Research Auto - Full Lifecycle Automation

Trigger

/research-auto <topic> or /research-auto <topic> --auto

Execution Modes

Interactive Mode (Default)

•Asks 4 initial questions before starting
•Displays progress at each phase
•Produces final report with summary

Autonomous Mode (`--auto`)

•Uses sensible defaults for all decisions
•No prompts or confirmations
•Optimized for batch processing and CI/CD pipelines
•Higher throughput with larger batch sizes

Mode Comparison:

Setting	Interactive	Autonomous
Initial questions	4 questions	Skip all
Time range	User specified	"last 2 years"
Batch size	3	5
Batch approval	N/A (auto)	N/A (auto)
TOC fields	Auto-select	Auto-select
Error handling	Log and continue	Log and continue

Workflow

Step 0: Parse Arguments

Extract from user command:

•{topic}: Research topic (required)
•{auto_mode}: True if --auto flag present, False otherwise

Generate {topic_slug}:

•Convert to lowercase
•Replace spaces with hyphens
•Remove special characters (keep alphanumeric and hyphens)
•Truncate to 64 characters max

Step 1: Initial Configuration (Interactive Only)

Interactive Mode: Use AskUserQuestion to ask 4 questions:

Question 1 - Research Scope:

code

What is the scope of this research?
- Comprehensive (thorough coverage of all major items)
- Focused (specific subset or recent developments only)
- Quick overview (top 5-10 most important items)

Question 2 - Time Range:

code

What time range should the research cover?
- Last 6 months (recent developments)
- Last 1 year
- Last 2 years (Recommended)
- All time (historical perspective)

Question 3 - Research Depth:

code

What level of detail do you need?
- Brief (key facts only, faster execution)
- Moderate (balanced detail and speed)
- Detailed (comprehensive data, slower execution)

Question 4 - Domain Type (for intelligent defaults):

code

What type of research is this?
- Technical (software, frameworks, APIs, benchmarks)
- Financial (stocks, companies, markets, valuations)
- Academic (papers, research, citations)
- General (products, services, comparisons)

Store responses as:

•{scope}: "comprehensive" | "focused" | "quick"
•{time_range}: "last 6 months" | "last 1 year" | "last 2 years" | "all time"
•{depth}: "brief" | "moderate" | "detailed"
•{domain}: "technical" | "financial" | "academic" | "general"

Autonomous Mode: Use defaults:

•{scope}: "comprehensive"
•{time_range}: "last 2 years"
•{depth}: "moderate"
•{domain}: Auto-detect from topic keywords (see Domain Detection)

Step 2: Domain Detection (Autonomous Mode)

Auto-detect domain from topic keywords:

Financial Keywords: stock, company, market, valuation, investment, equity, earnings, financ, portfolio, dividend, revenue, profit Technical Keywords: software, framework, library, api, benchmark, tool, code, programming, developer, tech, ai, ml, llm Academic Keywords: paper, research, study, journal, citation, academic, science, review, literature

Detection logic:

•Convert topic to lowercase
•Check for keyword matches
•If multiple domains match, use priority: financial > technical > academic > general
•Default to "general" if no matches

Step 3: Apply Domain-Specific Defaults

Based on detected/selected domain, configure research parameters:

Technical Domain:

yaml

suggested_fields:
  - github_stars
  - release_date
  - license
  - programming_language
  - swe_bench_score
  - documentation_quality
batch_size: 5
items_per_agent: 1

Financial Domain:

yaml

suggested_fields:
  - market_cap
  - pe_ratio
  - revenue
  - profit_margin
  - dividend_yield
  - analyst_rating
batch_size: 3  # Rate limit consideration for financial APIs
items_per_agent: 1

Academic Domain:

yaml

suggested_fields:
  - citation_count
  - publication_date
  - journal
  - authors
  - impact_factor
  - methodology
batch_size: 5
items_per_agent: 1

General Domain:

yaml

suggested_fields:
  - name
  - release_date
  - company
  - key_features
  - pricing
  - user_rating
batch_size: 5
items_per_agent: 1

Step 4: Phase 1 - Outline Generation

Display progress:

code

## Phase 1/3: Generating Research Outline
Topic: {topic}
Mode: {auto_mode ? "Autonomous" : "Interactive"}
Time Range: {time_range}
Domain: {domain}

Execute outline generation internally (equivalent to /research {topic} --auto):

•Generate initial framework from model knowledge
•Web search for recent items within {time_range}
•Apply domain-specific field suggestions
•Create {topic_slug}/outline.yaml and {topic_slug}/fields.yaml

Outline Generation Process:

Parameter Retrieval:

•{topic}: User input research topic
•{YYYY-MM-DD}: Current date (retrieve using appropriate method)
•{time_range}: From Step 1 configuration
•{domain}: From Step 1/2 configuration

Generate Initial Framework (Model Knowledge): Based on topic, generate:

•Items list relevant to the domain
•Field framework using domain-specific suggestions

Web Search Supplement: Launch 1 web-search-agent to supplement items and fields.

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

Prompt Template:

python

prompt = f"""## 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 relevant to {domain} research

## 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)
"""

Create Output Files: Write {topic_slug}/outline.yaml:

yaml

topic: "{topic}"

items:
  - name: "Item 1"
    category: "Category A"
    description: "Brief description"
  # ... more items

execution:
  batch_size: {batch_size}
  items_per_agent: 1
  output_dir: ./results
  auto_mode: true

Write {topic_slug}/fields.yaml:

yaml

field_categories:
  - category: "Basic Info"
    fields:
      - name: "name"
        description: "Official name"
        detail_level: "{depth}"
        required: true
      # ... domain-specific fields

Log any errors to {topic_slug}/errors.log with timestamp.

Display completion:

code

Outline generated: {item_count} items, {field_count} fields
Files: {topic_slug}/outline.yaml, {topic_slug}/fields.yaml

Step 5: Phase 2 - Deep Research Execution

Display progress:

code

## Phase 2/3: Executing Deep Research
Items to research: {item_count}
Batch size: {batch_size}
Estimated batches: {ceil(item_count / batch_size)}

Execute deep research internally (equivalent to /research-deep --auto):

•Read outline.yaml and fields.yaml
•Execute all batches without approval prompts
•Run validation on each JSON output
•Log errors but continue on failures

Batch Execution Loop: For each batch of items:

•Display batch progress: Processing batch {n}/{total}: {item_names}
•Launch web-search-agent for each item in parallel (background)
•Wait for batch completion
•Run validation on outputs
•Log any failures to {topic_slug}/errors.log

Agent Prompt (same as /research-deep):

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

python

prompt = f"""## Task
Research {item_related_info}, output structured JSON to {output_path}

## Field Definitions
Read {fields_path} to get all field definitions

## Output Requirements
1. Output JSON according to fields defined in fields.yaml
2. Mark uncertain field values with [uncertain]
3. Add uncertain array at the end of JSON, listing all uncertain field names
4. All field values must be in English

## Output Path
{output_path}

## Validation
After completing JSON output, run validation script to ensure complete field coverage:
python ~/.claude/skills/research/validate_json.py -f {fields_path} -j {output_path}
Task is complete only after validation passes.
"""

Display batch completion:

code

Batch {n}/{total} complete: {success_count}/{batch_size} successful

After all batches:

code

Deep research complete: {completed}/{total} items
Failed items: {failed_count}
Results: {topic_slug}/results/

Step 6: Phase 3 - Report Generation

Display progress:

code

## Phase 3/3: Generating Report
Aggregating {completed} research results...

Execute report generation internally (equivalent to /research-report --auto):

•Scan JSON results for summary fields
•Auto-select top 3-5 most populated metric fields for TOC
•Generate Python conversion script
•Execute script to produce report.md

Auto-Selection Algorithm:

•Read all JSON files from results directory
•For each field, calculate population rate (% with non-null, non-uncertain value)
•Filter to numeric/metric fields (numbers, dates, scores, versions)
•Sort by population rate descending
•Select top 3-5 fields for TOC

Generate and Execute Script:

•Write {topic_slug}/generate_report.py
•Execute: python {topic_slug}/generate_report.py
•Verify {topic_slug}/report.md exists and is non-empty

Display completion:

code

Report generated: {topic_slug}/report.md

Step 7: Completion Summary

Display final summary:

code

## Research Complete

### Topic
{topic}

### Execution Mode
{auto_mode ? "Autonomous" : "Interactive"}

### Configuration
- Domain: {domain}
- Time Range: {time_range}
- Depth: {depth}
- Batch Size: {batch_size}

### Results
- Total items researched: {total_items}
- Successful: {completed}
- Failed: {failed}
- Items with uncertain values: {uncertain_count}

### Output Files
- Outline: {topic_slug}/outline.yaml
- Fields: {topic_slug}/fields.yaml
- Results: {topic_slug}/results/ ({json_count} JSON files)
- Report: {topic_slug}/report.md
- Errors: {topic_slug}/errors.log (if any)

### Failed Items (if any)
- Item Name: Error reason
...

### Next Steps
- Review report: cat {topic_slug}/report.md
- Add more items: /research-add-items
- Add more fields: /research-add-fields
- Re-run deep research: /research-deep

Error Handling

All errors are logged to {topic_slug}/errors.log with format:

code

[YYYY-MM-DD HH:MM:SS] [PHASE] [SEVERITY] Error message
Item: {item_name} (if applicable)
Details: {error_details}
---

Error Severity Levels:

•ERROR: Task failed but workflow continues
•FATAL: Workflow must stop

Scenario	Severity	Response
Invalid topic	FATAL	Display error, exit
Web search fails	ERROR	Use model knowledge only, continue
Agent failure	ERROR	Log, mark item failed, continue
Validation failure	ERROR	Log, continue with warnings
Script generation fails	ERROR	Log, skip report generation
Script execution fails	ERROR	Log, display partial results
All items fail	FATAL	Display error summary, exit

Domain Detection Keywords

python

DOMAIN_KEYWORDS = {
    "financial": [
        "stock", "company", "market", "valuation", "investment",
        "equity", "earnings", "financ", "portfolio", "dividend",
        "revenue", "profit", "nasdaq", "nyse", "s&p", "fund",
        "bond", "etf", "ipo", "merger", "acquisition"
    ],
    "technical": [
        "software", "framework", "library", "api", "benchmark",
        "tool", "code", "programming", "developer", "tech",
        "ai", "ml", "llm", "agent", "model", "algorithm",
        "database", "cloud", "devops", "kubernetes", "docker"
    ],
    "academic": [
        "paper", "research", "study", "journal", "citation",
        "academic", "science", "review", "literature", "thesis",
        "dissertation", "arxiv", "peer-review", "methodology"
    ]
}

Output Artifacts

code

{topic_slug}/
  ├── outline.yaml        # Research items and config
  ├── fields.yaml         # Field definitions
  ├── results/            # JSON research results
  │   ├── item_1.json
  │   ├── item_2.json
  │   └── ...
  ├── generate_report.py  # Report generation script
  ├── report.md           # Final markdown report
  └── errors.log          # Error log (if any errors)

Examples

Interactive Mode

code

> /research-auto AI Coding Assistants

## Initial Configuration

What is the scope of this research?
> Comprehensive

What time range should the research cover?
> Last 2 years

What level of detail do you need?
> Moderate

What type of research is this?
> Technical

## Phase 1/3: Generating Research Outline
...

Autonomous Mode

code

> /research-auto AI Coding Assistants --auto

## Phase 1/3: Generating Research Outline
Topic: AI Coding Assistants
Mode: Autonomous
Time Range: last 2 years
Domain: technical (auto-detected)

Outline generated: 15 items, 12 fields
...

## Phase 2/3: Executing Deep Research
...

## Phase 3/3: Generating Report
...

## Research Complete
...

Related Commands

•/research - Generate outline only
•/research-add-items - Add items to existing outline
•/research-add-fields - Add fields to existing definition
•/research-deep - Execute deep research only
•/research-report - Generate report only

Research Auto - Full Lifecycle Automation

Trigger

Execution Modes

Interactive Mode (Default)

Autonomous Mode (--auto)

Workflow

Step 0: Parse Arguments

Step 1: Initial Configuration (Interactive Only)

Step 2: Domain Detection (Autonomous Mode)

Step 3: Apply Domain-Specific Defaults

Step 4: Phase 1 - Outline Generation

Step 5: Phase 2 - Deep Research Execution

Step 6: Phase 3 - Report Generation

Step 7: Completion Summary

Error Handling

Domain Detection Keywords

Output Artifacts

Examples

Interactive Mode

Autonomous Mode

Related Commands

Autonomous Mode (`--auto`)