Skip to content

feat: The grouped list_tasks answer adds open boards, unlisted boards, personal tasks and a summary EXO-94360 - #739

Merged
bmestrallet merged 2 commits into
feature/rtdfrom
feat/EXO-94360-grouped-list-tasks-summary
Nov 16, 2026
Merged

bmestrallet merged 2 commits into
feature/rtdfrom
feat/EXO-94360-grouped-list-tasks-summary

Conversation

@bmestrallet

Copy link
Copy Markdown
Contributor

Why. In evaluation runs, asked how many tasks are late across their boards, the assistant called the grouped list_tasks once, then kept paging or calling it board by board: the grouped answer said has_more true whenever more tasks matched than the page held, and gave no ready figure to answer with, so the assistant either looped or wrote its own totals.

How.

  • The grouped answer (group_by_project) also returns:
    • open_project_count: the boards the user can access that hold open tasks;
    • projects_not_listed: the boards past the 100-board cap of projects;
    • personal_count: the selected tasks on no board;
    • summary: one sentence built from those counts and naming the filters the call applied, such as « 260 overdue open tasks in 29 of 86 boards ».
  • The summary says « open », and counts the boards « of » the open boards, only when the call selects open tasks only: overdue_only, or hide_completed_tasks without completed_since.
  • The open tasks are counted once on every board the user can access, with the listing's accessibility. That count gives open_project_count and each listed board's open_count, so a grouped call still runs two grouped counts.
  • A grouped call with limit 1, made for the counts, returns has_more false and no next_offset. A grouped call reading a real page keeps both, since list_tasks also serves other agents and MCP clients that read the page and the counts in one call.
  • The list_tasks description names the new fields and tells the model to answer from them about the call's own filters, never reusing a summary for another question. Its earlier definition joins ai-tool-definition-upgrades.json, pinned in the upgrades test.

Stacked on #738. Part of EXO-94360 with a Meeds-io/ai change whose Tasks agent reads projects_not_listed: this PR lands before it or with it, never after.

Knowledge: pending — EXO-94360
Classification: N2: a changed definition of an existing read-only MCP tool on a list endpoint (description and parameter text; no require_approval or annotations change), output fields added; no REST, DAO, schema or ACL change. Its approver must be an Architect or Developer who knows it is N2.

🤖 Generated with Claude Code

@bmestrallet

Copy link
Copy Markdown
Contributor Author

Self-review close-out — 8472bab

Delivery checks before each round; reviewer round 1, then a closing round with a fresh reviewer, whose findings were fixed and confirmed clean by the same reviewer; one test gap found by the mutation run, fixed and confirmed.

# Round Sev Finding Status
1 delivery checks 🟠 The summary wrote « open » with hide_completed_tasks and completed_since, which lists completed tasks fixed
2 delivery checks 🟡 The limit-1 sentence of the description was not limited to grouped calls fixed
3 1 🟢 « one task » where an empty selection returns none fixed: « at most one task »
4 1 🟢 Javadoc wrap fixed
5 closing 🟢 Javadoc wrap fixed
6 closing 🟢 An all-personal selection read « 0 in 0 boards » fixed: « …, all personal, on no board »
7 mutation run 🟡 personal_count from the listed boards only survived: the test had fewer boards than the cap fixed: the 101-board test pins it

Verified conform.

  • personal_count = count less the tasks of every board, both distinct counts over the same conditions; open_project_count and each open_count come from one count on every board the user can access, with the listing's accessibility.
  • has_more is false only on a grouped call with limit 1; a grouped page and an ungrouped call keep has_more and next_offset.
  • The summary says « open » and « of » only when the selection is open-only, matching TaskServiceImpl#addChangeFilter and TaskQuery#setOverdueAt.
  • The shipped list_tasks digest is in ai-tool-definition-upgrades.json (23 earlier definitions, after a sweep of every branch's and pull request's history); the new one is not.
  • Targeted: TaskMcpToolTest 141/0, TaskToolDefinitionUpgradesTest 2/0.
  • Mutation-verified, 9/9 killed: has_more kept on the counting call; has_more dropped on every grouped call; the open count restricted to the listed boards; personal tasks from the listed boards; « overdue » always; « open » with completed_since; the all-personal wording removed; projects_not_listed 0; the upgrades digest removed.
  • Full suite -pl services -am verify -Prun-its -Pcoverage at 8472bab: UT 494/0, IT 13/0; « All coverage checks have been met ».

Classification (computed on origin/feature/rtd...8472bab16, this PR's own commit on top of #738): N2. A changed definition of an existing read-only MCP tool, no require_approval or annotations change; no REST, DAO, schema or ACL touched. Its approver must be an Architect or Developer who knows it is N2.

@bmestrallet
bmestrallet force-pushed the feat/EXO-94360-grouped-list-tasks-summary branch from 8472bab to 8c104a9 Compare November 15, 2026 18:13
@bmestrallet

Copy link
Copy Markdown
Contributor Author

Rebased onto #738's new head 22683ae, after #737 merged into feature/rtd 111fb84. No conflict, and the commit is identical by git range-diff: 8472bab → 8c104a9. The tree at 8c104a9 is byte-identical to the one at 8472bab, so the local runs in the close-out still apply.

bmestrallet and others added 2 commits November 16, 2026 07:24
…EXO-94377

The tool paged through every task of the project and returned each one
with its change log and comments of the period, so one answer for a large
project reached about a megabyte. It now returns at most 20 tasks in all:
the ones updated in the period, the most recently active first, then as
many older uncompleted ones as there is room for. has_more says that the
project holds more such tasks; uncompleted_count gives its uncompleted
tasks. Only the tasks returned are read in full. The period still starts
at midnight of the user's day, in the user's zone.

The description changes, so the upgrades file lists the tool's one
earlier definition, the only one any ref ever shipped.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…, personal tasks and a summary EXO-94360

- group_by_project also returns open_project_count (the boards holding
  open tasks the user can access), projects_not_listed (the boards past
  the list's cap), personal_count (the selected tasks on no board) and
  summary, one sentence naming the filters the call applied, such as
  « 260 overdue open tasks in 29 of 86 boards ». It says « open », and
  counts the boards « of » the open boards, only when the listing selects
  open tasks only: overdue_only, or hide_completed_tasks without
  completed_since.
- The open tasks are counted once on every board the user can access:
  that count gives open_project_count, and the listed boards read their
  open_count from it, instead of a second count on the listed boards.
- has_more: a grouped call with limit 1, made for the counts, returns at
  most one task, with has_more false and no next_offset. A grouped call reading
  a real page keeps has_more and next_offset, since list_tasks also serves
  agents and MCP clients that read the page and the counts in one call. A
  grouped shape without a has_more key would break those callers' paging.
- The list_tasks description names the new fields, says to answer from
  them about the call's own filters, and states the limit-1 rule for
  grouped calls. Its shipped definition joins the upgrades list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@bmestrallet
bmestrallet force-pushed the feat/EXO-94360-grouped-list-tasks-summary branch from 8c104a9 to 739444d Compare November 16, 2026 06:27
@sonarqubecloud

Copy link
Copy Markdown

@bmestrallet
bmestrallet merged commit 739444d into feature/rtd Nov 16, 2026
5 checks passed
@bmestrallet
bmestrallet deleted the feat/EXO-94360-grouped-list-tasks-summary branch November 16, 2026 10:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant