academic-research

# Academic Research Capability Pack > Teaches AI agents HOW to do academic research — depth enforcement, citation integrity, quality scoring, and self-evaluation. Derived from ScienceClaw's 629-line SCIENCE.md protocol + ScholarEval rubric + VOYAGER-inspired Reflexion Cycle. > > **CONSUMES**: Research question + optional domain constraints (field, date range, databases). > **PRODUCES**: Evidence-grounded research report + methodology section + citation audit trail + Reflexion Cycle self-evaluation (for non-trivial tasks). --- ## Scope Disambiguation This pack covers **academic and scientific research methodology**. **This pack — use when:** - User asks about 学术/academic/论文/paper/文献/literature tasks - Systematic or literature reviews, meta-analyses - Tasks mentioning PRISMA, PubMed, citation analysis, ScholarEval - Evaluating research quality or scholarly writing **NOT this pack — defer to `*research` when:** - Market analysis, technology trend analysis, competitive analysis - General non-academic "research X" tasks without scientific context - Business strategy investigation, product analysis If ambiguous, ask the user: "Is this academic/scientific research or general market/technology research?" --- ## Step 1: Detect Research Task Type Classify the user's request into one of four tiers. Each tier has a mandatory minimum depth derived from ScienceClaw SCIENCE.md lines 111-121. | Tier | Signal Keywords | Min Phases | Min Tool Calls | Session Model | |------|--------...