Built-in prompts

Ready-made workflows that call the right Single Origin tools in the right order.

Besides tools, the MCP server provides prompts: written workflows your client can start for you. In Claude Code they appear as slash commands; type / and pick the singleorigin server's prompt, for example /mcp__singleorigin__list_recommendations. Other clients list prompts in their own menu.

A prompt is offered only when every tool it needs is served on your workspace.

PromptArgumentsWhat it does
list_recommendationsnoneCalls list_recommendations. Shows APPLIED_UNCONFIRMED ones first under Waiting on you, with what was measured, then the rest with id, title, type, status, savings and signatures. Offers the next page or a review.
review_recommendationrecommendation_idReads the recommendation and its top 3 queries by cost, checks with query_detail that the problem really appears in them, and ends with a High, Medium or Low confidence rating and caveats.
did_my_change_helprecommendation_id, job_id (both optional)With a recommendation, reads recommendation_outcome. Otherwise calls job_run_comparison with the job and the day the change took effect, asking you for it if needed. Ends on an end state such as fixed_and_verified or cant_tell_yet.
table_scan_analysisfqn; start_date, end_date (optional; default the last 30 days)Snowflake only. Resolves the table or view, pulls table_scan_cost once over the range, detects a table-to-view migration, and shows scan cost and pruning per signature by day, plus each physical table's size and cluster key. Makes no cluster-key recommendation.
parse_dag_task_from_signaturenoneExplains the signature format: `"<dag_id> &&& <task_id>

Did this page help you?