Name: sf-datacloud-retrieve
Author: jaganpro

SKILL.md

$27

Required Context to Gather First

Ask for or infer:

target org alias

whether the user needs quick count, medium result set, large export, schema inspection, or semantic search

table/index name if known

whether the task is read-only SQL or search-index lifecycle management

Core Operating Rules

Treat Data Cloud SQL as its own query language, not SOQL.

Run the shared readiness classifier before relying on query/search surfaces: node ~/.claude/skills/sf-datacloud/scripts/diagnose-org.mjs -o <org> --phase retrieve --json.

Use describe before guessing columns.

Prefer sqlv2 or async query flows for larger result sets.

Use vector search or hybrid search only when the search index lifecycle is healthy.

Keep STDM/parquet/session-tracing workflows out of this skill family.

Recommended Workflow

1. Classify readiness for retrieve work

node ~/.claude/skills/sf-datacloud/scripts/diagnose-org.mjs -o <org> --phase retrieve --json

# optional query-plane probe, only with a real table name

node ~/.claude/skills/sf-datacloud/scripts/diagnose-org.mjs -o <org> --phase retrieve --describe-table MyDMO__dlm --json

2. Choose the smallest correct query shape

sf data360 query sql -o <org> --sql 'SELECT COUNT(*) FROM "ssot__Individual__dlm"' 2>/dev/null

sf data360 query sqlv2 -o <org> --sql 'SELECT * FROM "ssot__Individual__dlm"' 2>/dev/null

sf data360 query async-create -o <org> --sql 'SELECT * FROM "ssot__Individual__dlm"' 2>/dev/null

3. Use describe before guessing fields

sf data360 query describe -o <org> --table ssot__Individual__dlm 2>/dev/null

4. Use vector or hybrid search only when an index exists

sf data360 search-index list -o <org> 2>/dev/null

sf data360 query vector -o <org> --index Knowledge_Index --query "reset password" --limit 5 2>/dev/null

sf data360 query hybrid -o <org> --index Knowledge_Index --query "reset password" --limit 5 2>/dev/null

sf data360 query hybrid -o <org> --index Insurance_Index --query "weather damage coverage" --prefilter "Type_of_Insurance__c='Home'" --limit 10 2>/dev/null

5. Reuse curated search-index examples when creating indexes

Use the phase-owned examples instead of inventing JSON from scratch:

examples/search-indexes/vector-knowledge.json

examples/search-indexes/hybrid-structured.json

High-Signal Gotchas

Data Cloud SQL is not SOQL.

Table names should be double-quoted in SQL.

sqlv2 is better than ad hoc OFFSET paging for medium result sets.

async query is preferable for large results.

search-index operations and vector/hybrid queries depend on the index lifecycle being healthy.

Hybrid search can use --prefilter, but only on fields configured as prefilter-capable when the search index was created.

HNSW index parameters are typically read-only on create; leave userValues: [] unless the platform explicitly documents otherwise.

query describe is not a universal tenant probe; only run it with a known DMO or DLO table after broader readiness has been confirmed.

Output Format

Retrieve task: <sql / sqlv2 / async / describe / vector / search-index>

Target org: <alias>

Target object: <table or index>

Commands: <key commands run>

Verification: <query rows / schema / status>

Next step: <segment / harmonize / follow-up>