Skip to content

The pane's Areas section shows only busy lanes and waits for a busy Fleet - #9

Merged
marvin-bitterlich merged 1 commit into
mainfrom
mod-areas-timeouts
Nov 6, 2026
Merged

marvin-bitterlich merged 1 commit into
mainfrom
mod-areas-timeouts

Conversation

@marvin-bitterlich

Copy link
Copy Markdown
Contributor

Why?

Marvin keeps the Work Board pane beside his Claude sessions to see goals, Fleet work and areas. Two parts of it did not tell the truth.

The Areas section listed about 60 "areas" for Platform. These included .pnpm-store, .worktrees, .work, .claude and every package, and almost all were "free". It looked as if every folder was an area, and the few busy lanes were hard to find. The section must show at a glance which lanes are busy and what waits on them.

The pane also said that Fleet on port 3457 "didn't answer as expected" while Fleet was healthy and only busy. The Board gave Fleet 5 seconds per read. A busy Fleet often needs more time than that.

How?

Areas. The list came from Fleet. Fleet passed every repository folder that the Board's areaFolders found to its area lanes, so each folder became an idle lane. Now daemon.fleetItems lists only the areas that work holds, queues or reserves. The plugin also drops free entries (from an older Fleet), puts held areas first, and says "No area is held or queued." when no lane is busy. The areaFolders and repositoryRoot exports on the Board's board integration object had no other user, so this change removes them.

Timeouts. I measured against a copy of Platform's Board and Fleet, on other ports and another database. The copy's Fleet had a busy queue and the Board files changed at the same time:

Read Idle Busy
daemon.fleetItems under 0.2 s 7 to 15 s
daemon.briefProblems — 5 to 9 s
Board overview with the old 5 s wait 0.2 to 0.4 s 5.1 s, Fleet "didn't answer" every time

The Board now waits up to 20 seconds for each Fleet read. The Fleet view makes two reads, and Claude Code waits only 30 seconds for each plugin request. So the two reads share one 25-second budget, and the Board always answers in time. A Fleet that does not answer within that time is reported as "didn't answer within 20 seconds". The old "didn't answer as expected" message now shows only for a bad answer.

On the plugin side, Claude Code stops waiting for a plugin request after 30 seconds. The pane reported that case as "the Work Board isn't running". Now it says that the Board is busy, and the next refresh asks again.

Decisions

  • Free areas are omitted, not counted. A count of idle folders gives no information. With no folder list, Fleet also stops reading the repository on each pane refresh. So git-ignored folders are never listed. The Board still accepts a brief that names an ignored folder that exists, because a per-brief git check-ignore on each Board scan would cost too much.
  • 20 seconds per read, 25 seconds for the Fleet view. The values come from the measurements above: the slowest single read was about 15 seconds. The view must stay inside Claude Code's fixed 30-second plugin fetch.

Callouts

  • The plugin finds a Claude Code fetch timeout by the text "no complete answer within …ms" in the host's error. If Claude Code changes that text, the pane again says that a busy Board isn't running. src/claude-code.d.ts documents no typed timeout error to use instead.

…leet

Fleet's daemon.fleetItems lists only the areas that work holds, queues or
reserves, instead of every repository folder. The Work Board pane shows those
areas with held ones first and says when no area is busy.

Board waits up to 20 seconds for each Fleet read, and keeps its whole Fleet
view within 25 seconds so it answers inside Claude Code's 30-second plugin
fetch. A Fleet that does not answer in time is named as such, and the pane
calls a Board that the host stopped waiting for busy instead of stopped.
@marvin-bitterlich
marvin-bitterlich added this pull request to the merge queue Nov 6, 2026
Merged via the queue into main with commit b2cf57b Nov 6, 2026
16 checks passed
@marvin-bitterlich
marvin-bitterlich deleted the mod-areas-timeouts branch November 6, 2026 22:26
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