with one click
duckdb-docs
// Search DuckDB and DuckLake documentation and blog posts. Returns relevant doc chunks for a question or keyword using full-text search against a locally cached index.
// Search DuckDB and DuckLake documentation and blog posts. Returns relevant doc chunks for a question or keyword using full-text search against a locally cached index.
Explore and query data on S3, Cloudflare R2, GCS, MinIO, or any S3-compatible storage. Use when the user mentions an s3://, r2://, gs://, or gcs:// URL, asks "what's in this bucket", wants to list remote files, preview remote Parquet/CSV/JSON, or query data on object storage without downloading it. Also triggers when the user wants to know the size, schema, or row count of remote datasets.
Answer questions about spatial data using DuckDB. Use when the user mentions locations, coordinates, lat/lng, distances, maps, addresses, "near", "within", "closest", geographic names, or spatial file formats (GeoJSON, Shapefile, GeoPackage, GPX, GeoParquet). Also triggers when the user wants to find places, buildings, or roads — Overture Maps provides free global data on S3 with zero API keys. Handles spatial joins, distance calculations, containment checks, density analysis, and format conversions for geographic data.
Convert any data file to another format: CSV, Parquet, JSON, Excel, GeoJSON, and more. Use when the user says "convert to parquet", "save as xlsx", "export as JSON", "make this a CSV", "turn into parquet", or any variation of format-to-format conversion for data files. Also triggers when the user wants to write Parquet, Excel, or other binary formats that Claude cannot produce natively.
Read any data file (CSV, JSON, Parquet, Avro, Excel, spatial, SQLite) or remote URL (S3, HTTPS). Use when user references a data file, asks "what's in this file", or wants to preview/profile a dataset. Not for source code.
Search past Claude Code session logs to recall prior decisions, patterns, or unresolved work. Use when user says "do you remember", "what did we do", references past conversations, or you need context from prior sessions.
Install or update DuckDB extensions. Each argument is either a plain extension name (installs from core) or name@repo (e.g. magic@community). Pass --update to update extensions instead of installing.
| name | duckdb-docs |
| description | Search DuckDB and DuckLake documentation and blog posts. Returns relevant doc chunks for a question or keyword using full-text search against a locally cached index. |
| argument-hint | <question or keyword> |
| allowed-tools | Bash |
You are helping the user find relevant DuckDB or DuckLake documentation.
Query: $@
Follow these steps in order.
command -v duckdb
If not found, delegate to /duckdb-skills:install-duckdb and then continue.
duckdb :memory: -c "INSTALL httpfs; INSTALL fts;"
If this fails, report the error and stop.
The query is: $@
There are two search indexes available:
| Index | Remote URL | Local cache filename | Versions | Use when |
|---|---|---|---|---|
| DuckDB docs + blog | https://duckdb.org/data/docs-search.duckdb | duckdb-docs.duckdb | lts, current, blog | Default — any DuckDB question |
| DuckLake docs | https://ducklake.select/data/docs-search.duckdb | ducklake-docs.duckdb | stable, preview | Query mentions DuckLake, catalogs, or DuckLake-specific features |
Both indexes share the same schema:
| Column | Type | Description |
|---|---|---|
chunk_id | VARCHAR (PK) | e.g. stable/sql/functions/numeric#absx |
page_title | VARCHAR | Page title from front matter |
section | VARCHAR | Section heading (null for page intros) |
breadcrumb | VARCHAR | e.g. SQL > Functions > Numeric |
url | VARCHAR | URL path with anchor |
version | VARCHAR | See table above |
text | TEXT | Full markdown of the chunk |
By default, search DuckDB docs and filter to version = 'lts'. Use different versions when:
current/nightly features → version = 'current'version = 'blog'version = 'stable'If the input is a natural language question (e.g. "how do I find the most frequent value"), extract the key technical terms (nouns, function names, SQL keywords) to form a compact BM25 query string. Drop stop words like "how", "do", "I", "the".
If the input is already a function name or technical term (e.g. arg_max, GROUP BY ALL), use it as-is.
Use the extracted terms as SEARCH_QUERY in the next step.
The cache lives at $HOME/.duckdb/docs/CACHE_FILENAME (where CACHE_FILENAME is duckdb-docs.duckdb or ducklake-docs.duckdb per Step 3).
First, ensure the directory exists:
mkdir -p "$HOME/.duckdb/docs"
Then check whether the cache file exists and is fresh (≤2 days old):
CACHE_FILE="$HOME/.duckdb/docs/CACHE_FILENAME"
if [ -f "$CACHE_FILE" ]; then
MTIME=$(stat -f %m "$CACHE_FILE" 2>/dev/null || stat -c %Y "$CACHE_FILE")
CACHE_AGE_DAYS=$(( ( $(date +%s) - MTIME ) / 86400 ))
else
CACHE_AGE_DAYS=999
fi
echo "Cache age: $CACHE_AGE_DAYS days"
If CACHE_AGE_DAYS ≤ 2 → skip to Step 5.
Otherwise (stale or missing) → fetch the index:
duckdb -c "
LOAD httpfs;
LOAD fts;
ATTACH 'REMOTE_URL' AS remote (READ_ONLY);
ATTACH '$HOME/.duckdb/docs/CACHE_FILENAME.tmp' AS tmp;
COPY FROM DATABASE remote TO tmp;
" && mv "$HOME/.duckdb/docs/CACHE_FILENAME.tmp" "$HOME/.duckdb/docs/CACHE_FILENAME"
Replace REMOTE_URL and CACHE_FILENAME per Step 3. If the fetch fails (network error), report the error and stop.
duckdb "$HOME/.duckdb/docs/CACHE_FILENAME" -readonly -json -c "
LOAD fts;
SELECT
chunk_id, page_title, section, breadcrumb, url, version, text,
fts_main_docs_chunks.match_bm25(chunk_id, 'SEARCH_QUERY') AS score
FROM docs_chunks
WHERE score IS NOT NULL
AND version = 'VERSION'
ORDER BY score DESC
LIMIT 8;
"
Replace CACHE_FILENAME, SEARCH_QUERY, and VERSION per Step 3. Remove the AND version = 'VERSION' line if searching across all versions.
If the user's question could benefit from both DuckDB docs and blog results, run two queries (one with version = 'stable', one with version = 'blog') or omit the version filter entirely.
httpfs or fts not found): run duckdb :memory: -c "INSTALL httpfs; INSTALL fts;" and retry.https://duckdb.org/data/docs-search.duckdb and the DuckLake index at https://ducklake.select/data/docs-search.duckdb.For each result chunk returned (ordered by score descending), format as:
### {section} — {page_title}
{url}
{text}
---
After presenting all chunks, synthesize a concise answer to the user's original question ($@) based on the retrieved documentation. If the chunks directly answer the question, lead with the answer before showing the sources.