routing table with automatic recording and updating. 自学习型技能记忆表，支持自动记录和更新。

How It Works / 工作机制

1. Check Before Use / 使用前检查

When executing a task, check memory table first:

text

Task → Look up memory.json → Found? → Use preset params
                              ↓ Not found
                         Use default → Record to memory

2. Auto Record / 自动记录

When using a skill not in memory:

json

{
  "task_type": "新任务类型",
  "skill": "技能名称",
  "parameters": {...},
  "scenario": "使用场景描述",
  "created_at": "2026-03-27T13:00:00+08:00",
  "usage_count": 1
}

3. Auto Update / 自动更新

When same task type but different scenario/params:

json

{
  "task_type": "网页截图",
  "skill": "Playwright CDP",
  "parameters": {"resolution": "1920x1080"},  // 新参数
  "scenario": "高清截图",
  "updated_at": "2026-03-27T14:00:00+08:00",
  "usage_count": 5
}

Memory File Location / 记忆文件位置

Memory is stored in workspace:

text

~/.openclaw/workspace/skills/skill-memory/references/memory.json

Core Mapping Table / 核心映射表

Task Type	Preferred Skill	Preset Parameters
Analyze Image / 分析图片	doubao-api-toolkit	Vision analysis
Analyze Video / 分析视频	doubao-video-analyzer	Doubao 2.0 model
Web Screenshot / 网页截图	Playwright CDP	1280×1024, 20s wait
Text-to-Image / 文生图	doubao-api-toolkit	Doubao API
Image-to-Image / 图生图	doubao-api-toolkit	Doubao API
Text-to-Video / 文生视频	doubao-api-toolkit	Doubao API
TTS Voice / TTS配音	mambo-tts	Mambo +50Hz
News Aggregation / 新闻聚合	news-aggregator-skill	28-source
Weibo Hot Search / 微博热搜	weibo-hot-search	Hot + Entertainment

Web Screenshot Details / 网页截图详细参数

Technical Solution: Playwright connects to existing Chrome (CDP port 9222)

Preset Parameters / 预设参数:

text

Wait Time / 等待时间: 20 seconds (page load)
Resolution / 分辨率: 1280×1024
Capture Scope / 截取范围: Webpage content only
Style / 风格要求: Clean, no scrollbars, no borders

Usage / 调用方式:

python

# Connect to existing Chrome / 连接已有Chrome
browser = await playwright.chromium.connect_over_cdp("http://localhost:9222")
page = browser.contexts[0].pages[0]
await page.wait_for_timeout(20000)  # Wait 20s / 等待20秒
await page.screenshot(path="output.png", full_page=False)

TTS Voice Presets / TTS音色预设

Preset	Voice	Pitch	Use Case / 适用场景
Mambo / 曼波	zh-CN-XiaoyiNeural	+50Hz	Lively, energetic / 活泼有活力
Xiaoyi / 晓伊	zh-CN-XiaoyiNeural	default	Fast-paced, news / 快节奏新闻
Xiaoxiao / 晓晓	zh-CN-XiaoxiaoNeural	default	Natural, general / 自然通用
Yunyang / 云扬	zh-CN-YunyangNeural	default	Narration, documentary / 旁白纪录片

Update Script / 更新脚本

Use the update script to record or update skill memory:

bash

python scripts/update_memory.py --task "任务类型" --skill "技能名称" --params '{"key":"value"}' --scenario "场景描述"

Script Options / 脚本参数

Option	Description
--task	Task type (任务类型)
--skill	Skill name (技能名称)
--params	JSON parameters (参数JSON)
--scenario	Usage scenario (使用场景)
--list	List all memory entries (列出所有记忆)

Usage Workflow / 使用流程

Step 1: Check Memory / 检查记忆

bash

python scripts/update_memory.py --list

Step 2: Execute Task / 执行任务

• If task in memory → Use preset params
• If task NOT in memory → Use best skill → Record to memory

Step 3: Auto Update / 自动更新

• Same task, new scenario → Update memory entry
• Increment usage_count automatically

Example Scenarios / 示例场景

Scenario 1: New Task Type / 新任务类型

text

User: "帮我分析这个音频"
  ↓
Check memory: No "音频分析" entry
  ↓
Use best skill: doubao-api-toolkit (audio feature)
  ↓
Record to memory:
{
  "task_type": "分析音频",
  "skill": "doubao-api-toolkit",
  "parameters": {"mode": "audio"},
  "scenario": "音频内容分析",
  "usage_count": 1
}

Scenario 2: Updated Parameters / 参数更新

text

User: "截个高清网页图，要1920x1080"
  ↓
Check memory: Found "网页截图" but params differ
  ↓
Use new params: 1920x1080
  ↓
Update memory:
{
  "task_type": "网页截图",
  "parameters": {"resolution": "1920x1080", "high_res": true},
  "scenario": "高清截图",
  "updated_at": "...",
  "usage_count": 6  // incremented
}

Extension Rules / 扩展规则

• If task type not in table, use default skill selection
• User's explicit skill choice overrides table mapping
• Preset parameters can be overridden by user specification
• Always record new skill usage to memory
• Always update memory when scenario/params differ