Skip to content
Back to skills

Feat

ASecurity

Add New Feature

  • 6,081 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added August 31, 2026
ai-agentsfrontend

Security analysis

A100/100

Scanned August 31, 2026

npx -y skills add UfoMiao/zcf --skill feat --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Feat?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Feat
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ufomiao-feat/badge)](https://www.skillsdirectory.com/skills/ufomiao-feat)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: feat
description: Add New Feature
disable-model-invocation: true
---

$ARGUMENTS

## Core Workflow

### 1. Input Analysis and Type Determination

When receiving user input, first perform type determination and clearly inform the user:

**Determination Criteria:**

- **Requirement Planning Type**: User proposes new feature requirements, project ideas, or needs to formulate plans

- **Discussion Iteration Type**: User requests to continue discussion, modify, or refine existing planning

- **Execution Implementation Type**: User confirms planning is complete and requests to start specific implementation work

### 2. Classification Processing Mechanism

#### A. Requirement Planning Processing

**Trigger Condition**: Identified as functional requirement input

**Execution Actions**:

- Enable Planner Agent

- Generate detailed markdown planning document

- Store document in `./.claude/plan` directory, named in plan/xxx.md format

- Include: objective definition, feature breakdown, implementation steps, acceptance criteria

#### B. Discussion Iteration Processing

**Trigger Condition**: User requests to continue discussion or modify planning

**Execution Actions**:

- Retrieve and analyze previously generated planning files

- Identify user feedback and confirmation content

- Enable Planner Agent

- Generate detailed markdown planning document

- Create a new document, for example, if the last one was plan/xxx.md, then this time it's plan/xxx-1.md, if the last one was plan/xxx-1.md, then this time it's plan/xxx-2.md, and so on

- Reorganize pending implementation task priorities

#### C. Execution Implementation Processing

**Trigger Condition**: User confirms planning is complete and requests to start execution

**Execution Actions**:

- Start task execution in the order of planning documents

- Perform task type identification before each subtask begins

- **Frontend Task Special Processing**:

- Check if available UI design exists

- If no design solution exists, must use UI-UX-Designer Agent

- Complete UI design before proceeding with development implementation

### 3. Key Execution Principles

#### Mandatory Response Requirements

- **Must first state in every interaction**: "I determine this operation type as: [specific type]"

- Type determination must be accurate and clearly communicated to users

#### Task Execution Standards

- Strictly execute according to documented planning

- Must clarify task nature and dependencies before subtask startup

- Frontend tasks must ensure UI design completeness

#### State Management Mechanism

- Maintain task completion status tracking

- Timely update planning document status

- Ensure user visibility of progress

## Quality Assurance Points

1. **Type Determination Accuracy**: Type identification at the beginning of each interaction must be accurate

2. **Document Consistency**: Planning documents and actual execution remain synchronized

3. **Dependency Management**: Pay special attention to UI design dependencies of frontend tasks

4. **Transparent User Communication**: All judgments and actions must be clearly communicated to users

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.