scratchpad_skill

Overview

This skill provides an isolated scratchpad environment for quick, result-focused experiments and hypothesis validation. It creates a temporary workspace, runs short tests (code, shell, API), extracts concise conclusions, and then cleans up. Use it to keep the main context clean and return only key outcomes.

How this skill works

On demand the skill creates a temporary directory and writes test files or scripts. It executes the tests, reads outputs, and extracts compact results (success/failure, key data, correct usage). Finally it optionally deletes temp files and returns a conclusion-only summary without revealing internal steps or tooling calls.

When to use it

• Validating a single hypothesis quickly without polluting the main context
• Trying API usage or library behavior in isolation
• Debugging an isolated reproducible issue
• Comparing small performance differences between implementations
• Testing shell commands, scripts, or language-specific snippets

Best practices

• Test one hypothesis per scratchpad session
• Exit immediately when a decisive conclusion is reached
• Return only concise results: conclusion, key data, and recommended code or commands
• Manually clean temporary files when practical; rely on TTL only as fallback
• Name files predictably (e.g., test_<topic>.<ext>, data_<topic>.<ext>)

Example use cases

• Check whether Python re.split supports multiple separators and show correct usage
• Experiment with requests.get(stream=True) usage and provide a short example for large downloads
• Reproduce and fix a small ImportError or API mismatch, then return the correct API call and compatibility note
• Compare two small implementations and return a performance table with recommended option
• Validate a shell command flag (e.g., find -mtime) and return succinct usage examples