Installs into .claude/skills of the current project.
Are you the author of 2225 Devkitlraadd Feature Dc63f95c?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/tools-only-2225-devkitlraadd-feature-dc63f95c)
---
description: Add a new feature to the feature list
allowed-tools: Read, Write, Edit
argument-hint: "[category] [priority] [description]"
---
# Long-Running Agent - Add Feature
Add a new feature to the project's feature list.
## Arguments
- **$1** (category): core | ui | api | database | auth | testing | other
- **$2** (priority): critical | high | medium | low
- **$3+** (description): Description of the feature
## Example Usage
```
/developer-kit:devkit.lra.add-feature api high Add endpoint to retrieve user preferences
/developer-kit:devkit.lra.add-feature ui medium Implement dark mode toggle in settings
```
## Your Task
1. Read the current `.lra/feature-list.json`
2. Determine the next feature ID:
- Find the highest existing ID (e.g., F042)
- Increment by 1 (e.g., F043)
3. Create the new feature object:
```json
{
"id": "F[next_number]",
"category": "$1",
"priority": "$2",
"description": "$3 (and remaining arguments)",
"acceptance_criteria": [
"Generate 2-4 reasonable acceptance criteria based on the description"
],
"status": "pending",
"completed_at": null,
"notes": ""
}
```
4. Add the feature to the `features` array in the appropriate position:
- Critical priority: near the top (after other critical)
- High priority: after critical features
- Medium/Low: at the end
5. Save the updated feature-list.json
## Output
Confirm the feature was added:
```
✅ Feature Added
ID: F043
Category: api
Priority: high
Description: Add endpoint to retrieve user preferences
Acceptance Criteria:
- GET /api/users/{id}/preferences returns user preferences
- Response includes all preference categories
- Unauthorized requests return 401
Total features: 43 (38 pending, 5 completed)
```
## Execution Instructions
**Agent Selection**: To execute this LRA task, use the following approach:
- Primary: Use `general-purpose` agent with task management and state persistence capabilities
- Or use `plan` agent for complex multi-step workflows