devis

Interviewing to understand requirements, and then implementing them using a Manus-style approach.

devis is a Claude Code plugin that understands requirements through structured interview workflows, then implements features based on Manus's context engineering principles. It uses persistent Markdown files as "working memory on disk" to ensure goals and progress are never lost during complex tasks.

Key Features

• Structured Interviewing (/devis:intv) - Clarify requirements, technical solutions, and trade-offs through in-depth interviews
• Progressive Implementation (/devis:impl) - Manus-style workflow using the filesystem as external memory
• Context Engineering - Based on Manus AI agent best practices, generates three core files: task_plan.md, findings.md, progress.md
• Context Engineering - Built on production-grade AI agent best practices from pre-acquisition Manus

Installation

Method 1: Claude Code Marketplace (Recommended)

Run in Claude Code:

/plugin marketplace add st01cs/devis
/plugin install devis@devis

Method 2: Manual Installation

# Clone repository to Claude Code plugins directory
git clone https://github.com/st01cs/devis.git ~/.claude/plugins/devis

Usage

Workflow

devis adopts a two-stage workflow:

User Requirements → /devis:intv (Interview) → Planning Docs → /devis:impl (Implementation) → Final Delivery

Step 1: Interview & Plan

Use the /devis:intv command for requirements interview:

/devis:intv path/to/your/plan.md

Interview Process:

• Claude will ask in-depth questions about technical implementation, UI/UX, concerns, trade-offs, etc.
• After the interview, three files are automatically generated:
  • task_plan.md - Phase breakdown, progress tracking, decision records
  • findings.md - Research findings, interview content, technical decisions
  • progress.md - Session logs, test results

Step 2: Implement

Use the /devis:impl command to implement according to plan:

/devis:impl path/to/your/plan.md

Implementation Process:

• Executes plan phase by phase, updating status after each completion
• Records all errors and solutions
• Uses "2-Action Rule" to prevent loss of multimodal information
• Verifies all phases complete upon finish

File Structure

devis/
├── .claude-plugin/
│   ├── plugin.json          # Plugin metadata
│   └── marketplace.json      # Marketplace configuration
├── commands/
│   ├── intv.md              # /intv command definition
│   └── impl.md              # /impl command definition
├── templates/
│   ├── task_plan.md         # Task plan template
│   ├── findings.md          # Findings template
│   └── progress.md          # Progress log template
├── refs/
│   ├── manus.md             # Manus principles reference
│   └── examples.md          # Practical examples
├── scripts/
│   └── check-complete.sh    # Completion check script
└── README.md

Example Scenario

# 0. Requirements draft
# Create a file dev-docs/plan/feature-xxx/feature-draft.md with a simple description of requirements

# 1. Interview requirements
/devis:intv dev-docs/plan/feature-xxx/feature-draft.md

# Interview will ask: design preferences, state management, compatibility requirements, etc.
# After interview, three files are automatically generated:
# - dev-docs/plan/feature-xxx/task_plan.md
# - dev-docs/plan/feature-xxx/findings.md
# - dev-docs/plan/feature-xxx/progress.md

# 2. Implement feature
/devis:impl dev-docs/plan/feature-xxx/task_plan.md

# Implements by phases, tracks progress

Advanced Usage

Custom Templates

You can modify files under templates/ to customize the planning process:

# Edit template
nano ~/.claude/plugins/devis/templates/task_plan.md

FAQ

Q: How to resume interrupted work?

A: Simply run /devis:impl path/to/task_plan.md again, it will read existing files and continue from the current phase.

Acknowledgments

The core concepts and methodology of this project are deeply inspired by:

• Manus - Context Engineering for AI Agents

• planning-with-files - planning-with-files

License

MIT License - See LICENSE for details

Author

st01cs - GitHub