academic-research

これはレビュー時に保存された抜粋です。完全で最新の内容は外部ソースを確認してください。

---
name: academic-research
description: Academic-grade research (L4) — runs /expert-research (L3) then adds academic sources (arXiv, Google Scholar), full multi-agent crew (Planner + 2x Researchers + Critic + Editor), timeline analysis, methodology section, and annotated bibliography with source quality ratings. ~40 min, 80-120 sources, 5000+ word report. Use for scientific overviews, research-grade investigations.
user_invocable: true
---

# Academic Research — L4

Academic-grade research tier. **Composes** on top of L3 (`/expert-research`) by adding:
- **Academic sources** — arXiv, Google Scholar, SSRN, research.google, research papers
- **Full multi-agent crew** — Planner, Researcher × 2 (parallel), Critic, Editor
- **Timeline / history analysis** — how the topic evolved over time
- **Methodology section** — explicit description of research approach
- **Cross-language search** — EN + RU + domain-specific languages
- **Quality-rated bibliography** — A/B/C/D ratings per source
- **Open questions section** — what the literature hasn't answered

**Position in ladder:** L4. Calls L3 (which chains through L2 and L1). Called by L5.

## When to Use

- Scientific/technical deep-dives: "как работают LLM router-агенты"
- Research-grade investigations needing peer-reviewed backing
- Topics with substantial academic literature (ML/AI, distributed systems, algorithms)
- When user says "академический", "научный", "peer-reviewed", "с научной точки зрения"
- Complex topics where vendor-driven sources aren't enough

**Escalate to:**
- `/ultra-research` (L5) — when you want a full knowledge base + playbooks + memory sync

## Budget

- Time: ~40 min total (L1: 5 + L2: 7 + L3: 8 + L4: 20)
- Tavily credits: ~400
- Sub-questions: 12
- Sources: 80–120 (including 15–25 academic)
- Output: 5000+ word academic report + annotated bibliography + timeline + methodology

## Pipeline

```
STAGE 1: Execute /expert-research (L3)
    → L1/ + L2/ + L3/ artifacts

STAGE 2: L4 Academic Layer
  1. ACADEMIC PLANNING           — scope the academic angle
  2. MULTI-AGENT CREW SETUP      — spawn crew in parallel
     ├─ Planner agent            — generates 4 new academic sub-qs
     ├─ Researcher A (parallel)  — primary academic sources
     ├─ Researcher B (parallel)  — cross-domain / adjacent fields
     ├─ Critic (after R A+B)     — adversarial review of academic layer
     └─ Editor (last)            — final synthesis pass
  3. ACADEMIC SOURCE HARVESTING  — arXiv, Scholar, SSRN
  4. CROSS-LANGUAGE SEARCH       — EN + RU + other relevant
  5. TIMELINE CONSTRUCTION       — chronological development
  6. METHODOLOGY DOCUMENTATION   — how research was done
  7. QUALITY-RATED BIBLIOGRAPHY  — A/B/C/D per source
  8. ACADEMIC SYNTHESIS          — formal structure
```

## Artifacts Directory

```
.firecrawl/research/<slug>/
├── L1/ L2/ L3/                  # from lower levels
└── L4/
    ├── academic-plan.md
    ├── crew/
    │   ├── planner-output.md
    │   ├── researcher-a-output.md
    │   ├── researcher-b-output.md
    │   ├── critic-output.md
    │   └── editor-output.md
    ├── sources/
    │   ├── academic/            # arXiv, Scholar papers
    │   │   ├── 50-<paper>.md
    │   │   └── 50-<paper>.sum.md
    │   └── secondary/
    │       ├── 70-...
    │       └── ...
    ├── timeline.md
    ├── methodology.md
    ├── open-questions.md
    ├── report.md                # 5000+ word academic report
    ├── executive-summary.md     # updated from L3
    ├── bibliography.md          # A/B/C/D rated
    └── report.pdf
```

## Stage 1: Execute L3

Invoke `expert-research` skill. Use `--go` flag semantic (no plan approval prompt — we're already committed at L4):

```
Skill: expert-research
Args: <query> (note: running as L4 foundation, skip plan approval)
```

Wait for L3 completion. Verify L3 artifacts exist before proceeding.

## Stage 1b: Exa as primary academic source (optional, v0.5.0+)

> **Biggest L4 upgrade in v0.5.0.** Previously L4 used Tavily with "academi