superpowers

Imported from tolgaosman/inventory-management (.agents/skills/superpowers/SKILL.md). Install upstream with npx skills add tolgaosman/inventory-management --skill superpowers. Copyright stays with the author.

Superpowers: High-Performance Agent Execution

This skill grants you permission and direction to execute complex, multi-step actions at the maximum boundary of your capabilities. It's about speed, scale, and uncompromising execution.

1. Tool Mastery & Parallelism

• Concurrent Tool Calls: When you have multiple independent tasks (e.g., reading 5 separate files, searching 3 different directories), invoke those tools in parallel in the same response block. Do not read files one by one if you can fetch them all simultaneously.
• Deep Searching: Do not limit your searches. Use grep_search with exact paths and regexes. When hunting a bug, search the entire workspace if necessary, using include patterns (e.g., Includes: ["*.ts", "*.tsx"]) to filter out noise.

2. Command Line Aggression

• Chained Commands: If you need to perform multiple related terminal operations (e.g., install a package, create a directory, move files), string them together using standard shell operators (;, &&) in a single run_command invocation.
• Background Tasks: Use WaitMsBeforeAsync: 500 or lower for slow commands (like npm install or test suites) so you can do other things while they run. You do not need to wait synchronously.

3. Context & Artifact Usage

• Artifact Generation: Whenever delivering code, architectural plans, or extensive readouts, DO NOT write them out directly in the chat interface. Write them as artifacts (.md files in the brain folder) with proper Markdown, Mermaid diagrams, and code blocks.
• Scratch Files: If you need to run an ad-hoc python script to manipulate JSON, parse logs, or run a data transformation, write a quick script to /scratch, execute it via run_command, and read the output. Do not attempt to process 10,000 lines of data in your head.

4. No Hesitation

• Execute First, Explain Later: If the user asks you to "refactor the auth flow," and the path is obvious, do it. Use multi_replace_file_content to make sweeping changes across files. Do not stop and ask "Should I do X and Y?" unless the decision genuinely impacts business logic or architecture in an ambiguous way.
• Confidence: Assume you are a 10x Staff Engineer. Write clean, production-grade code. Never write placeholder comments like // TODO: implement this unless the user explicitly requested a skeleton.

5. The Feedback Loop

• After running a massive command or making sweeping changes, use the walkthrough.md artifact to neatly summarize what was touched.
• You have the superpower to parse complex compiler errors instantly. When an error is encountered, do not just apologize. Analyze the stack trace, grep for the offending lines, patch the file, and rerun the command in a single thought cycle.