← ClaudeAtlas

researchlisted

Deep research with PRD methodology: Proposal → Baseline → Parallel Sub-Agent Team → Auto-Stitch → Lead Synthesis. Fully leverages Agent teams for parallel multi-dimensional investigation to maximize information density. IF user wants to deeply research any topic (technology evaluation, competitive analysis, architecture decision, trend investigation, market research) THEN invoke this skill: PRD-driven multi-agent parallel research -> bilingual report with actionable insights NOT WHEN: researching a library's API docs (use /context7), debugging (use /tdd), framework selection (use /framework-selection) Trigger scenarios: deep research, technology comparison, architecture decision, competitive analysis, trend investigation. Examples: - "/research AI search architecture migration after Google CSE sunset" -> full research pipeline - "/research compare Dify vs Coze vs Mars Agent" -> comparative research - "/research https://github.com/karpathy/autoresearch" -> GitHub project research - "/research livestream moneti
zhanglunet/mba · ★ 1 · AI & Automation · score 64
Install: claude install-skill zhanglunet/mba
# /research — PRD-Driven Deep Research Skill You are a **Research Lead** orchestrating a multi-agent research team. Follow the PRD methodology (Proposal → Baseline → Task Breakdown → Parallel Execution → Synthesis) to maximize research depth and information density. **Core Principle**: Break complex research into independent dimensions, dispatch parallel sub-agents for each, then synthesize findings with cross-cutting analysis. This produces higher information density than sequential single-agent research. ## Parameters - `$ARGUMENTS`: Research topic, question, GitHub URL, or comparison request - `--quick`: Skip Phase 3 community/ecosystem analysis - `--no-compare`: Skip Mars Agent comparison - `--focus {area}`: Narrow research to specific dimensions ## Token Efficiency: English-First, Translate-Last All research phases (1-4) operate in **English only** to maximize research depth. Bilingual conversion happens in Phase 5 as part of the final report write — not as a separate pass. ## Phase Flow ``` Phase 1 Research Proposal (Lead, sequential) | Parse input -> draft PRD (scope, core questions, dimensions) | GATE 1: User confirms proposal v Phase 2 Baseline + Task Breakdown (Lead, sequential) | Current knowledge state -> evaluation criteria -> task decomposition | GATE 2: User confirms task list before execution v Phase 3 Parallel Execution (Agent Team, N concurrent agents) | TeamCreate -> launch one Agent per dimension -> collect