RawDrive Doc Co-Authoring Workflow

Structured workflow for collaborative documentation in the RawDrive project. Three stages: Context Gathering, Refinement & Structure, and Reader Testing.

RawDrive Documentation Structure

When to Offer This Workflow

Trigger conditions:

• •"write a spec", "create a proposal", "draft a plan"
• •Feature planning: "let's design the X feature"
• •Document types: "PRD", "ADR", "RFC", "design doc"
• •Any substantial writing in specs/ or docs/

Initial offer:

I can guide you through a structured workflow:
1. Context Gathering - You provide context, I ask clarifying questions
2. Refinement & Structure - Build sections iteratively
3. Reader Testing - Test with fresh Claude to catch blind spots

Want to try this approach, or prefer freeform?

Stage 1: Context Gathering

Goal: Close the gap between what you know and what Claude knows.

Initial Questions

1. •What type of document? (feature spec, ADR, technical spec, plan)
2. •Who's the audience? (developers, product, stakeholders)
3. •What impact should it have when read?
4. •Any existing templates or format requirements?
5. •Related features or dependencies?

RawDrive-Specific Context

For feature specs, gather:

• •Workspace isolation - How does workspace_id affect this?
• •Storage implications - R2/BYOS considerations?
• •AI/MCP integration - Any AI features or MCP tools?
• •Multi-tenancy - Billing, limits, permissions?
• •Frontend components - Which UI components needed?
• •Backend services - New repositories/services required?

Info Dumping

Encourage dumping all context:

• •Background on the feature/problem
• •Related Slack discussions or docs
• •Why alternatives aren't used
• •Timeline pressures
• •Technical dependencies
• •Stakeholder concerns

Exit condition: When edge cases and trade-offs can be discussed without needing basics explained.

Stage 2: Refinement & Structure

Goal: Build document section by section through brainstorming and iteration.

RawDrive Document Templates

Feature Spec (specs/{feature}/spec.md)

# {Feature Name}

## Overview
Brief description of the feature.

## User Stories
- US1: As a [role], I want [action] so that [benefit]

## Technical Requirements
### Backend
- New endpoints, services, repositories

### Frontend
- New components, hooks, pages

### Database
- New tables, migrations, indexes

## Security Considerations
- Workspace isolation
- Permission requirements
- Data encryption needs

## Performance Considerations
- Caching strategy
- Query optimization
- Rate limiting

ADR (docs/ADR/{number}-{title}.md)

# ADR-{number}: {Title}

## Status
Proposed | Accepted | Deprecated | Superseded

## Context
What is the issue we're addressing?

## Decision
What is the change we're making?

## Consequences
What are the trade-offs?

## Alternatives Considered
What other options were evaluated?

Technical Spec (docs/TechnicalSpecs/{name}.json)

{
  "name": "feature_name",
  "version": "1.0.0",
  "description": "Brief description",
  "endpoints": [],
  "models": [],
  "workflows": []
}

Section-by-Section Process

For each section:

1. •Clarify - Ask 5-10 questions about what to include
2. •Brainstorm - Generate 5-20 options
3. •Curate - User selects: "Keep 1,4,7" / "Remove 3" / "Combine 5+6"
4. •Draft - Write the section
5. •Refine - Iterate on feedback using surgical edits

Key instruction: Ask user to indicate changes rather than edit directly. This helps learn their style.

Quality Checks

After 3 iterations with no substantial changes:

• •Ask if anything can be removed without losing value
• •Check for RawDrive-specific requirements (workspace_id, permissions)
• •Verify code examples use correct tech stack (Python/FastAPI, React/TS)

Stage 3: Reader Testing

Goal: Test document with fresh Claude to catch blind spots.

With Sub-Agents (Claude Code)

// Test document with fresh context
const testQuestions = [
  "What is this feature trying to solve?",
  "What are the security implications?",
  "What migrations are needed?",
];

// Invoke sub-agent with just the document
for (const question of testQuestions) {
  const response = await subAgent.ask(docContent, question);
  // Check if answers are correct
}

Manual Testing (Claude.ai)

1. •Open fresh Claude conversation
2. •Paste the document
3. •Ask generated questions
4. •Check if Reader Claude gets correct answers
5. •Ask: "What's ambiguous?", "What context is assumed?"

Exit Condition

Document is ready when Reader Claude:

• •Answers questions correctly
• •Doesn't surface new ambiguities
• •Understands the RawDrive context from the doc alone

RawDrive-Specific Checklist

Before finalizing any RawDrive document:

• • Workspace isolation addressed (workspace_id in all queries)
• • Permissions defined (RBAC requirements)
• • Code examples use correct stack:
  • •Backend: Python 3.11 + FastAPI + SQLAlchemy
  • •Frontend: React 19 + TypeScript + TailwindCSS
• • File paths correct (not TurboRepo apps/ structure)
• • Migrations mentioned if database changes
• • Security considerations documented
• • Performance implications noted

Tips for Effective Guidance

Tone: Direct and procedural. Don't oversell the approach.

Handling Deviations:

• •If user wants to skip: "Want to skip this and write freeform?"
• •If frustrated: "This is taking longer than expected. Ways to speed up?"

Context Management:

• •Address gaps as they come up
• •For RawDrive, always check workspace isolation early

File Management:

• •Create files in appropriate specs/ or docs/ directory
• •Use str_replace for edits, never reprint whole doc
• •Follow RawDrive naming conventions