[bmestrallet]

Why

Task tool answers were far larger than an assistant needs: on the Tribe, list_projects over 427 projects returned ≈ 89k characters and list_tasks for 200 tasks ≈ 457k, so clients cut them without saying so and agents concluded on partial data. Each project carried its allowed statuses and labels (two queries per project); each task its full HTML description, full user profiles for the assignee and every coworker, and two constant strings. limit had no upper bound, and nothing let an assistant find a board by name.

How

• list_projects takes an optional name: it lists the projects whose name holds all its words, ignoring case (ProjectQuery#setNameWords, AND); only when none does and the name has several words, those holding any one of them (setKeyword, OR). Same in a run limited to a space. Items become {project_id, name, space_id, url}; get_project_by_id keeps the full project. The total is counted with the same ProjectQuery selection as the list, so a project with several managers or participants counts once and has_more is exact.
• list_tasks, list_assigned_tasks, list_unscheduled_tasks return compact items (task_id, title, link, completed, project_id, project_name, status, labels, assignee{username, display_name}, due_date, created_date, days_overdue, priority, comments_count, plus changes with changed_since/completed_since). The compact path skips HTML rendering, coworker and user-model lookups. include_change_log=true keeps the full shape; get_task_by_id stays full.
• On request (include_description: true), each compact task item also carries its description as plain text — HTML and entities removed with Jsoup, mentions and links shown as their text, one line per block, cut at 500 characters on a word or line boundary with description_truncated: true when cut — for readers that need a card's qualification lines without one get_task_by_id per task. Without it the item is unchanged and the description is not read.
• limit is capped at 100 (local MAX_LIMIT), a zero or negative one means 10. Collections carry has_more and next_offset (absent when nothing remains); count and used_limit keep their meaning.
• list_assigned_tasks takes offset. It now reads getAssignedTasksQuery + findLastUpdatedTasks instead of getUncompletedTasks, which had no offset and no order: same selection (open, assignee or coworker), now most recently active first.
• list_unscheduled_tasks reports has_more but has no offset (it scans the project's 100 most recently active open tasks).
• The constant strings list_task_comments_tool / list_allowed_project_labels leave TaskModel (so get_task_by_id too); their content moves to the tool descriptions, which are trimmed to keep the per-call prompt small.

Measured on a local instance (50 projects, 206 tasks)

Call	Before	After
Find a board: full list_projects traversal vs list_projects name=	31,637 chars	199
list_tasks, 110 tasks	259,740	45,517
list_assigned_tasks, 206 tasks	483,603	87,259
list_tasks title= (filter existed before)	4,740	878

Depends on

• Stacked on feat: The task tools list the tasks that moved since a date, and filter by assignee and status EXO-91902 #681 (EXO-91902), itself on feat: An assistant searches the tasks of a project by text EXO-91903 #680 (EXO-91903): this PR targets feat: The task tools list the tasks that moved since a date, and filter by assignee and status EXO-91902 #681's branch and is retargeted when it lands.
• Meeds-io/ai#414 must reach the same acceptance server (agent instructions switch to has_more, name, and the create-task actions get list_project_statuses).
• The changed descriptions and parameters reach an instance that already booted once feat: the tool definitions are re-read from the add-ons once, so changed tool parameters reach booted instances EXO-91672 mcp-server#63 (definitions re-import) is deployed.
• fix: a project's description and its link are no longer swapped in the AI tools' project results EXO-91949 #682 (EXO-91949) fixes the swapped description/url still visible in get_project_by_id.

Tests

mvn -o test -pl services at 6a0407a3d: 375 tests, 0 failures. They pin the compact shapes (exact JSON key sets, created_date equal to the full task's), the all-words query and its fallback (structure, and run through the engine in TestProjectDAO: count, AND, case, exclusion outside the memberships), a project counted once with several participants, the one-word name without fallback, the clamp, has_more/next_offset on full, last and empty pages, the list_assigned_tasks offset, and the opt-in description (default key set unchanged and the description never read; plain text, line per block, cut at a boundary, surrogate pairs kept whole, flag only when cut); each was mutation-checked.

Classification: N1 proposed — the space-scope allowed_args gain name (list_projects) and include_description (list_tasks), so a run limited to a space may pass a new argument (it only narrows projects already selected by the scope check), and a new DAO query condition (setNameWords). No ACL, getTaskQuery or require_approval change.
Knowledge: Meeds-io/eng-standards#668, Meeds-io/eng-standards#669, Meeds-io/eng-standards#630, Meeds-io/eng-standards#670

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
90.0% Coverage on New Code
2.4% Duplication on New Code

See analysis details on SonarQube Cloud