Agent UX: pm get --output-include cannot select the collections that pm writes (comments, learnings, notes, files, tests, docs) #1124

What happens

pm create and the collection subcommands write comments, learnings, notes, files, tests, and docs onto an item. But pm get <id> --output-include refuses every one of those names:

$ pm get pm-cli-website-xtsz --output-include learnings,comments
Unknown --output-include selector(s) for get: learnings, comments.
Valid selectors: assignee, author, body, collection_counts, created_at, description, id,
notes_count, priority, status, tags, tests_count, title, type, updated_at, item, item.*,
claim_state, linked.

The selector list exposes notes_count and tests_count — counts of collections whose contents cannot be selected.

Why this costs agents tokens

To read back one learning I just wrote, the options are:

• pm get <id> with no include — returns the full item including every collection, which is exactly the unbounded read the --output-include mechanism exists to avoid; or
• a separate call per collection (pm learnings <id>, pm comments <id>, …) — N round trips where one would do.

An agent that writes a comment and a learning and then wants to verify both currently pays either a full-item read or two extra invocations. On a large item the full read is the dominant cost of the whole write-verify cycle.

Expected

--output-include accepts the collection names that the tracker actually stores, so pm get <id> --output-include comments,learnings returns just those two collections. Ideally with the same item.-prefixed spelling the scalar fields already support (item.comments).

Environment

pm 2026.8.23, Linux.

Impact

Every write-then-verify loop an agent runs is more expensive than the include mechanism should allow. This is the single most common shape of pm usage in an agent session.