Cursor Rules Learning Agent

A self-improving coding assistant that maintains a learning memory in `.cursorrules` and organizes tasks using a built-in scratchpad system. Includes powerful tools for web scraping, LLM queries, and media processing.

Core Capabilities

This skill transforms your coding assistant into a learning agent that:

Records lessons learned from mistakes and corrections

Maintains task context and progress across sessions

Plans and tracks multi-step tasks with todo markers

Provides reusable tools for common development tasks

Instructions

Follow these steps when assisting the user:

1. Memory and Learning System

**Before starting any task:**

Read the `.cursorrules` file to review previous lessons and current scratchpad state

Check the "Lessons" section for project-specific patterns and past mistakes to avoid

Review the "Scratchpad" section for any in-progress tasks

**During task execution:**

When you make a mistake or receive a correction, add an entry to the "Lessons" section in `.cursorrules`

When you discover reusable information (library versions, model names, configuration patterns), record it in "Lessons"

All `.cursorrules` content must be in English regardless of chat language

2. Task Planning with Scratchpad

**For new tasks:**

1. Review current scratchpad content

2. Clear old/unrelated tasks if necessary

3. Write a clear task explanation in the scratchpad

4. Break down the task into steps using todo markers:

```

[X] Completed step

[ ] Pending step

```

**During execution:**

Update todo markers as you complete each subtask

At milestones, reflect in the scratchpad and plan next steps

Keep the big picture visible by maintaining scratchpad organization

Always consult the scratchpad before planning the next step

3. Code and Documentation Standards

Use English for ALL generated content: code, comments, docstrings, scratchpad entries

Read files before editing them

Include debugging information in program outputs

For git operations requiring multiline messages:

- Write message to a temporary file

- Use `git commit -F <filename>`

- Include "[Cursor]" prefix in commit messages and PR titles

- Delete the temporary file afterward

Always ask for user confirmation before committing unless explicitly requested

Use `git push --set-upstream origin main` for first push to new repositories

4. Available Tools

All tools are Python 3 based. Use `venv/bin/python3` to run them.

**Web Scraping:**

```bash

venv/bin/python3 ./tools/web_scraper.py --max-concurrent 3 URL1 URL2 URL3

```

**Search Engine:**

```bash

venv/bin/python3 ./tools/search_engine.py "search keywords"

```

Output format: URL, Title, Snippet

**LLM Query (multiple providers):**

```bash

venv/bin/python3 ./tools/llm_api.py --prompt "Your question" --provider {openai|anthropic|azure|deepseek|gemini|local}

```

Providers: OpenAI (gpt-4o), Anthropic (claude-3-sonnet), Azure, DeepSeek, Gemini, Local (Qwen)

**Screenshot Capture:**

```bash

venv/bin/python3 tools/screenshot_utils.py URL [--output OUTPUT] [--width WIDTH] [--height HEIGHT]

```

**Screenshot Verification with LLM:**

```bash

venv/bin/python3 tools/llm_api.py --prompt "Verification question" --provider openai --image path/to/screenshot.png

```

**Video to GIF Conversion:**

```bash

Single video

venv/bin/python3 playgrounds/video2gif/video_to_gif.py input.mp4 [output.gif]

Batch conversion

venv/bin/python3 playgrounds/video2gif/batch_convert_videos.py input_dir

GIF resizing

venv/bin/python3 playgrounds/video2gif/gif_resize.py input.gif [output.gif] [--width WIDTH] [--height HEIGHT]

```

**Video Transcription:**

```bash

venv/bin/python3 playgrounds/video_transcriber/video_transcriber.py VIDEO_URL [--output output.json] [--raw-only]

```

5. Known Lessons (Apply These)

Use the python3 venv in `./venv`

Handle UTF-8 encoding properly for international content

Use 'seaborn-v0_8' instead of 'seaborn' for matplotlib styles

Use 'gpt-4o' for OpenAI GPT-4 with vision

Use Pillow < 10.0.0 (e.g., 9.5.0) for moviepy to avoid ANTIALIAS issues

Keep stdout clean, use stderr for debug info when building pipeline tools

Project Structure Reference

```

tools/

llm_api.py - Multi-provider LLM integration

web_scraper.py - Web content scraping

search_engine.py - Web search

screenshot_utils.py - Screenshot capture

playgrounds/

video2gif/ - Video to GIF tools

video_transcriber/ - Video transcription

```

Example Workflow

```python

1. Update scratchpad with task

2. Use tools as needed

from screenshot_utils import take_screenshot_sync

from llm_api import query_llm

screenshot = take_screenshot_sync('https://example.com', 'screenshot.png')

result = query_llm(

"Describe this webpage's layout",

provider="anthropic",

image_path=screenshot

)

3. Update scratchpad with progress

4. Record any lessons learned in .cursorrules

```

Important Notes

The learning system depends on maintaining the `.cursorrules` file

Batch processing can be done by consulting Python files and writing custom scripts

Screenshot verification and LLM features require API key configuration

Always maintain English in generated content for consistency

The scratchpad is your memory - keep it organized and up-to-date

Cursor Rules Learning Agent

Cursor Rules Learning Agent

Core Capabilities

Instructions

1. Memory and Learning System

2. Task Planning with Scratchpad

3. Code and Documentation Standards

4. Available Tools

Single video

Batch conversion

GIF resizing

5. Known Lessons (Apply These)

Project Structure Reference

Example Workflow

1. Update scratchpad with task

2. Use tools as needed

3. Update scratchpad with progress

4. Record any lessons learned in .cursorrules

Important Notes

Reviews (0)