Skip to content
AssureSwarm Docs

Agent Operating Guide

/coach-query-data

Run a read-only GraphQL query against the Canvas tenant from a natural-language question and get the results back as a clean table, count, or record.

Turn a plain-language data question into a GraphQL query, run it, and present the result. The skill translates what you ask, find, list, search, count, look up, into a valid query over the tenant’s items, workflows, steps, and users, runs it, and formats the answer as a table, a count, or a single record. It is read-only: it never mutates anything, so there is no suggestion to approve and results come back directly.

When to use

Section titled “When to use”

When you want to find, list, search, count, or look up records in the tenant. To change what you find, this skill only reads: hand off to a mutate skill like /coach-item-create once you know the record you want to act on.

Inputs

Section titled “Inputs”
Flag Required Notes
--question <text> No The natural-language data question. The default input.
--query <graphql> No A pre-built GraphQL string. If supplied, the translation step is skipped and it runs directly.
--limit <n> No Page-size override.
--no-format No Return raw JSON instead of a formatted table. Use sparingly: it defeats the readability goal.

Example

Section titled “Example”

/coach-query-data --question "list open issues assigned to Maria" reads the schema if it needs to, builds an items query with a full-text search and a status filter, runs it, and returns a short markdown table of id, title, and status. Counts work the same way: “how many audits are in progress?” returns just the number.

Good to know

Section titled “Good to know”
  • Search vs. filter. A free-text search hits the full-text index over titles and searchable fields; a structured filter narrows by status, type, owner, or date. The skill combines them for precision: search a corpus, then narrow.
  • Counts come from a pre-aggregated path. For “how many issues by status,” it uses the tenant’s pre-computed stats rather than fetching every row and counting: cheaper and faster.
  • Results are paged. Lists default to 20 rows; larger sets are paged rather than over-fetched, since the platform may reject very high limits.
  • Your permissions gate the results. An empty result can mean the records exist but are outside your access, not that there are none.
Section titled “Related”