researching

Parallel Research Workflow

Gather the research question from the user, then delegate to the research-orchestrator agent for autonomous parallel investigation.

Input

Research question from $ARGUMENTS.

If $ARGUMENTS is empty, ask via AskUserQuestion:

- question: "What would you like to research?"
  options:
    - "How does X work?"
    - "Where is X implemented?"
    - "What patterns does X use?"
    - "Compare approaches for X"

Step 1: Generate Slug

Create a URL-safe slug from the research question (lowercase, hyphens, max 50 chars, strip articles).

mkdir -p docs/research

Step 2: Dispatch Research Orchestrator

Spawn the research-orchestrator agent via Task tool with:

• question: The research question
• slug: Generated slug
• project_path: Current project root

The agent autonomously decomposes the question, dispatches parallel specialist agents (locator, analyzer, pattern-finder), reconciles results, and writes the report to docs/research/{slug}-report.md.

Step 3: Present Results

After the orchestrator completes, read and present the report summary.

Offer next steps via AskUserQuestion:

- question: "Research complete. What next?"
  options:
    - label: "Deep dive"
      description: "Investigate a specific finding further"
    - label: "Create plan"
      description: "Use findings to create an implementation plan"
    - label: "Done"
      description: "Research is sufficient"

• "Deep dive" — re-run with a narrowed sub-question
• "Create plan" — load the planner:planning skill, passing the report path
• "Done" — report the report path and exit

Rules

Rule	Rationale
file:line references mandatory	Every technical claim must be verifiable
Synonym expansion for searches	Single terms miss aliased concepts
Progressive disclosure	Overview first, details on demand
Max 5 parallel agents	Resource and context limits
No fabricated paths	If not found, report that clearly