Harness APIs use different paging mechanisms internally. The CLI abstracts these behind a uniform set of flags so callers — including AI agents — don't need to know which model a given endpoint uses.
| Flag | Description |
|---|---|
--offset N |
Skip the first N items (item-level, not page-level) |
--limit N |
Return at most N items |
--all |
Fetch all pages |
--count |
Print total item count instead of items |
--offset and --limit are always item-level. If a page boundary falls in the middle of the requested window, the CLI fetches the necessary pages and slices the result transparently.
--all overrides --offset and --limit.
--count is mutually exclusive with --offset, --limit, and --all. It returns the total count for the active query, respecting any other filters (e.g. --search). The count reflects the full matching set, not a page window.
--all and --count are only shown for endpoints that support them. For endpoints where counting or full enumeration is not available, those flags are hidden.
Normal output is a flat list of items with no paging envelope. Agents can infer position without any metadata:
- Got back fewer items than requested → you've reached the end
- Got back exactly as many as requested → there may be more; try
--offset N+limit