Skip to content

ericrisco/rsc-harness / skills/duckdb

duckdb

Flagged

Use when analytical SQL must run in-process with no server: Parquet/CSV/JSON/Arrow queried in place, OLAP embedded in an app or notebook, a slow pandas groupby on multi-GB data, or S3/lakehouse data read without downloading. NOT a multi-user analytics server (that is clickhouse-analytics), NOT an app's transactional CRUD store (that is postgresdb).

FileKindSize
SKILL.mdskill md10.5 KB
evals/README.mdmarkdown2.0 KB
evals/cases.yamlconfig4.0 KB
references/python-and-interop.mdmarkdown4.0 KB
references/remote-and-lakehouse.mdmarkdown3.1 KB
scripts/verify.shexecutablescript2.6 KB

evals/cases.yaml

View on GitHub
skill: duckdb

should_trigger:
  - prompt: "I have a directory full of monthly parquet files and I want to sum revenue by region across all of them. What's the fastest way without setting up infrastructure?"
    why: "Core case: analytical aggregate over a folder of Parquet files in place, explicitly 'without infrastructure' — DuckDB's killer read_parquet-glob feature."
  - prompt: "My pandas groupby on a 5GB dataframe takes forever and sometimes runs out of memory. Is there a faster way to do this aggregation?"
    why: "Replacing a slow/OOM pandas groupby/merge with a vectorized out-of-core engine is a named when-to-use; the user never says DuckDB."
  - prompt: "Necesito analizar un CSV gigante de varios GB en mi portátil sin levantar un servidor de base de datos. ¿Qué uso?"
    why: "Spanish phrasing for ad-hoc analytics on a large CSV on a single machine with no server — exactly DuckDB's niche."
  - prompt: "I want to run SQL directly over a dataframe inside my Jupyter notebook without exporting it first. Possible?"
    why: "Embedded analytics in a notebook via replacement scan over an in-scope frame — a core embedded-OLAP use case."
  - prompt: "Can I query these parquet files straight from our S3 bucket and aggregate them, without downloading them to disk first?"
    why: "Non-obvious: most people download first; DuckDB reads s3:// in place via httpfs with row-group pushdown. Tests intent over keywords."
  - prompt: "I need an out-of-core aggregate that spills to disk because the join result won't fit in RAM, but I don't want to stand up Spark or a cluster."
    why: "Non-obvious single-node out-of-core spilling workload that explicitly rejects distributed infra — DuckDB scales up, not out."

should_not_trigger:
  - prompt: "Build a production analytics API that serves dashboards to hundreds of concurrent users with sub-second queries all day."
    route_to: "clickhouse-analytics"
    why: "Many concurrent users + always-on production serving is the single-writer/single-process boundary DuckDB explicitly cedes to ClickHouse."
  - prompt: "Set up the database for my web app to store users, orders, and handle lots of small writes with foreign keys and a connection pool."
    route_to: "postgresdb"
    why: "Transactional OLTP app database with many small writes and FKs — row-store territory, not columnar OLAP."
  - prompt: "What's the correct syntax for a SQL window function using ROW_NUMBER with PARTITION BY and ORDER BY?"
    route_to: "sql"
    why: "Engine-agnostic SQL language question, not an engine choice; belongs to the generic sql skill."
  - prompt: "I need to embed a small transactional database inside my mobile/edge app for local reads and writes that syncs later."
    route_to: "sqlite-turso"
    why: "Embedded single-file TRANSACTIONAL store at the edge — same niche as DuckDB but opposite (OLTP) workload."
  - prompt: "Build a similarity search over our document embeddings to find the nearest vectors for a query."
    route_to: "vector-db"
    why: "Vector/embedding search as the core workflow is vector-db's job even though DuckDB ships a VSS extension."

capability:
  - scenario: "User has a folder of monthly Parquet files at `sales/2025-*.parquet`, each with columns region, amount, ts. They want to aggregate total revenue by region across ALL files and write the result to `summary.parquet`, with no infrastructure."
    must_include:
      - "Reads the files with a glob in place — read_parquet('sales/2025-*.parquet') or FROM 'sales/2025-*.parquet' — no intermediate pandas/Polars load step"
      - "Aggregates with GROUP BY (ideally GROUP BY ALL) and an aggregate function like sum(amount) AS revenue"
      - "Exports with COPY (query) TO 'summary.parquet' (FORMAT parquet) rather than writing a CSV or pickling a dataframe"
      - "Uses DuckDB (pip install duckdb / duckdb.sql or the CLI) and does not stand up a server or a cluster"
      - "Optionally notes only the needed columns are scanned (columnar pushdown) and/or pins the version, without over-engineering"
deprecation-and-migrationaddyosmani/agent-skillsManages deprecation and migration. Use when removing old systems, APIs, or features. Use when migrating users from one implementation to another. Use when migrating a database schema in production, such as renaming or dropping a column without downtime (expand/contract). Use when deciding whether to maintain or sunset existing code.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
performance-optimizationaddyosmani/agent-skillsOptimizes application performance across frontend, backend, queries, and databases. Use when performance requirements exist, when you suspect performance regressions, when Core Web Vitals or load times need improvement, when N+1 query patterns need fixing, or when profiling reveals bottlenecks.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
agent-cost-reportthedotmack/claude-memBelievable agent cost report for any period, default the last 7 full days PT, not counting today. Measured tokens from Claude Code transcripts priced at OpenRouter list prices (ESTIMATED), measured provider spend when a sanctioned source exists, note-taker cost separate, Timing-style HTML/PDF plus report.json, line-items.csv, evidence.json.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
agent-cost-reportthedotmack/claude-memBelievable agent cost report for any period, default the last 7 full days PT, not counting today. Measured tokens from Claude Code transcripts priced at OpenRouter list prices (ESTIMATED), measured provider spend when a sanctioned source exists, note-taker cost separate, Timing-style HTML/PDF plus report.json, line-items.csv, evidence.json.Stars: 118kUpdated Claude CodeCodexVerifiedInstall