lsm / neokai / 24759292318

82%

push

github

test(daemon): switch 5-space unit test SQLite to in-memory (#1582)

* fix(space): clicking Review node on canvas now opens the node's agent session

Match activity members by m.role (agent slot name) instead of m.label against
agentDisplayNames. The old code resolved agent display names from the store and
compared them to the human-readable m.label field, which is unreliable — when
they diverged (e.g. 'Code Reviewer' vs 'reviewer') the match failed silently and
fell through to the Task Agent session.

SpaceTaskActivityMember.role is explicitly the slot name (e.g. 'reviewer',
'coder') and directly maps to the _agentSlotNames parameter, making the match
deterministic.

Adds a regression test covering the label-vs-slot-name mismatch scenario.

* test(daemon): use in-memory SQLite in NodeExecutionManager tests

The file-based makeDb() helper created a temp directory and ran all
migrations on each beforeEach call. Under CI load this exceeded the 10s
bun hook timeout, causing consistent failures in the 1-core test shard
that were unrelated to the code under test.

Switch to :memory: as recommended by CLAUDE.md for unit tests: faster,
no filesystem I/O, and no temp-dir cleanup needed.

* test(daemon): switch workflow test helpers to in-memory SQLite

workflow-executor.test.ts and completion-actions-persistence.test.ts
both used file-based makeDb() helpers with mkdirSync/runMigrations in
beforeEach. Under CI load this exceeded the 10s bun hook timeout.

Switch all three makeDb() calls to :memory: — no filesystem I/O, no
temp-dir cleanup, same migration coverage.

* test(web): fix ReadOnlyWorkflowCanvas mock onNodeClick type signature

The mock typed onNodeClick as (nodeId: string) => void but the real
component passes three args: (nodeId, nodeName, agentNames). Align the
mock type to match.

* test(daemon): switch RoomRepository test helper to in-memory SQLite

Same beforeEach hook timeout pattern as the other fixed test files:
file-based DB creation + create... (continued)

9070 of 12356 branches covered (73.41%)

Branch coverage included in aggregate %.

1 of 1 new or added line in 1 file covered. (100.0%)

66915 of 75758 relevant lines covered (88.33%)

302.31 hits per line