capture-request

# Capture Request

Captures user intent as structured work requests. **Does not execute work.**

This skill implements intent separation: capture what needs to be done, then hand off to the work-loop for execution.

## Purpose

- Accept raw user input describing work
- Detect if input contains multiple disjoint ideas
- Split into separate requests when appropriate
- Group only when dependency exists
- Write structured request files to queue

## Constraints

**CRITICAL: This skill NEVER:**
- Executes code
- Makes repository changes (beyond writing request files)
- Performs speculative planning
- Implements features

**This skill ONLY:**
- Captures intent
- Structures requests
- Writes to `do-work/requests/`

## Request Contract

Every request must follow this schema:

```markdown
# Request: <short descriptive title>

## Intent
What outcome is desired. Plain language.

## Context
Relevant files, constraints, assumptions.
Explicit exclusions.

## Tasks
- Step 1
- Step 2
- Step 3

## Done When
Objective completion criteria.
```

## Workflow

### 1. Parse User Input

When user provides work description:

1. Identify the core intent(s)
2. Check for multiple disjoint ideas
3. Determine dependencies between ideas

### 2. Splitting Rules

**Split into SEPARATE requests when:**
- UI change AND database change (different domains)
- Unrelated features mentioned together
- No execution dependency between items

**Keep as SINGLE request when:**
- Sequential dependency exists (B requires A)
- Tightly coupled changes (model + its tests)
- User explicitly groups them

### 3. Handle Ambiguity

If input is ambiguous:
- Ask ONE clarification question
- Then proceed with best interpretation
- Do not over-clarify

### 4. Write Request Files

For each request:

1. Generate filename: `YYYYMMDD-HHMMSS-slug.md`
2. Write to `do-work/requests/`
3. Confirm to user what was captured

## Directory Setup

If directories don't exist, create them:

```
do-work/
├── requests/    # Pending work
├── in