Claude Commands
https://github.com/badlogic/claude-commands- Category
- AI Agents
- Rank
- No. 1498Tools index
Previous survey · No. 1504 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- badlogic
- GitHub
- 522 stars
- Date
About
Global Claude Code commands and workflows curated by Mario Zechner (badlogic).
What it does
Claude Commands turns a rough todo into a staged, human-approved implementation process. You choose either isolated worktrees with incremental commits and pull requests, or a simpler current-branch flow ending in one commit. Both guide the agent through setup, task selection, refinement, implementation, validation, and completion.
Why it's ranked here
This is a strong fit for engineers who want agents doing scoped execution without surrendering control. Explicit approval points, saved research, resumable task state, and validation steps address common failure modes. Its value comes from understandable process design, not hidden automation, though that also leaves reliability dependent on prompt-following.
What's good
The workflow converts vague ideas into descriptions, implementation checklists, tests, and durable research notes before coding starts. Each change requires human approval. Unexpected work must be added to the agreed plan first. The worktree option adds branch isolation, incremental commits, editor handoff, and pull-request creation without introducing software dependencies.
Tradeoffs
It is prompt-based, so complex tasks can cause the model to ignore parts of the process. The branch workflow keeps uncommitted changes during implementation. Worktree resumption detection only works when agents share a machine. Concurrent edits to the todo list can create duplicate work, and neither workflow solves merge conflicts or distributed coordination.
How to use it well
Use it for small, focused engineering tasks where a developer wants to approve the plan and each implementation step. Choose worktrees for isolated or parallel features, and the current-branch variant for quick fixes. Start from a clean tree. It does not replace project management, merge-conflict handling, or supervision of broad autonomous builds.
Technical notes+
README.md describes two approximately 150-line Markdown state machines. todo-worktree.md uses git worktrees, per-step commits, task.md and analysis.md persistence, editor handoff, validation, GitHub CLI pull-request creation, and orphan detection through stored agent PIDs. todo-branch.md runs the same INIT → SELECT → REFINE → IMPLEMENT → COMMIT phases on the current branch, progressively stages changes, and creates one final commit. Installation copies a chosen workflow into ~/.claude/commands or project/.claude/commands; other agents can receive its contents as an initial instruction. publish.md separately documents an npm build, version bump, tag, push, and public publish routine, although the supplied text includes no package manifest.
Observed
- Primary format
- Markdown workflows containing Git and shell instructions
- Dependencies
- The README states zero dependencies
- Claude Code interface
- Installed as a global or project command and invoked through the Claude CLI
- Other-agent interface
- Workflow contents can be supplied as an initial instruction to another coding agent
- Workflow variants
- Worktree isolation with incremental commits, or current-branch execution with one completion commit
- Required tooling
- Git and shell commands; the worktree flow also uses an editor command and GitHub CLI for pull requests
Read from README.md, publish.md, todo-branch.md, todo-worktree.md.
What it can do
Execute Claude code commands globally
Code snippets or commands → Executed results or processed code
Run curated coding workflows
Workflow parameters and source code → Transformed or processed code following predefined patterns
Apply badlogic's code transformations
Source code files → Modified code following established conventions
Access global Claude command library
Command name or identifier → Available commands and their documentation
Process code using AI-powered commands
Code and command instructions → AI-generated code modifications or suggestions
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.