Brutal Deepresearch

date +%Y-%m-%d

0.3 Determine Next DR Number

Find the highest existing DR number:

ls workspace/research/ 2>/dev/null | grep -oE '[0-9]{4}' | sort -rn | head -1

New DR number = highest + 1. If no research sessions exist, start at 0001.

0.4 Generate Session Slug

Generate a slug from the research topic:

• Lowercase
• Replace spaces and special characters with hyphens
• Max 40 characters
• Remove trailing hyphens

Session directory: workspace/research/DR-<NNNN>-<slug>/

Do not create the directory yet. Wait until Gate 2 confirmation.

Step 0.5: Resume Mode (Conditional)

Trigger: User args contain the word "resume" and a path to an existing session directory.

Example invocation: /brutal-deepresearch resume research workspace/research/DR-0001-ai-coding

If resume mode is NOT detected, skip to Step 1.

0.5.1 Read Existing Session

1. Read outline.yaml from the given session path to get items list
2. Read fields.yaml from the given session path to get field definitions
3. Read progress.yaml (if it exists) for previous execution context

0.5.2 Check Completed Results

ls <session_path>/results/*.json 2>/dev/null
ls <session_path>/results/*.started 2>/dev/null

Determine item status:

• .json exists → completed (skip this item)
• .started exists but no .json → interrupted (re-research this item)
• Neither exists → never started (research this item)

0.5.3 Calculate Remaining Items

Compare items in outline.yaml against completed results:

• Log which items are completed and will be skipped
• Log which items were interrupted and will be re-researched
• Log which items never started and will be researched

0.5.4 Branch to Execution or Report

• If remaining items exist: Skip directly to Step 6 (Execute Deep Research), launching agents only for remaining items
• If all items are complete: Report "All items already completed" and skip directly to Step 7 (Report Configuration)

Resume mode skips Steps 1-4 entirely — the outline and fields are already confirmed from the previous session.

Step 1: Generate Initial Framework

Based on the user's research topic, use model knowledge to generate:

1. Items List: The main research objects/items in this domain. Each item should have:

   • name: Item name
   • category: Classification (if applicable)
   • description: Brief description of why this item is relevant

2. Field Framework: Suggested research field categories and fields per category. Each field should have:

   • name: Field name (snake_case)
   • description: What this field captures
   • detail_level: One of brief, moderate, or detailed

Present the framework to the user in a readable format.

Step 2 - GATE 1: Confirm Initial Framework

This is a hard gate. Do not proceed past this step without explicit user confirmation.

Present:

• The items list with names, categories, and descriptions
• The field framework organized by category

Use AskUserQuestion to ask:

• "Are these items and fields correct? Add/remove anything?"

Hard gate: Do not proceed until user confirms. User can request additions or removals here.

Step 3: Web Search Supplement

3.1 Get Time Range

Use AskUserQuestion to ask for time range:

• Last 6 months
• Since 2024
• Since 2025
• Unlimited

3.2 Launch Web Search Agent

Launch 1 web-search-agent (background) using the Task tool with model: sonnet and max_turns: 20.

Parameter Retrieval:

• {topic}: User's research topic
• {YYYY-MM-DD}: Current date from Step 0.2
• {step1_output}: Complete output from Step 1 (items list + field framework)
• {time_range}: User-specified time range

Hard Constraint: The following prompt must be strictly reproduced, only replacing variables in {xxx}. Do not modify structure or wording.

Prompt Template:

You are an elite internet researcher. Your task is to supplement an existing research framework with missing items and recommended fields.

## Research Methodology

Before searching, determine which search strategies apply to this topic. Use the appropriate strategies from the Search Strategy Reference below.

Get today's date first:
date +%Y-%m-%d

Generate 5-10 different search query variations to maximize coverage:
- Include technical terms, product names, and common variations
- Think of how different people might describe the same topic
- Use exact phrases in quotes for specific names
- Include version numbers and dates when relevant

## Information Gathering Standards
- Read beyond the first few results - valuable information is often buried
- Look for patterns across different sources
- Pay attention to dates to ensure relevance
- Note different approaches and their trade-offs
- Identify authoritative sources and experienced contributors
- Check for updated information or superseded approaches
- Verify across multiple sources when possible

## Task
Research topic: {topic}
Current date: {YYYY-MM-DD}

Based on the following initial framework, supplement latest items and recommended research fields.

## Existing Framework
{step1_output}

## Goals
1. Verify if existing items are missing important objects
2. Supplement items based on missing objects
3. Continue searching for {topic} related items within {time_range} and supplement
4. Supplement new fields

## Output Requirements
Return structured results directly (do not write files):

### Supplementary Items
- item_name: Brief explanation (why it should be added)
...

### Recommended Supplementary Fields
- field_name: Field description (why this dimension is needed)
...

### Sources
- [Source1](url1)
- [Source2](url2)

One-shot Example (assuming researching AI Coding History):

## Task
Research topic: AI Coding History
Current date: 2025-12-30

Based on the following initial framework, supplement latest items and recommended research fields.

## Existing Framework
### Items List
1. GitHub Copilot: Developed by Microsoft/GitHub, first mainstream AI coding assistant
2. Cursor: AI-first IDE, based on VSCode
...

### Field Framework
- Basic Info: name, release_date, company
- Technical Features: underlying_model, context_window
...

## Goals
1. Verify if existing items are missing important objects
2. Supplement items based on missing objects
3. Continue searching for AI Coding History related items within since 2024 and supplement
4. Supplement new fields

## Output Requirements
Return structured results directly (do not write files):

### Supplementary Items
- item_name: Brief explanation (why it should be added)
...

### Recommended Supplementary Fields
- field_name: Field description (why this dimension is needed)
...

### Sources
- [Source1](url1)
- [Source2](url2)

3.3 Merge Findings

After the web search agent completes, merge its findings with the initial framework:

• Add supplementary items to the items list (avoid duplicates)
• Add recommended fields to the field framework
• Note sources for traceability

Step 4 - GATE 2: Confirm Final Outline

This is a hard gate. Do not proceed past this step without explicit user confirmation.

Present the merged outline:

• Complete items list (original + web search additions, clearly marked)
• Complete field framework (original + web search additions, clearly marked)

Use AskUserQuestion to confirm the outline is correct.

Add-Items/Add-Fields Loop

User can say "add X item" or "add Y field" at this gate. If they do:

1. Add the requested item/field to the framework
2. Re-present the updated framework
3. Ask for confirmation again

Repeat until user explicitly confirms. Do not generate files until confirmed.

4.1 Create Session Directory and Write Files

After confirmation:

mkdir -p workspace/research/DR-<NNNN>-<slug>/results

Write outline.yaml: