リサーチ・学習のスキル

Think and work like an expert Machine Learning Engineer. Use when a task calls for Machine Learning Engineer judgment. Reasons from decision-policy framing, Google's Rules of ML, point-in-time data, and prefill/decode inference physics through GBDT/PyTorch baselines, vLLM/SGLang/KServe serving with FP8/AWQ quantization, hybrid-retrieval RAG, RAGAS and human-validated LLM judges, OpenTelemetry GenAI tracing, and post-Omnibus EU AI Act obligations while treating train-serve skew, temporal leakage, prompt injection, LLM nondeterminism, and degenerate feedback loops as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Machine Learning Researcher. Use when a task calls for Machine Learning Researcher judgment. Reasons from population risk, double descent, and inductive bias; enforces sacred test sets, hierarchical ablations, nested CV, and HELM/Dynabench-aware benchmarking; reports with NeurIPS and Pineau reproducibility checklists while treating leakage, meta-overfitting, benchmark contamination, Goodhart gaming, and seed variance as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Global Health Researcher. Use when a task calls for Global Health Researcher judgment. Reasons from burden, equity, and health-system building blocks through DHS/MICS/DHIS2/GBD triangulation, cluster and stepped-wedge designs, RE-AIM/CFIR implementation science, and CIOMS-fair partnership while treating survey weights, HMIS completeness, and GBD smoothing as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Human–Computer Interaction Researcher. Use when a task calls for Human–Computer Interaction Researcher judgment. Reasons from situated context, Fitts/GOMS/KLM, and CHI contribution types; runs contextual inquiry through LMM/CLMM analysis with SUS/NASA-TLX triangulation; uses Prolific/OSF and treats demand characteristics, novelty effects, ordinal misuse, and WEIRD samples as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Ethologist. Use when a task calls for Ethologist judgment. Reasons from Tinbergen's four questions and versioned species vs experimental ethograms; scores with BORIS (Cohen's κ per behavior), Altmann focal/scan budgets, and ARRIVE 2.0/study-plan lab reporting while treating observer expectation, field–lab arena mismatch, habituation, and pseudoreplication as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Environmental Policy Analyst. Use when a task calls for Environmental Policy Analyst judgment. Reasons from statutory authority, baseline definition, and monetization boundaries through NEPA/ESA compliance, Circular A-4 RIAs, EPA SC-GHG and benefit transfer, IAM/IPCC scenario use, and APA regulatory comment while treating discount-rate dominance, weak transfer extrapolation, IAM structural uncertainty, and baseline inflation as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Emergency Medicine Researcher. Use when a task calls for Emergency Medicine Researcher judgment. ED trial design expert for pragmatic and cluster RCTs, time-zero and immortal-time bias in acute cohorts, NIHSS/SOFA/qSOFA and ESI/CTAS triage, NEDS/NHAMCS registries, and CONSORT/SPIRIT reporting with Hawthorne and selection-bias failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Formal Methods Researcher. Use when a task calls for Formal Methods Researcher judgment. Reasons from operational semantics and temporal logics through SPIN/TLA+/PRISM, Coq/Lean/Isabelle, Z3/CVC5, refinement and separation logic, vacuity and false-positive diagnosis, and Dafny/F* versus property-based testing boundaries.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Distributed Systems Researcher. Use when a task calls for Distributed Systems Researcher judgment. Reasons from failure models, consistency contracts, and tail-latency-and-recovery performance through TLA+ model checking, Jepsen and Porcupine linearizability checking, and YCSB/DeathStarBench benchmarking with iptables and kill -9 fault injection, while treating unbounded leases and split-brain, clock skew under NTP, and GC-induced p99 spikes as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Computer Graphics Researcher. Use when a task calls for Computer Graphics Researcher judgment. Reasons from the rendering equation, Monte Carlo bias/variance, sampling theory, and BSDF energy conservation through pbrt-v4/Mitsuba 3 references, MIS/ReSTIR/path guiding, DXR 1.2 real-time stacks, 3DGS and NerfBaselines protocols, FLIP/ColorVideoVDP evaluation, OpenUSD/MaterialX/OpenPBR interchange, and ACES 2/OCIO color while treating unconverged references, unequal-time comparisons, denoiser and temporal-reuse bias, scene- vs display-referred color errors, and white-furnace energy failures as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Critical Care Researcher. Use when a task calls for Critical Care Researcher judgment. Reasons from acute physiology trajectories, modular organ dysfunction, and cluster-aware trial design through APACHE/SAPS/SOFA scoring, Berlin/Sepsis-3/KDIGO definitions, MIMIC-IV phenotyping, and target-trial emulation with clone-censor-weighting while treating immortal-time bias, cluster contamination, competing risks from early death, and sepsis-phenotype cohort inflation as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Computer Architecture Researcher. Use when a task calls for Computer Architecture Researcher judgment. Reasons from ISA semantics, AMAT/CPI, MESI coherence, and branch prediction through gem5/SPEC/MLPerf evaluation, Amdahl and roofline discipline, TPU/GPU dataflow accelerators, DVFS/EDP, and Spectre/Meltdown mitigation overhead at ISCA/MICRO/HPCA rigor.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Deep Learning Scientist. Use when a task calls for Deep Learning Scientist judgment. Reasons from CNN/Transformer inductive bias, Li et al. loss landscapes, grokking/mode connectivity, and Kaplan/Chinchilla scaling (~20 tokens/param); designs ResNet/ViT/DiT/MoE/FlashAttention stacks with FLOPs-matched ablations; trains AdamW+cosine/WSD via Megatron-FSDP/DeepSpeed; evaluates FID/MMLU-Pro/MMLU-CF with lm-eval decontamination and Pineau/NeurIPS reproducibility checklists.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Database Systems Researcher. Use when a task calls for Database Systems Researcher judgment. Reasons from storage hierarchy, concurrency semantics, query-optimization theory, and declared workload models through TPC-C/H, YCSB, and JOB benchmarks, Jepsen/Elle correctness checkers, and perf/blktrace/fio profiling while treating cardinality-estimation plan regressions, tail-latency spikes under skew, unfair fsync-disabled speedups, and benchmark-trick wins as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Comparative Medicine Researcher. Use when a task calls for Comparative Medicine Researcher judgment. Reasons from species biology, translational validity, and the 3Rs through model-validity frameworks, IACUC protocols, ARRIVE 2.0 reporting, and FELASA/AALAS health surveillance while treating substrain drift, subclinical colony infection (murine norovirus, pinworm, Mycoplasma), analgesia-pathway confounds, and unstated husbandry variables as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert AI Researcher. Use when a task calls for AI Researcher judgment. Reasons from data generating processes, inductive biases, and compute-data-algorithm trade-offs through train/val/test discipline, seed sweeps, ablation ladders, and standards like NeurIPS reproducibility checklists, model cards, and lm-eval-harness, while treating data leakage (Kapoor & Narayanan taxonomy), benchmark contamination, spurious correlations, and reward-model overoptimization as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Algorithms Researcher. Use when a task calls for Algorithms Researcher judgment. Reasons from separating problem, model, and cost model (comparison, word-RAM, arithmetic, online) through exchange/matroid greedy proofs, subproblem-DAG dynamic programming, max-flow min-cut and Goemans–Williamson primal-dual rounding, Karp–Rabin fingerprinting, competitive ratio and Yao's principle, PTAS/FPTAS (Williamson–Shmoys), and Instance Space Analysis over DIMACS10/MIPLIB 2017/SuiteSparse while treating amortized-versus-average-case conflation, unproven greedy killed by a 4-node counterexample, Monte Carlo without a false-match probability, DIMACS10 suite overfitting, and 'linear time' hiding word-size tricks over bit-length L as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial verification of research theses — cross-run/cross-thesis contradiction via the entity index, pre-mortem (Klarman/Kahneman: assume the loss already happened, reverse the path), inversion (Munger: invert, always invert), and wrong_if falsifiability review. ≈50-finding cap sorted by severity; content-derived finding IDs; incremental scoping with three backstops; lifecycle hooks at pillar-complete / thesis-complete / pre-reduction.

日本語の概要は準備中です。原文の説明を表示しています。

The research-domain clarification skill — find underspecified items in a thesis spec.md (prose wrong_if, universe rows without rationale, missing budget/expiry/pins, ambiguous pillars), ask the human structured questions WITH options (max 5 per round), encode the answers back into spec.md's Clarifications section, and re-evaluate checklists/thesis-quality.md after every write (Q32 bidirectional maintenance).

日本語の概要は準備中です。原文の説明を表示しています。

Execute research tasks — checklist soft gate, phase dispatch with explicit thesis_dir, budget enforcement (halt + approval card on overrun), skill_pin recording (version_hash content-hashed per skill directory), [x] written on completion.

日本語の概要は準備中です。原文の説明を表示しています。

Author the research plan — fundamentals first, trade ideas last; Constitution Check evaluated twice; Deviation Register with mandatory Expiry; emits the four side artifacts (brief/entities/reproduce/contracts). Technical (market_data_stage: early) theses MUST emit the entities.md bars schema before implement (Q42).

日本語の概要は準備中です。原文の説明を表示しています。

Create a research thesis — theses/{nnn}-{slug}/ via mkdir-as-CAS, pillar-prioritized spec.md, initialized thesis.md, and the thesis-quality checklist. Refuses creation while the workspace constitution is unratified (spec 046 Q83 A+).

日本語の概要は準備中です。原文の説明を表示しています。

The Auto 档 scenario — a soft-plan orchestrator for a complete fundamental equity research thesis: understand → financials → valuation → synthesize. The body is a soft plan (stages), not a rigid DAG; correctness is enforced by each referenced skill's own requires: preconditions, never by this plan's rigidity (Q2).

日本語の概要は準備中です。原文の説明を表示しています。

Manage the operational learnings lifecycle — load prior learnings to inform current work, harvest new patterns worth preserving, and keep the document tight over time. Provides a protocol for accumulating actionable patterns from practice that complement standards and defaults. Use when a workflow session completes and produced insights worth persisting, when starting a session that should benefit from prior patterns, or when the user says 'harvest learnings', 'what have we learned', 'capture this pattern', 'tighten learnings', 'compress learnings', or 'operational learnings'.

日本語の概要は準備中です。原文の説明を表示しています。