name: deep-research description: Create educational documents that build understanding progressively with citations

Research Document

A systematic approach to creating educational documentation through tiered research.

Purpose

Creates in-depth explanatory documents that:

Build understanding from fundamentals to specifics
Use progressive disclosure to guide the reader
Provide complete citations for all information
Use visual summaries and tables for dense information

Roles

You (reading this file) are the orchestrator.

Role	Reads	Does
Orchestrator (you)	SKILL.md, next-step.py output	Runs state machine, spawns subagents, writes outputs
State machine	progress.json, workspace files	Decides next action, validates gates
Subagent	Phase file (e.g., SCOUT.md)	Executes phase instructions

⚠️ Do not read the files in phases/ — they are instructions for your subagents. Pass them via context_files.

When to Use

User asks for comprehensive explanations of systems or features
Need to document how components work end-to-end
Creating educational content about architecture or processes
Questions like "Explain how X works" or "What is the Y process?"

For quick factual lookups, use fact-find instead.

Directory Structure

skills/deep-research/
├── SKILL.md              # This file (for orchestrator)
├── next-step.py          # State machine
└── phases/               # For subagents only - do not read
    ├── SCOUT.md
    ├── RESEARCH.md
    ├── ASSEMBLE.md
    └── VERIFY.md

Workspace Layout

All artifacts go to planning/<question-slug>/:

planning/how-twoliter-builds-kits/
├── progress.json         # State machine state
├── question.txt          # Original question
├── 00-scout.md           # Scout findings + sub-questions
├── 01-kit-structure.md   # Sub-question answer
├── 02-build-command.md   # Sub-question answer
├── verify-1.txt          # Citation verification
├── verify-2.txt          # Citation verification
└── FINAL.md              # Assembled document

Orchestrator Loop

The orchestrator runs the state machine and spawns subagents.

Pseudocode (any agentic system)

slug = slugify(user_question)
workspace = "planning/" + slug

create workspace directory
write user_question to workspace/question.txt

loop:
    # Ask state machine what to do next
    action = run("python3 skills/deep-research/next-step.py <workspace>")
    parse action as JSON
    
    if action.type == "done":
        read workspace/FINAL.md
        break
    
    if action.type == "gate_failed":
        log "Gate failed: " + action.reason
        break
    
    if action.type == "spawn":
        # Spawn subagent with phase file - DO NOT read it yourself
        result = spawn_subagent(
            prompt = action.prompt,
            context_files = action.context_files,  # includes phase file
            context_data = action.context_data
        )
        write result to workspace/<action.output_file>

Python variant (run_agent_program systems)

import json

slug = "question-slug"  # derive from user question
workspace = f"planning/{slug}"

bash(f"mkdir -p {workspace}", on_error="raise")
write("create", f"{workspace}/question.txt", file_text=user_question)

while True:
    result = bash(f"python3 skills/deep-research/next-step.py {workspace}", on_error="raise")
    action = json.loads(result)
    
    if action["type"] == "done":
        final = fs_read("Line", f"{workspace}/FINAL.md", 1, 9999)
        break
    
    if action["type"] == "gate_failed":
        log(f"Gate failed: {action['reason']}")
        break
    
    if action["type"] == "spawn":
        r = spawn(
            action["prompt"],
            context_files=action["context_files"],
            context_data=action.get("context_data"),
            allow_tools=True
        )
        write("create", f"{workspace}/{action['output_file']}", file_text=r.response)

Handling Exceptions

The state machine handles the happy path. When things go wrong, exercise judgment:

Exception	Response
Spawn times out	Assess: retry with longer timeout? Report partial progress?
Spawn returns error	Report failure to state machine, let it track retries
Empty/invalid response	Treat as failure, report to state machine

Don't silently advance past failures. Either retry, fail explicitly, or document gaps.

Phases

Phase 1: Scout

Discovers what exists and formulates sub-questions. Writes 00-scout.md.

Phase 2: Research

Answers each sub-question from the scout phase. Writes NN-*.md files. The state machine loops until all sub-questions are answered.

Phase 3: Assemble

Combines all research files into FINAL.md.

Phase 4: Verify

Checks each citation in FINAL.md. Spawns one verifier per citation.

Gates

The state machine validates between phases:

Gate	Validation
Scout → Research	`00-scout.md` exists with sub-questions
Research → Assemble	Count of `NN-*.md` files matches sub-question count
Assemble → Verify	`FINAL.md` exists
Verify → Done	All citations verified

Research Quality Indicator

Documents end with one of:

✅ Answered from documentation - Fully answered from README files, design docs
⚠️ Answered from source code - Had to read implementation files
🔍 Partial documentation - Required both docs and source code
❓ Gaps remain - Some sub-questions could not be answered

deep-research

$ 설치