# Deep Research Skill

> Multi-phase research agent system that collects data on any topic and generates comprehensive, citation-verified research reports. Use when the user asks to: research a topic, analyze a market, study an industry, investigate policy or regulatory trends, analyze technology trends, or generate any data-driven report. Triggers on requests like "research X", "analyze the Y market", "what are the trends in Z", "research this topic", or any request for a structured research report with sources. Requires AGENTS.md to be present in the working directory for full procedures.

- Skill: `godstale/deep-research-skill` (Agent Skill, multi-file: 8 files)
- Install (CLI): `npx skillmds@latest add godstale/deep-research-skill`
- Raw SKILL.md: https://api.skillmd.com/api/skills/godstale/deep-research-skill/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: godstale (https://skillmd.com/u/godstale)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/godstale/deep-research-skill

---


# Research Agent System

Read `AGENTS.md` in the current working directory for complete per-phase procedures, sub-agent
prompts, and verification criteria **before starting any research**.

## Session Start (mandatory)

1. Check `docs/` for folders with incomplete `TODO.md` -> resume if found, else create new
2. Create `docs/[SLUG]/` -- SLUG is uppercase English words separated by `_` (e.g. `NIGERIA_COSMETICS`)
3. Write `TODO.md` using the **exact template** defined in `AGENTS.md` (no simplification)
4. Create `LOG.txt` with first session-start entry
5. If `SOURCES.md` exists in topic folder -> read it before searching

## Workflow

```
Phase 1-A: Broad Search -- 3 Sub-Agents in parallel (SA-1-A, SA-1-B, SA-1-C)
     |
Phase 2-1: First-pass data collection
     |
  [Data sufficiency check]
  Sufficient ---------> Phase 3: Report generation
  Insufficient |                   |
Phase 1-B: Trusted source discovery   Phase 4: Verification
     |                                    |
Phase 1-C: Targeted search          Pass -> done / Fail -> re-run Phase 1 (max 3x)
     |
Phase 2-2: Second-pass collection -> Phase 3 -> Phase 4
```

## Report Language

- Default: **English**
- Request written in non-English -> use that language
- Explicit override (e.g. "write in Korean") -> use that language
- Record language in `TODO.md` at project creation

## Tool Priority

| Purpose | 1st (free) | 2nd (free) | Last resort (credits) |
|---------|-----------|-----------|----------------------|
| Search | WebSearch | -- | firecrawl search |
| Page fetch | WebFetch | agent-browser / Playwright MCP | firecrawl scrape |

## Core Rules

1. **Never fabricate numbers, URLs, or sources** -- mark as `unknown` if unclear
2. Report must use **only data in `SEARCH_RESULT.md`**
3. Every stat must include reference year + source: `$2.3B (2023, World Bank)`
4. **Never delete existing files** -- only append/update
5. Save `TODO.md` and `LOG.txt` before every session end

## Files Per Phase

| Phase | Files to create/update |
|-------|------------------------|
| Session start | `TODO.md`, `LOG.txt` |
| Phase 1-A/1-C | `WEB_SEARCH_RAW.md` (append), `TODO.md` |
| Phase 1-B | `SOURCES.md` |
| Phase 2 | `scraped/[domain]-[date].txt`, `SEARCH_RESULT.md`, `TODO.md` |
| Phase 3 | `REPORT.md` |
| Phase 4 | `VERIFICATION.md` |
| Every phase | `LOG.txt` (append) |

## Data Sufficiency (assessed after Phase 2-1)

**Sufficient** if ALL: 15+ pages collected, core quantitative data present, 2+ independent sources, data within last 3 years.

**Insufficient** if ANY: fewer than 10 pages, missing core metrics, single source only, data older than 5 years.

---

For detailed sub-agent prompts, Phase 1-A query groups, Phase 2 extraction targets, `REPORT.md`
structure template, and `VERIFICATION.md` checklist -> see `AGENTS.md`.