pmem-style-extraction

Style Extraction

Extract writing style features from multimodal input and store them in the powermem knowledge base, supporting intelligent conflict detection and manual arbitration.

Core Process

Execute the following steps to extract and store style features:

0. Environment Preparation (First Execution)

Before running any script, first prepare .venv. The following approach does not depend on a fixed absolute path:

ROOT_DIR="$(cd ../.. && pwd)"
PYTHON_BIN="$ROOT_DIR/.venv/bin/python"
test -d "$ROOT_DIR/.venv" || python3 -m venv "$ROOT_DIR/.venv"
"$PYTHON_BIN" -m pip install -U pip
"$PYTHON_BIN" -m pip install -U powermem python-dotenv

All subsequent Python scripts use $PYTHON_BIN for execution, without depending on whether the current shell has activated the virtual environment.

Configuration Management Notes：

On first use, the script will automatically create the ../pmem-config/ shared configuration directory
containing pmem-key.env(user configuration) and .env(full template)
Configuration priority: system environment variables > pmem-config/pmem-key.env > pmem-config/.env
The two skills (extraction and generation) share the same configuration, so configuring once is sufficient

0.1 Configuration Detection (Required Step)

Configuration detection must be executed before each run：

"$PYTHON_BIN" scripts/check_config.py

If configuration detection fails (exit code is 1)：

Detection output description (fixed text is retained in the Skill; only missing items are dynamically output by the script)：

# 运行配置检测（脚本会动态输出“当前缺失的配置项”）
"$PYTHON_BIN" scripts/check_config.py

Output structure example (the list of missing items is subject to the script's actual detection results):

⚠️  检测到以下配置项缺失或未设置:

[这里的缺失项列表由 check_config.py 动态输出]

📝 共享配置文件位置: /path/to/skills/pmem-config/pmem-key.env

请点击上面的共享配置文件路径，进入文件夹后手动编辑文件：`pmem-key.env`。
手动填写完成后，返回聊天框回复：已完成

The Agent should perform the following operations：
- Clearly inform the user of the shared configuration directory path:../pmem-config/
- Clearly inform the user of the filename that needs to be edited:pmem-key.env
- Do not collect configuration values item by item in the conversation; only wait for the user to fill them in manually
- After the user replies "completed", rerun check_config.py to verify the configuration

Configuration writing example：

# 用户手动打开并编辑：../pmem-config/pmem-key.env
# 在文件中填写缺失项，例如：
LLM_API_KEY=sk-xxxxx
EMBEDDING_API_KEY=sk-yyyyy

# 重新检测
"$PYTHON_BIN" scripts/check_config.py

Important Notes：
- The configuration will be persistently saved in ../pmem-config/pmem-key.env
- There is no need to reconfigure the next time either skill is used
- Users can also manually edit this file to modify the configuration

After configuration detection passes, continue to the subsequent steps.

1. Obtain Content

Choose the processing method based on the input type:

URL input: call baoyu-url-to-markdown to obtain the webpage content

SKILL_URL_DIR="$HOME/.cursor/skills/baoyu-url-to-markdown"
npx -y bun ${SKILL_URL_DIR}/scripts/main.ts <url> -o /tmp/extracted_content.md

Then read /tmp/extracted_content.md as the input for subsequent analysis.

Text/file input: directly use the provided content.

Skill file input: read the content of the SKILL.md file.

Image input: use visual capabilities to extract the text layout and visual style elements from the image.

2. Feature Extraction

Use an LLM to strip factual information from the content, retaining only the style skeleton. Use the following Prompt template:

你是一位专业的写作风格分析师。请从下面的内容中提取纯粹的风格特征，忽略所有具体的事实、观点和信息内容。

请从以下维度提取风格特征：

1. **叙事节奏 (narrative_rhythm)**：句子长度偏好、段落节奏、信息密度
2. **金句位置 (key_phrase_placement)**：强调内容的位置习惯（段首/段中/段尾）
3. **排版符号 (formatting_symbols)**：特殊引号、列表符号、分隔符使用习惯
4. **标点习惯 (punctuation_habits)**：省略号、破折号、感叹号等使用频率和场景
5. **段落结构 (paragraph_structure)**：段落组织方式（总分总、递进、并列等）
6. **情感基调 (emotional_tone)**：理性/感性、克制/热烈、严肃/幽默等
7. **修辞手法 (rhetorical_devices)**：比喻、排比、反问等惯用修辞
8. **开篇风格 (opening_style)**：如何引入话题（直入主题/铺垫/引用/提问）
9. **结尾风格 (closing_style)**：如何收束内容（总结/升华/留白/号召）

请以 JSON 格式输出，每个维度用简洁的描述性语言概括：

{
  "narrative_rhythm": "描述",
  "key_phrase_placement": "描述",
  "formatting_symbols": "描述",
  "punctuation_habits": "描述",
  "paragraph_structure": "描述",
  "emotional_tone": "描述",
  "rhetorical_devices": "描述",
  "opening_style": "描述",
  "closing_style": "描述"
}

待分析内容：
---
{CONTENT}
---

Save the extracted features in JSON format (for example /tmp/extracted_features.json），Then automatically proceed to the next step of conflict detection。

3. Automatic Conflict Detection

No user action required; the system automatically uses the script to search for similar Profiles that already exist in powermem and detects whether conflicts exist:

"$PYTHON_BIN" scripts/search_profiles.py \
  --features /tmp/extracted_features.json \
  --threshold 0.85

The script will output:

Whether a similar Profile is found
Similarity score
Whether conflicting features exist
Conflict details (if any)

4. Manual arbitration (only when there is a conflict)

Pause only when a conflict is detected, show the comparison to the user and wait for a choice:

检测到风格特征冲突：

【已存 Profile: default】
- 叙事节奏: 短句为主，每段2-4行
- 情感基调: 理性克制

【新提取特征】
- 叙事节奏: 长句排比，情绪化段落
- 情感基调: 热烈煽情

请选择处理方式：
1. 抛弃 - 仅作参考，不写入
2. 覆盖 - 用新特征替换旧 Profile
3. 分支 - 保留旧 Profile，新建独立风格变体（需指定名称）

Wait for the user to enter a choice (1/2/3).

If there is no conflict, skip directly to step 5 to collect metadata。

5. Interactively collect metadata

Regardless of whether there is a conflict, after determining the storage mode, immediately proceed to metadata collection, asking the user one by one through dialogue:

Interaction flow(the Agent asks one by one, and after the user answers one, asks the next):

Step 1/6 - Applicable platform：

Agent: 📱 这个风格适合哪些平台？请从以下选项中选择（可多选，用逗号分隔）：
1. 小红书
2. Twitter / X
3. 公众号
4. 知乎
5. LinkedIn
6. 通用（不限平台）
7. 其他（请说明）

请输入序号或平台名称（例如：2,3 或 Twitter,公众号）