A collection of Claude Code skills for enhanced AI-assisted development workflows.
The main skill implements the Manus AI workflow pattern - a file-based "working memory" system that enables Claude to handle complex, multi-step tasks without losing context.
| File | Purpose |
|---|---|
task_plan.md |
Track phases and progress with checkboxes |
notes.md |
Store research findings (not in context) |
deliverable.md |
Final synthesized output |
"Markdown is my working memory on disk. Since I process information iteratively and my active context has limits, Markdown files serve as scratch pads for notes, checkpoints for progress, and building blocks for final deliverables."
This pattern allows Claude to:
- Maintain focus across 50+ tool calls
- Never repeat failed approaches
- Resume interrupted sessions
- Produce consistent, thorough results
# Clone this repo
git clone https://github.com/UMWai/claude-code-workflows.git
# Copy skills to your Claude config
cp -r claude-code-workflows/skills/* ~/.claude/skills/git clone https://github.com/UMWai/claude-code-workflows.git
ln -s $(pwd)/claude-code-workflows/skills/manus-workflow ~/.claude/skills/manus-workflowmkdir -p ~/.claude/skills
curl -L https://github.com/UMWai/claude-code-workflows/archive/main.tar.gz | \
tar -xz --strip-components=2 -C ~/.claude/skills claude-code-workflows-main/skills/manus start [your goal here]
Example:
/manus start Research and implement rate limiting for our API
| Command | Description |
|---|---|
/manus start [goal] |
Initialize new workflow |
/manus status |
Check current progress |
/manus checkpoint |
Force save to notes |
/manus complete |
Finalize deliverable |
Use this workflow for:
- Multi-step research tasks
- Complex implementations
- Bug investigations
- Anything requiring >5 tool calls
Skip for:
- Simple questions
- Single file edits
- Quick lookups
An iterative task completion workflow based on the Ralph Wiggum technique. Uses PRD checkboxes to track progress across iterations.
| File | Purpose |
|---|---|
PRD.md |
Requirements with checkboxes to track completion |
progress.md |
Log of actions (append-only memory) |
PROMPT.md |
Agent instructions for autonomous work |
/um-loop start Build a REST API with CRUD operations
Then continue with /um-loop to work on the next item.
| Command | Description |
|---|---|
/um-loop start [task] |
Initialize new loop |
/um-loop |
Continue (do next item) |
/um-loop status |
Check progress |
| Scenario | Use |
|---|---|
| Research, exploration, context-heavy | Manus Workflow |
| Implementation, clear deliverables | UM Loop |
| Complex project | Manus first, then UM Loop |
| Skill | Description | Status |
|---|---|---|
manus-workflow |
File-based planning workflow | ✅ Ready |
um-loop |
Iterative task completion with PRD checkboxes | ✅ Ready |
claude-code-workflows/
├── README.md
├── LICENSE
├── skills/
│ ├── manus-workflow/
│ │ ├── SKILL.md # Main skill definition
│ │ ├── reference.md # Deep dive on principles
│ │ ├── examples.md # Usage examples
│ │ └── templates/
│ │ ├── task_plan.md
│ │ ├── notes.md
│ │ └── deliverable.md
│ └── um-loop/
│ ├── SKILL.md # Main skill definition
│ └── templates/
│ ├── PROMPT.md # Agent instructions
│ ├── PRD.md # Task requirements
│ └── progress.md # Progress log
└── docs/
└── installation.md
PRs welcome! To add a new skill:
- Create
skills/your-skill/SKILL.md - Add any templates or supporting files
- Update README with skill description
- Submit PR
- Workflow pattern inspired by Manus AI
- Technical analysis from @renschni
- planning-with-files reference implementation
MIT