SuperClaude Framework

Organize engineering work around defined outputs: clarify, plan, implement, diagnose and verify.

Hosts listed by the author

Claude Code

GitHub stars: 23,914
Version: 4.3.0
Checked: 2026-10-30
Categories: Engineering Product & Project

Validate one workflow before adding more tools.

Start with requirements → plan, check that the repository was understood, then implement one bounded stage. Connect search, documentation or browser tools when the task needs them, and verify each connection.

Pick a task and get started

Run the installation commands in a terminal. Run the /sc: tasks in the Claude Code conversation after preparing the named input files.

What would you like to make?

Turn a feature brief into an implementation planImplement one stage and verify its behaviorDiagnose a failing build before applying a fixBuild a traceable case for a technical choice

Start here. In the project root, create brief.md with users, current and desired behavior, scope and acceptance criteria. A one-line request to “add search” leaves important decisions unresolved.

Install & configure
Use an existing pipx installation with Python 3.10+. This installs a pinned Python package and copies commands and agents into personal Claude Code directories. Existing names are skipped by default; confirm loading in the client afterward.
```
pipx install superclaude==4.3.0
superclaude install
superclaude install --list
```
Read the author’s setup guide
Check that Claude Code recognizes the commands
superclaude install writes commands to ~/.claude/commands/sc and agent files to ~/.claude/agents; existing same-named files are skipped by default. Restart Claude Code and check the / menu for sc:workflow, sc:implement, sc:troubleshoot and sc:research. A file listed in the terminal is not proof that the client loaded it, and doctor does not check command loading in Claude Code or MCP connectivity.
Inspect the installer
Give your AI this first task
This editorial example follows the documented capabilities. Attach your materials and send it to your configured AI assistant.
/sc:workflow brief.md --strategy systematic --depth deep Read brief.md, project guidance, repository structure, relevant routes and data-access code. Plan search for the existing order list using paths and conventions found in this repository. Resolve the search fields, empty queries, no results, access boundaries, case and fuzzy matching, combined filters, and pagination when the query changes. Record missing product decisions instead of inventing requirements. Write claudedocs/workflow_order-search.md with phases, affected files, reusable behavior, dependencies, risks and concrete verification steps. Separate independent tasks from work that requires earlier results. Produce only the plan: do not modify application code, run builds or tests, or start implementation. Use tools that are actually available; when an MCP server is missing, use existing file-reading tools for feasible work and disclose the limitation. Map every acceptance criterion to a planned check and recommend the smallest useful first implementation stage.
Check what you receive
- A plan grounded in real code paths, dependencies and execution order.
- Acceptance coverage and unresolved product decisions.
- A bounded first stage, without claiming code or tests are complete.
Read the author’s notes for this task

Before you start

Repository and task material: Bring readable code, project conventions and a specific feature or failure. Use brief.md for requirements, issue.md for commands and logs, and research-brief.md to bound a research question. These filenames are this guide’s working conventions, not files generated by the installer.
Claude Code and Python: Have Claude Code installed, authenticated and working, plus Python 3.10+ and pipx. The commands below use the 4.3.0 Python package. The older installation guide’s Python 3.8 requirement and some capitalized commands and flags do not match the current entry point.
Tools and costs: The framework is MIT licensed and is not affiliated with or endorsed by Anthropic. Claude Code access, model usage and external MCP services retain their own costs and limits. Existing tools can handle basic code reading and planning; web research and browser checks require working tools for those activities.

Who it’s for

Developers · Professionals · Founders

Know the boundaries

/sc:workflow ends with a plan and /sc:research ends with a report. Their completion is not feature acceptance.
/sc:troubleshoot defaults to diagnosis without editing project files or applying repairs; reproduction commands may produce logs or build artifacts. Its --fix flow still asks for confirmation of the concrete repair. /sc:test runs existing tests; it should not be described as automatically creating all missing coverage.
Roles, modes and flags such as --depth are instructions to the model. They do not guarantee duration, source counts, speedups or token savings.
This guide checks author documentation and installation code, without installing or running the framework for you. Verify command loading, connections and output in your own environment.

Install and run in your chosen tools, then check the output against your own materials.

More about this skill

SuperClaude supplies reusable working instructions for Claude Code. It separates discovery, code analysis, planning, implementation, testing and research into distinct entry points, making the expected output and next step explicit. It is useful when you have a repository and a concrete goal and want to stop restating the same working process.

The reviewed source contains 30 command files and 20 agent definitions. The default installer copies them into personal Claude Code directories. They are Markdown instructions interpreted by the model with its available tools, rather than a deterministic pipeline. Twenty role definitions do not mean twenty independent agents start on every request.

Begin with a small feature plan. Check whether it points to real code, whether implementation follows acceptance criteria, and whether research leads back to primary sources. Those outcomes are more informative than command counts or self-reported confidence scores.

Sources and verification

Sources checked 2026-10-30

Usage notes are based on author documentation. We have not tested every skill’s generated output.

Author documentation

Explore related skills

Continue exploring by task.

Karpathy SkillsA community guide from Karpathy's observations: state assumptions, keep it simple, verify the result.gstackGarry Tan's specialist skills link product exploration, plan and code review, browser tests and PRs.Deep ResearchScope a question, keep sources and evidence, and write a cited report with counterevidence and limits.Matt Pocock SkillsComposable skills for specs, tickets, test-first development, debugging and architecture reviews.

Validate one workflow before adding more tools.

Pick a task and get started

Install & configure

Check that Claude Code recognizes the commands

Give your AI this first task

Check what you receive

Sources and verification

Explore related skills