agnes-the-ai-analyst/cli/skills/agnes-data-querying.md

name	description
agnes-data-querying	Use when querying any data in Agnes — discovery first, estimate before fetch, materialize scoped subsets locally

Querying Agnes data

When asked about ANY data in Agnes, follow this protocol: discover → choose tool → fetch (with estimate) → query locally → clean up.

Discovery first

Before writing ANY query, understand what's available:

da catalog --json | jq <filter>     # know what's available
da schema <table>                    # learn columns + types
da describe <table> -n 5             # see real values for shape

Never write SELECT * FROM <table> blindly. For local-mode tables it's wasteful; for remote-mode tables it can blow up at 225M+ rows.

Choose the right tool

Tables in da catalog have a query_mode:

Mode	Means	How to query
local	parquet synced on laptop	da query "SELECT …" directly
remote (BigQuery)	parquet NOT on laptop	da fetch subset → snapshot, OR da query --remote one-shot

For remote tables, you MUST either:

1. da fetch a filtered subset → query the local snapshot (preferred), OR
2. da query --remote for one-shot server-side execution, OR
3. da query --register-bq for hybrid joins (rare; see docs)

The da fetch workflow (preferred for remote tables)

1. Estimate first

Always estimate before fetching:

da fetch web_sessions_example \
    --select event_date,country_code,session_id \
    --where "event_date >= DATE_SUB(CURRENT_DATE(), INTERVAL 30 DAY) 
             AND country_code = 'CZ'" \
    --estimate

Output tells you scan cost, expected rows, and local bytes — so you know if it's reasonable.

2. If reasonable, fetch to snapshot

da fetch web_sessions_example \
    --select event_date,country_code,session_id \
    --where "event_date >= DATE_SUB(CURRENT_DATE(), INTERVAL 30 DAY) 
             AND country_code = 'CZ'" \
    --as cz_recent

3. Query the local snapshot

da query "SELECT event_date, COUNT(*) FROM cz_recent GROUP BY 1 ORDER BY 1"

Heuristics for da fetch

Requirement	Why
Always --select specific columns	Avoid implicit SELECT * on remote (expensive)
Always --where for remote tables	Otherwise add --limit to keep result bounded
Always --estimate first if unsure	Partition/clustering metadata + shape matters; dry runs are free
Reuse snapshots across questions	da snapshot list before fetching — existing snapshot? Skip the fetch

BigQuery SQL flavor for --where

For source_type=bigquery (per da catalog), use BigQuery SQL syntax:

Syntax	Example
Date literal	DATE '2026-01-01' (NOT '2026-01-01'::date)
Timestamp literal	TIMESTAMP '2026-01-01 00:00:00 UTC'
Now	CURRENT_DATE(), CURRENT_TIMESTAMP()
Date arithmetic	DATE_SUB(CURRENT_DATE(), INTERVAL 30 DAY)
Regex	REGEXP_CONTAINS(col, r'pattern') (raw string!)
NULL check	col IS NOT NULL (standard)
Cast	CAST(x AS INT64) (NOT INT)

For source_type=keboola / source_type=jira (local), use DuckDB SQL in your da query calls — there's no --where on local since fetch is implicit.

Snapshot hygiene

• Reuse snapshots across questions in the same conversation
• Use descriptive names: cz_recent, orders_q1_us, sessions_today
• Drop with da snapshot drop <name> when done with a topic
• Check total cache size with da disk-info

When NOT to use da fetch

Scenario	Use instead
Single aggregate on remote table (SELECT COUNT(*))	da query --remote "SELECT COUNT(*) FROM web_sessions_example" — cheap, no fetch needed
Throwaway exploration with raw BQ syntax	da query --remote — one-shot, no snapshot
Cross-table JOIN with both remote	Use da fetch for one side + da query --remote for the other; full cross-remote JOIN needs design (see #101)

When the table you need isn't in da catalog

The catalog reads from system.duckdb::table_registry — entries land there only via admin registration, not auto-discovery. If da catalog doesn't show what the user is asking about:

1. Tell the user the table isn't registered
2. Hand off to an admin (or, if you have admin role yourself, follow the agnes-table-registration skill)
3. Don't da query --remote your way around it — the catalog gap means the registry doesn't track this dataset, RBAC can't gate it, and quotas don't apply

Protocol summary

1. Discover: da catalog, da schema, da describe
2. Check query_mode: local (direct) or remote (fetch or --remote)?
3. For remote: --estimate first, then da fetch with --select + --where
4. Snapshot name: descriptive (cz_recent), reuse across questions
5. Query: da query against snapshot; DuckDB SQL syntax
6. Cleanup: da snapshot drop when done; da disk-info to check size