Squashed history of PR #4892 (pr-4784-fix). Makes spaces the primary
navigation and working surface, on the "associations bind at creation
time only" model.
Model
- Issue <-> space is the only enforced ownership (per-space numbering).
Parent/child and project<->space associations only seed defaults at
creation; cross-space/child and project-association validations are
removed.
- Moving an issue renumbers it and records the old identifier in
issue_identifier_alias; API/CLI lookups and GitHub branch/PR
auto-linking fall back to the alias, so old identifiers resolve forever.
- Membership drives only the sidebar and personal defaults — never
access. Anyone can configure any space's member set wholesale
(PUT /api/spaces/{id}/members); saving an empty set archives the
space behind a confirm.
- Per-user space order (workspace_space_member.sort_order, fractional):
drag-sorted sidebar, "my first space" is the personal issue-creation
default; the workspace default space backs headless creation
(agents/CLI/Slack) and system placement.
Surfaces
- Sidebar: joined-spaces section (drag reorder, row -> space page,
per-group persisted collapse), Workspace group with a More menu,
Settings demoted to a footer icon.
- /space/:key/{issues,projects,autopilots,settings} — space surfaces
reuse shared page components; a routed /space/new create page
(replacing the earlier create-space modal), reserved key "NEW" so it
can never collide with a real space's /space/:key detail page.
- Issue detail moves to /issue/:id (identifier-first, Linear-style; old
/issues/:id redirects); create dialogs lead with a required space pill.
- Agent runtime brief now carries Space context (id/key/name) through
the daemon claim -> TaskContextForEnv -> prompt pipeline, with a
"## Space Context" section and --space on issue create/update in both
brief renderers, matching Project's existing treatment.
- zh-Hans: Space translated to 空间 across locales and conventions.zh.mdx.
Fixes along the way
- Cache membership judgment gains the space dimension + space_changed
WS flag.
- Silent skip on default-space lookup during invite acceptance is now a
hard failure (no space-less members).
- Backfilled space_id into ~28 raw-SQL Go test fixtures across
internal/handler and cmd/server that predated migration 132's NOT
NULL cutover.
- Fixed a resolve-loop bug in the IssueDetail identifier wrapper (mount/
unmount cycle on resolution failure) and gave it its own loading
skeleton instead of a blank screen while resolving.
- reserved-slugs generator's stale hardcoded doc-comment example synced
back to /create-space.
Verification
- go build ./..., go vet ./..., go test ./... all clean.
- pnpm typecheck (core/views/web/desktop) clean.
- packages/views: 162 files / 1656 tests passing.
- pnpm generate:reserved-slugs produces no diff.
Follow-ups tracked in docs/follow-ups/space-rollout.md.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Re-introduces the `involves_user_id` filter on the issues list / open-list /
count / grouped paths, but with the semantics nailed down for the second time
around: tab 3 surfaces issues whose assignee is an *indirect* extension of the
user (owned agent, or a squad they're a human member of / lead via owned agent
/ have an owned agent inside) — and explicitly NOT direct member assignment,
which is tab 1's meaning.
- server/pkg/db/queries/issue.sql: 4-branch filter on ListIssues /
ListOpenIssues / CountIssues. Each subquery clamps workspace_id because
issue.assignee_id is polymorphic with no FK. Leader resolution reads
squad.leader_id directly, not the squad_member copy row (squad.go ignores
errors when seeding that copy, so it can be missing). FindActiveDuplicateIssue
switched from positional $2/$3/$4 to named sqlc.arg() — pure hygiene so the
generated struct field names don't drift when new nargs are added.
- server/internal/handler/issue.go: parse involves_user_id and plumb it into
the three sqlc params; ListGroupedIssues (hand-written dynamic SQL) gets a
mirrored 4-branch fragment, no shortcut.
- packages/core: ListIssuesParams / ListGroupedIssuesParams / MyIssuesFilter /
api.listIssues / api.listGroupedIssues all carry the new param through.
- packages/views/my-issues: tab 3 switches from client-side agent-fanout to
involves_user_id=user.id. agentListOptions import and the myAgentIds memo
go away.
- server/internal/handler/issue_involves_test.go: 13 integration tests cover
every branch (positive + cross-workspace negatives) plus the critical
ExcludesDirectMemberAssignee negative on BOTH the sqlc and the grouped paths,
locking tab 3 ∩ tab 1 = ∅.
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* feat(my-issues): cover squad assignees via involves_user_id (MUL-2364)
The "My Agents" tab on /my-issues only resolved agents owned by the
caller, so issues assigned to squads (member, leader, or agent-member of
mine) never surfaced. This added a UNION-based involves_user_id filter
that the backend expands to "me + agents I own + squads I relate to" in
a single query.
- SQL: ListIssues / ListOpenIssues / CountIssues accept narg
involves_user_id and OR a workspace-scoped 3-branch UNION on the
squad assignee subquery. Leader is sourced from canonical
squad.leader_id (not the best-effort squad_member copy row whose
AddSquadMember error is dropped in squad.go:177-188 and :259-263).
- Handler: parses involves_user_id via parseUUIDOrBadRequest, plumbs
into all three list params, and mirrors the same UNION fragment into
the grouped dynamic SQL path.
- Frontend: ListIssuesParams / ListGroupedIssuesParams / MyIssuesFilter
gain involves_user_id; api client forwards it to the querystring.
- My Issues page: "agents" scope now passes involves_user_id instead of
fanning out owned-agent IDs client-side. Tab label widens to
"我的智能体 / 小队" / "My Agents / Squads".
- Tests: Go suite covers all three squad relations including the
canonical-leader-without-squad_member-copy variant, cross-workspace
isolation for agent / leader / squad_member branches, combination
with creator_id, and the malformed-UUID 400 path. Client test pins
the involves_user_id querystring wiring for both list endpoints.
The FindActiveDuplicateIssue query gets explicit sqlc.arg() names so
sqlc regeneration keeps the existing struct field names regardless of
the local sqlc version (no behavior change).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* test(my-issues): tighten cross-workspace negatives for involves_user_id UNION
Cross-workspace negative tests previously put both the foreign actor and the
foreign issue in the foreign workspace, so the outer i.workspace_id = $1
already excluded the row before the UNION branches were exercised. Stripping
a.workspace_id = $1 / s.workspace_id = $1 from any of the UNION subqueries
would not have failed the tests.
Rewrite the three existing negative cases to seed the issue in
testWorkspaceID with a polymorphic assignee_id pointing at a foreign-workspace
agent or squad (issue.assignee_id has no FK per migrations/001_init.up.sql:61).
Now each UNION branch must enforce its own workspace scoping for the issue to
stay out of the result.
Also add ExcludesOtherWorkspaceSquadAgentMember: the squad_member.agent UNION
branch had only positive coverage; this test pins that s.workspace_id = $1
and a.workspace_id = $1 must both hold there too.
Verified by mutation: stripping the workspace clause from each branch makes
the corresponding test fail.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
