Data & Analytics for Claude Code

Synthesizes BigQuery schema optimization findings from all prior phases into a prioritized implementation plan with rollback procedures, timelines, and success metrics. Use when all previous phase reports are complete. Trigger with "generate recommendations", "finalize schema optimization plan".

Selects chart types and encodings, designs EDA workflows, and audits existing visualizations for misleading charts, accessibility failures, and wrong library choices. Use when choosing how to visualize a dataset, building a dashboard spec, or catching bad charts in a notebook. Trigger with \"design this visualization\", \"audit our charts\".

Designs production RAG pipelines covering chunking strategy, embedding selection, retrieval patterns (basic, reranked, hybrid, multi-query), and evaluation metrics. Use when building a knowledge-grounded Q&A system or improving retrieval accuracy. Trigger with "design a RAG system", "help me build a knowledge base".

Designs retrieval reranking pipelines, relevance scoring systems, and learning-to-rank models with rigorous NDCG/MRR evaluation. Use when ranking quality is poor or a reranker is needed. Trigger with \"improve my search ranking\", \"design a reranking pipeline\".

Compiles specialist agent outputs into tiered analytics narratives (mini pulse, daily brief, full deep-dive) in an advisory business voice — never touches raw data. Use when generating a polished analytics report for console, email, or Slack. Trigger with \"generate analytics report\", \"compile analytics brief\".

Designs honest model evaluation frameworks — metric selection matched to business cost functions, statistical significance testing, calibration, and confusion analysis. Use when choosing eval metrics or comparing models. Trigger with \"design a model evaluation framework\", \"compare these models statistically\".

Analyzes web traffic source attribution, channel performance, AI referral trends, and redirect domain tracking to explain where visitors come from and why traffic changed. Use when diagnosing channel shifts or measuring AI-referral growth. Trigger with \"analyze traffic sources\", \"why did traffic change\".

Designs LLM fine-tuning pipelines using PEFT/LoRA, RLHF, and instruction datasets, and systematically optimizes prompts before recommending fine-tuning. Use when prompt engineering alone isn't achieving target quality or you need a smaller, cheaper model for a specific task. Trigger with \"design a fine-tuning pipeline\", \"optimize my prompts\".

Implement comprehensive data validation at database and application layers — type, range, format, referential integrity, and custom business-rule checks. Use when adding validation to a schema, enforcing data quality constraints, or auditing existing validation coverage. Trigger with \"add data validation\", \"implement validation rules\".

Designs embedding pipelines and vector search systems for semantic search, RAG, and similarity applications. Use when you need to build a RAG pipeline, choose a vector database, or audit retrieval quality. Trigger with \"design my RAG pipeline\", \"help me choose a vector database\".

Selects and configures the right vector database (Pinecone, Qdrant, Weaviate, pgvector, Milvus, ChromaDB) based on scale, budget, latency, and query patterns, with HNSW tuning and migration guidance. Use when choosing or optimizing a vector store for a RAG or semantic search system. Trigger with "which vector database should I use", "optimize my vector DB".

Adversarially verifies specialist analytics agent outputs — checks every number against raw data, flags hallucinated insights, logical contradictions, and sampling bias before the report reaches the user. Use when quality-gating an analytics pipeline before final delivery. Trigger with \"verify analytics outputs\", \"quality check the report\".

Orchestrates YouTube data extraction via Apify actors — channel metadata, video details, search results — polling until complete and saving all datasets to disk. Use when collecting raw YouTube data for strategy analysis. Trigger with "scrape youtube channels", "fetch youtube data".

Analyst Ray — Health Data Analyst

Post-hoc Analyzer Agent

Experiment Runner Agent

用于在特定学术数据库执行文献检索的Agent。通过浏览器自动化访问数据库检索页面。

用于在特定学术数据库执行文献检索的Agent。通过浏览器自动化访问数据库检索页面。

🔭 露娜 — Research Agent

你是团队中的投资分析专家。你的职责是进行市场分析、数据解读和投资策略建议。

| 属性 | 值 | |------|-----| | 名称 | Verify Agent | | 类型 | Task Agent | | 用途 | 元数据验证 |

| 属性 | 值 | |------|-----| | 名称 | Verify Agent | | 类型 | Task Agent | | 用途 | 元数据验证 |

Hello Researcher Agent

职责：负责把 Sorftime MCP 原始返回整理成结构化中间数据，不直接写结论。

Use this agent when you need to perform specialized task X with deep analysis

Insight Writer Agent（v2）

职责：基于已经准备好的结构化数据，输出 Markdown 正式报告与 HTML 精简报告。确保每份报告达到「充电宝 v2」级别的分析深度。

Bottom-up cluster-finder for the learnbiostats vault. Reads the evergreen/ knowledge graph and proposes structure notes when a real cluster has formed — a minimum of 4 related evergreen notes before a cluster is "ready," and flags over-large clusters (>12 notes) for splitting. Mirrors the reference vault's Emergence agent: structure is discovered from the notes, never imposed a priori. Use when the writer asks "find clusters", "what structure notes are ready?", "is this idea solid enough to writ

Lab and project mentor for the learnbiostats studio. Designs and reviews the experiential projects and capstones (`projects/`, `projects/capstone/`), scaffolds and runs analysis code against real genomics/phenotype data, and enforces the two disciplines that keep an ML genetics result honest: cross-validation (no leakage across folds, families, or environments) and predict-then-check (state the expected result before running). Grounded in the design-of-experiments and code-data-analysis-scaffold

Stage 1 ingestion worker for the conference-scheduling pipeline. Parses a conference program (any format) into normalized event records with per-field confidence and independent axes, flags thin abstracts, fans out the high-volume web enrichment to conf-enrichment-researcher sub-workers, merges the enrichment back, and writes a single events.json plus a human index. Reusable across conferences — takes the source and output paths as explicit inputs and hardcodes no conference specifics. Returns t

Stage 2 clustering worker for the conference-scheduling pipeline. Turns event records into a navigable theme map — 6-8 coarse themes with sub-clusters, an explicit outlier bucket, and soft top-k affinities for every event — using the embed-then-label pipeline (resources/cluster.py) when the libraries are present and an LLM-reasoned hierarchical fallback when they are not. Embeddings group; the cartographer writes the human-readable labels. Reusable across conferences; hardcodes no conference spe

An orchestrating agent that collaboratively helps engineers build graph-based retrieval-augmented generation systems. Guides users through knowledge graph construction, embedding strategy design, retrieval orchestration, system integration, and evaluation. Use when user mentions knowledge graph, GraphRAG, graph retrieval, entity extraction for RAG, Neo4j with LLM, or building graph-augmented AI systems.

Weekly Substack performance analyzer for substacker. Pulls stats directly from the Substack dashboard via Claude-in-Chrome browser automation (primary) or ingests a manual CSV export (fallback). Computes rolling 4-week baseline, attributes over/under-performance in plain English with calibrated confidence, tracks per-section metrics once sections exist, produces a 400-800 word weekly report. Feeds Curator (section pruning) and Growth Strategist (quarterly rollups). Use Monday mornings. Trigger k

Generates the weekly static HTML dashboard for the household finance JSON store. Reads the canonical data files, applies cognitive-design principles for visual hierarchy and visual-storytelling-design for narrative annotations, runs cognitive-fallacies-guard to prevent visual misleads, and produces a single self-contained HTML file with six panels (net worth over time, cash flow, recurring & subscriptions, goals, asset allocation, vigilance feed). Output is a static file — no server, no live dat

Household spending analyst that answers, every month, where the money went, what changed versus prior months and same-month-last-year, what is recurring, what is an outlier, and what is coming up. Produces the Spending section of the monthly briefing, runs a quarterly subscription audit, projects 60-day daily cash flow per account, and maintains a 12-month forward seasonal calendar. Use as the third phase of a per-drop or monthly pipeline, when the user asks for a spending review, or when sizing

Search worker for one keyword query across bioRxiv, medRxiv, PubMed, and arXiv. Spawned by an orchestrator that's running a fan-out (one coach per expanded query). Receives a date window, a single query, and an arXiv category list; fetches all four sources in parallel via the fetch-* skills; dedupes within the four sources for this one query (DOI first, then normalized title + first-author surname); returns the deduped paper-records list as a JSON array directly in the response. Writes no files.

Evaluates Yahoo Fantasy Baseball trade offers. Runs advocate (Acceptor) + critic (Rejecter) variants, computes per-category deltas for both sides across all 10 categories (R/HR/RBI/SB/OBP, K/ERA/WHIP/QS/SV), factors in regression, positional scarcity, and playoff schedule. Verdict is ACCEPT, COUNTER (with specific counter), or REJECT. Use when a trade offer arrives, evaluating potential trade, or assessing a 2-for-1 or consolidation move.

Scans Yahoo Fantasy Baseball free agents for weekly waiver claims (FAAB league, $100 season budget). Fires in advocate (Buy) and critic (Pass) variants, synthesizes, and produces ranked ADD + BID $X recommendations with drop candidates. Use for weekly waiver priority review, FAAB bid sizing, prospect call-ups, closer-committee speculation, or injury replacement.

Per-paper structured-extraction worker for ONE paper at ONE pass depth. Spawned by an orchestrator that's running per-paper fan-out (one extractor invocation per paper). Receives a single paper record, an output directory, a week tag, and the pass to run (1, 2, or 3). Reads the paper at the requested depth (abstract for Pass 1; full PDF via Bash + curl + Read for Pass 2 and Pass 3), invokes the methodology skills in the right order, and writes the combined output to a markdown file. Pass 1 creat