deep-work-session

Files (1)

• SKILL.md

SKILL.md (first lines)

# Deep Work Session

## Overview

Set up a focused working session with preferences established upfront, enabling autonomous execution with minimal interruption.

**Core principle:** Front-load decisions so the agent can work autonomously.

**When to use:**
- User explicitly mentions "deep work" → definite trigger
- User signals substantial work ("let's build X", "let's implement X") → ask first

**When NOT to use:**
- Simple questions or quick fixes
- User dives directly into a specific task without signaling session intent
- Information requests

## Trigger Behavior

```dot
digraph trigger {
    "User message" [shape=doublecircle];
    "Mentions 'deep work'?" [shape=diamond];
    "Start deep work session" [shape=box];
    "Signals substantial work?" [shape=diamond];
    "Ask: Deep Work or interactive?" [shape=box];
    "Proceed normally" [shape=box];
    
    "User message" -> "Mentions 'deep work'?";
    "Mentions 'deep work'?" -> "Start deep work session" [label="yes"];
    "Mentions 'deep work'?" -> "Signals substantial work?" [label="no"];
    "Signals substantial work?" -> "Ask: Deep Work or interactive?" [label="yes"];
    "Signals substantial work?" -> "Proceed normally" [label="no"];
}
```

**Substantial work signals:**
- "Let's build/implement/create X"
- "I want to work on X"
- Multi-step requests
- Feature development, refactoring, complex fixes

**Use the question tool to ask:**
> "Would you like to start a Deep Work session (autonomous execution with upfront planning) or work interactively?"
> Options: Deep Work session, Interactive style

## Session Setup

**Announce:** "Starting Deep Work session. Let me gather a few preferences upfront."

### Step 1: Analyze Task and Present Defaults

Analyze what the user wants to do and present a proposed configuration:

```
Deep Work session for: [task description]

Proposed configuration:
- Commits: Auto-commit after each task
- Questions: Front-loaded (ask everything now, then work autonomously)
- Scope: [inferred - e.g., "src/components/", "no breaking changes"]
- Skills: [relevant skills - e.g., "TDD, verification-before-completion"]

Which would you like to change?
```

**Use the question tool with multiple selection:**
- Commit preference
- Question style  
- Scope boundaries
- Skills to use
- None - looks good

### Step 2: Gather Changes (if any)

Only ask follow-up questions for items the user selected to change.

**Commit options:**
- Auto-commit after each task (default)
- Ask before each commit
- Batch commits at end
- No commits (user handles)

**Question style options:**
- Front-loaded then minimal (default) - ask foreseeable questions now, then only interrupt if truly blocked
- Minimal - proceed unless blocked, no upfront gathering
- Checkpoint - pause at major decision points
- Collaborative - frequent check-ins

**Scope options** (ask whichever are relevant):
- Files/directories to focus on or avoid
- Change magnitude (small fix, refactor, rewrite)
- Testing requirements
- Time/effort budget

**Skills options:**
- Present skills that seem relevant to the task
- Let user confirm or decline each

### Step 3: Plan the Workflow

Based on the task, propose a sequence of skills:

```
Proposed workflow:
1. brainstorming - explore the design
2. writing-plans - create implementation plan
3. subagent-driven-development - execute with fresh agents per task

Does this workflow look right?
```

**Use the question tool:**
- Yes, proceed
- Adjust workflow (let me specify)

**Common workflows:**

| Task Type | Typical Workflow |

From nocksock/superpowers, MIT.

More from nocksock/superpowers