research

name	research
description	Conduct preliminary multi-source research on a topic and generate a reusable research outline. Use for academic research, benchmark research, technology selection, source-taxonomy planning, and preparation for research-deep.

Research - Preliminary Research

Workflow

Step 1: Generate the initial framework

Use model knowledge only to propose:

• Main research objects or items
• Suggested field categories and fields

Label this output as provisional and do not present unsourced model knowledge as verified fact. Ask the user to confirm additions, removals, and the field framework.

Step 2: Set search parameters

Ask for the time range, such as the last six months, since a specific year, or unlimited. Read references/source-taxonomy.md completely before launching web research.

Resolve these parameters:

• {topic}: User-provided topic
• {current_date}: Current date in YYYY-MM-DD
• {initial_framework}: Confirmed Step 1 output
• {time_range}: User-selected range
• {source_policy_path}: Absolute path to references/source-taxonomy.md

Launch one web-search agent. Reproduce the following prompt exactly except for variable substitution:

prompt = f"""## Task
Research topic: {topic}
Current date: {current_date}
Time range: {time_range}

Supplement the confirmed initial framework with verified items and research fields.

## Existing Framework
{initial_framework}

## Source Policy
Read {source_policy_path} completely and follow it strictly.

Search official, academic, implementation, industry, and community sources independently. Prioritize primary sources. Treat sources as independent only when they come from different organizations, publishers, or research groups. Record a coverage gap instead of adding a weak source.

## Goals
1. Verify whether important research objects are missing.
2. Propose missing objects supported by evidence.
3. Search for relevant objects within {time_range}.
4. Propose useful research fields supported by evidence.
5. Report coverage for all five source categories.

## Output Requirements
Return structured results directly. Do not write files.

### Supplementary Items
- item_name: Name
  explanation: Why it should be added
  evidence_source_ids: [S1, S2]

### Recommended Supplementary Fields
- field_name: Name
  description: What the field captures
  rationale: Why this dimension is needed
  evidence_source_ids: [S2]

### Source Findings
- id: S1
  title: Source title
  url: Canonical URL
  source_category: official|academic|implementation|industry|community
  source_type: Taxonomy subtype
  publication_date: YYYY, YYYY-MM, YYYY-MM-DD, or null
  relevance: Why the source is relevant
  key_finding: Finding supported by the source
  credibility:
    level: high|medium|low
    rationale: Concise assessment
  supports_fields: [supplementary_items.item_name]

### Source Coverage
- category: official|academic|implementation|industry|community
  status: found|searched_no_reliable_source|not_applicable
  notes: Coverage summary
"""

Step 3: Merge existing field definitions

Ask whether the user has an existing field-definition file. Read and merge it without silently overwriting conflicting definitions.

Step 4: Generate the outline

Merge the confirmed framework, verified supplements, and existing fields into two files.

outline.yaml must contain:

topic: Research topic
items: []
source_policy:
  categories: [official, academic, implementation, industry, community]
  prioritize_primary_sources: true
  minimum_independent_publishers: 2
  corroborate_important_claims: true
  allow_documented_coverage_gaps: true
execution:
  batch_size: 1
  items_per_agent: 1
  output_dir: ./results

Confirm batch_size and items_per_agent with the user.

fields.yaml must contain field categories and definitions. Give every field a name, description, detail_level, and required value. Use brief, moderate, or detailed for detail_level. Reserve uncertain, _sources, and _source_coverage; do not define them as domain fields.

Step 5: Save and confirm

Create ./{topic_slug}/, save outline.yaml and fields.yaml, and show both paths for confirmation.

Follow-up skills

• research-add-items: Add research objects
• research-add-fields: Add field definitions
• research-deep: Run item-level deep research