Skip to content
Back to skills

Deep Research

ASecurity

Disciplined multi-source research with triangulation, citations, and adversarial review. Use for high-stakes questions, comparisons, strategy groundwork, or hypothesis validation — not quick fact-checks.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 23, 2026
ai-agentsgogit

Security analysis

A100/100

Scanned September 23, 2026

npx -y skills add danmo-ai/danmo-work --skill deep-research --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Deep Research?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Deep Research
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/danmo-ai-deep-research/badge)](https://www.skillsdirectory.com/skills/danmo-ai-deep-research)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: deep-research
source: builtin
description: Disciplined multi-source research with triangulation, citations, and adversarial review. Use for high-stakes questions, comparisons, strategy groundwork, or hypothesis validation — not quick fact-checks.
license: MIT
compatibility: Requires web_search, web_fetch, write; read_file/grep for existing project context
metadata:
  author: danmo-work
  version: "1.0"
  category: work
  adapted_from: "https://github.com/alirezarezvani/claude-skills/tree/main/research/deep-research/skills/deep-research"
  upstream_license: MIT
---

# Deep Research

> Adapted from alirezarezvani/claude-skills `deep-research` (© Alireza Rezvani / contributors, MIT).
> Rewritten for Danmo Work tools and agents; not a verbatim copy.

Use when a wrong answer is expensive. Prefer a short direct answer for low-risk fact checks.

## Pipeline

1. **Reframe** — restate the decision behind the question; list 2–4 falsifiable hypotheses.
2. **Plan** — scope, report genre (qa / decision / landscape / validation), stop criteria, opposition queries.
3. **Existing work** — search the workspace/KB first; do not re-research what you already have.
4. **Search** — multiple `web_search` queries; fetch primary sources with `web_fetch`. Prefer diverse source types (primary, docs, academic/industry, reputable discussion).
5. **Persist sources** — when writing files is allowed, save each source (metadata + verbatim quotes) under a research folder; never invent URLs.
6. **Triangulate** — each major thesis needs ≥3 independent sources of different types, or mark “insufficient evidence”.
7. **Synthesize + adversarial** — answer hypotheses; steel-man counter-arguments; note confidence and gaps.
8. **Deliver** — final report with citations; optional `refresh_targets` (entities/numbers to revisit later).

## Output shape (when persisting)

```
research/<slug>/
  plan.md
  sources/01_....md
  report.md
```

If file writes are not available, keep the same structure in the chat report and quote sources with URLs.

## Anti-patterns

- One-shot search then confident prose
- Fabricated or unfetched citations
- Skipping the adversarial pass on medium/deep work
- Treating same-outlet reprints as independent sources
- Dumping only chat text when the user asked for reusable research files

## Stop condition

Report that confirms/refutes each hypothesis (or marks under-determined), with citations and explicit confidence. Do not propose implementation unless asked.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.