deep-research
Deep Research Agent for complex, multi-step research requiring web research, planning, and synthesis.
Skill by mahsumaktas
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/mahsumaktas-deep-research
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: writing, research, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/deep-research
What's inside
- Skill name
deep-research- Description
- Deep Research Agent for complex, multi-step research requiring web research, planning, and synthesis. Use instead of single web_search for any topic requiring multi-angle investigation. Trigger on: 'what is X', 'research X', 'compare X and Y', before content generation tasks.
- License
- MIT
- Instructions
- 215 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Deep Research Agent
> "Complexity is not an obstacle; it's the raw material for structured decomposition."
Kapsamlı, çok kaynaklı araştırma için sistematik metodoloji. Tek web_search yeterli olmadığında kullan.
## Core Principle
**Tek arama = yüzeysel bilgi.** Kaliteli çıktı = kaliteli araştırma. Asla genel bilgiden içerik üretme.
---
## Araştırma Metodolojisi (DeerFlow-inspired)
### Phase 1 — Broad Exploration (Geniş Tarama)
Ana konuyu anlamak için geniş sorgular:
1. **İlk tarama**: Konuyu genel olarak tara
2. **Boyutları belirle**: Hangi alt konular, açılar, perspektifler var?
3. **Harita çiz**: Farklı bakış açıları, paydaşlar, tartışmalı noktalar
```
Örnek: "Oracle agent mimarisi"
İlk sorgular:
- "AI agent orchestration architecture 2025"
- "multi-agent system design patterns"
- "LLM agent memory context management"
Tespit edilen boyutlar:
- Memory sistemleri (short/long term)
- Tool filtering ve güvenlik
- Trace/observability
- Subagent delegation patterns
```
### Phase 2 — Deep Dive (Derinlemesine)
Her önemli boyut için hedefli araştırma:
1. **Spesifik sorgular**: Her alt konu için ayrı keyword
2. **Farklı ifadeler dene**: Aynı konuyu farklı kelimelerle ara
3. **Tam içeriği oku**: web_fetch ile sadece snippet değil, tam sayfa
4. **Referansları takip et**: Kaynakların bahsettiği diğer kaynakları ara
### Phase 3 — Diversity & Validation (Çeşitlilik)
| Bilgi Tipi | Amaç | Sorgu Örnekleri |
|---|---|---|
| **Gerçek & Data** | Somut kanıt | "statistics", "benchmark", "numbers" |
| **Örnekler & Vakalar** | Gerçek uygulama | "case study", "example", "implementation" |
| **Uzman Görüşleri** | Otorite perspektifi | "expert analysis", "opinion", "commentary" |
| **Trendler** | Gelecek yönü | "trend 2025", "forecast", "future" |
| **Karşılaştırma** | Bağlam | "vs", "comparison", "alternatives" |
| **Eleştiriler** | Dengeli bakış | "challenges", "limitations", "criticism" |
### Phase 4 — Reflection Generation (Yansima Uretimi) [v2]
Her onemli kaynak icin distilled insight olustur. Ham icerik yerine yoğunlastirilmis bilgi kaydet.
**Her kaynak icin:**
1. URL + baslik kaydet
2. 2-3 cumlede distilled insight yaz (reflection)
3. `memory_store` ile kaydet: `category="research_reflection"`
**Reflection formati:**
```
- [Kaynak Basligi](URL): [2-3 cumle distilled insight]
```
**Ornekler:**
Iyi:
```
- [LATS: Language Agent Tree Search](https://arxiv.org/abs/2310.04406):
MCTS + LLM reflection birlestiren agent search. Her basarisiz denemede
self-reflection uretip sonraki denemeyi yonlendiriyor. ReAct'tan %22
daha iyi HumanEval'da.
```
Kotu:
```
- [LATS paper](URL): Ilginc bir calisma, agent'lar icin onemli gorusuyor.
```
**Kurallar:**
- Spesifik veri, somut bulgu, uygulanabilir cikti odakli ol
- YASAK: Genel ifadeler ("ilginc", "onemli", "dikkat cekici")
- Her reflection bagimsiz anlasilabilir olmali
- Minimum 5, maksimum 20 reflection (konu genisligine gore)
### Phase 5 — Synthesis from Reflections (Yansimalardan Sentez) [v2]
Icerik uretimini ham kaynak yerine SADECE reflection'lardan yap.
```
sessions_spawn(
task="Asagidaki arastirma reflection'larindan kapsamli sentez yaz:\n\n[TUM REFLECTION'LAR]\n\nKurallar:\n- Sadece reflection'lardaki bilgiyi kullan\n- Kaynak URL'lerini icerige dahil et\n- Celiskili bulgulari belirt\n- Bilgi boslugu varsa isaretleKonular arasi baglanti kur",
model="sonnet-4-6"
)
```
**Token tasarrufu:**
- ~50K ham veri → ~3K reflection = ~%94 azalma
- Writer subagent'a sadece reflection'lar gider (tam content DEGIL)
- Context penceresi korunur, sentez kalitesi artar (gurultu azalir)
**Reflection yetersizse:**
- Phase 4'teki minimum 5 kurali kontrole et
- Eksik boyut varsa Phase 2-3'e don, ek arastirma yap
- Yeni reflection'lar ekle, sonra Phase 5'i tekrarla
### Phase 6 — Synthesis Check (Sentez Kontrolu)
Icerik uretimine gecmeden once:
- [ ] En az 3-5 farkli acidan aradim mi?
- [ ] Onemli kaynaklari tam okudum mu?
- [ ] Somut veri, ornek, uzman gorusu var mi?
- [ ] Hem olumlu hem elestirel perspektif var mi?
- [ ] Bilgi guncel mi?
- [ ] Reflection'lar spesifik ve uygulanabilir mi? [v2]From mahsumaktas/agent-evolution-kit, MIT.
More from mahsumaktas/agent-evolution-kit
- 0.443score
prompt-guard
mahsumaktas
Advanced prompt injection defense system for Clawdbot.
SkillDFreshness: D★ 2Claude Code +9 - 0.443score
memory-hygiene
mahsumaktas
Audit, clean, and optimize Clawdbot's vector memory (LanceDB).
SkillDFreshness: D★ 2Claude Code +9 - 0.443score
news-aggregator-skill
mahsumaktas
Comprehensive news aggregator that fetches, filters, and deeply analyzes real-time content from 8 major sources: Hacker News, GitHub Trending, Product Hunt…
SkillDFreshness: D★ 2Claude Code +9 - 0.440score
agent-governance
mahsumaktas
Agent islemlerini policy'ye gore denetle.
SkillDFreshness: D★ 2Claude Code +9 - 0.436score
security-review
mahsumaktas
Security audit checklist for code changes.
SkillDFreshness: D★ 2Claude Code +9 - 0.436score
orchestrator-loop
mahsumaktas
Orchestration pipeline for development tasks.
SkillDFreshness: D★ 2Claude Code +9 - 0.436score
hn
mahsumaktas
Browse Hacker News - top stories, new, best, ask, show, jobs, and story details with comments.
SkillDFreshness: D★ 2Claude Code +9 - 0.434score
agent-system-output-hygiene-ops
mahsumaktas
Audit and remediate AgentSystem cron output hygiene by detecting raw tool logs, stack traces, and noisy summaries.
SkillDFreshness: D★ 2Claude Code +9 - 0.434score
agent-system-learning-loop
mahsumaktas
Evaluate and improve AgentSystem learning quality for memory/self-compound jobs.
SkillDFreshness: D★ 2Claude Code +9 - 0.434score
agent-system-stability-ops
mahsumaktas
Audit and harden AgentSystem scheduler reliability, cron contract compliance, and agent consistency.
SkillDFreshness: D★ 2Claude Code +9 - 0.430score
agent-system-memory-proactivity-ops
mahsumaktas
Audit and improve AgentSystem memory retention, proactive behavior, learning quality, and agent consistency.
SkillDFreshness: D★ 2Claude Code +9 - 0.427score
agent-system-live-quality-ops
mahsumaktas
Operate and validate AgentSystem live quality alarm (systemd timer + alert state) to keep memory/proactivity/consistency regressions visible in real time.
SkillDFreshness: D★ 2Claude Code +9
Signals
- Stars
- 2
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Mar 1, 2026 (8 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from mahsumaktas
- 0.443score
prompt-guard
mahsumaktas
Advanced prompt injection defense system for Clawdbot.
SkillDFreshness: D★ 2Claude Code +9 - 0.443score
memory-hygiene
mahsumaktas
Audit, clean, and optimize Clawdbot's vector memory (LanceDB).
SkillDFreshness: D★ 2Claude Code +9 - 0.443score
news-aggregator-skill
mahsumaktas
Comprehensive news aggregator that fetches, filters, and deeply analyzes real-time content from 8 major sources: Hacker News, GitHub Trending, Product Hunt…
SkillDFreshness: D★ 2Claude Code +9 - 0.440score
agent-governance
mahsumaktas
Agent islemlerini policy'ye gore denetle.
SkillDFreshness: D★ 2Claude Code +9 - 0.436score
security-review
mahsumaktas
Security audit checklist for code changes.
SkillDFreshness: D★ 2Claude Code +9 - 0.436score
orchestrator-loop
mahsumaktas
Orchestration pipeline for development tasks.
SkillDFreshness: D★ 2Claude Code +9
Cite this page
deep-research skill. Agent Mart. https://agentmart.dev/skills/mahsumaktas-deep-research (last updated 2026-10-24)