A MinVer dev build on main computes a prerelease of its own guessed next
version (e.g. 2.9.1-alpha.0.14 after tag v2.9.0), which already sorts
above the last tag by numeric core, so the update banner correctly
stays quiet there. But the reduction to a bare numeric core meant a
genuine new release landing at exactly that guessed version (v2.9.1
real) compared equal to the still-running prerelease and the banner
never fired. A prerelease now only loses to a real release of the same
core, not to another prerelease of the same core.
Verified against the live repo: latest published tag is v2.9.0; a
Release build of this worktree (14 commits ahead) reports
AssemblyInformationalVersion 2.9.1-alpha.0.14+<sha>.
MCP_TOOL_TIMEOUT (raised by ClaudeDo's launchers) is a wall-clock cap unrelated
to Claude Code's idle-silence abort (default 300s for HTTP-transport MCP
servers), which no launcher raises. A wait near the 900s recommendation was
silently killed at ~300s in any session, launcher or not.
WaitForTaskChange now reports MCP progress every 30s while polling, which
resets that idle timer. Verified against a real claude -p call (no launcher
env overrides) surviving a 341s wait via 60s pings -- the same silence that
previously aborted at 300s. Tool description and the list-handler prompt no
longer claim ClaudeDo's launchers guarantee a long wait survives.
batch_get_tasks with includeDescription=true had no size guard: 8 tasks'
full Description/Result serialized to a single 51k-char line, blowing past
the tool-result token budget and forcing a file/PowerShell workaround to
read it back. Roadblock text was also only reachable via a second get_task
call per task, since roadblockCount (in the lean ref) has no text sibling.
- descriptionMaxChars (default 1500, was unlimited) truncates Description
and Result independently, flagged via *Truncated/*FullLength so a caller
never silently works off a cut string.
- fields narrows taskFull to just the named properties, including the new
roadblockText (the tail of Result after TaskRunner's roadblock marker) —
reachable without pulling in the rest of Result/Description.
- taskFull is now BatchTaskDetailDto, a batch-only shape decoupled from
TaskDto so get_task's own contract is untouched.
- The whole response is capped (MaxResponseChars); over that, the call
throws naming which parameter (descriptionMaxChars/fields/taskIds) to
adjust instead of shipping an oversized payload.
10 tasks with 5000-char descriptions (default settings) serialize to
~13.8k chars, comfortably under the 25k-char cap.
Omitted fields on set_task_config/set_list_config used to be indistinguishable
from an explicit clear, so any single-field update silently wiped every other
override (including SessionSkills, which wasn't even a tool parameter). Now
only the fields you pass are changed; clearing an override requires naming it
in the new clearFields array (a sentinel string doesn't work uniformly since
maxTurns is an int). set_list_config's "all four null deletes the config" case
is preserved but only reachable via clearFields. Return values now reflect the
full resulting config.
MergeHelperSelectionModalViewModel now loads ParentTaskId and orders
rows hierarchy-aware (parent immediately followed by its children,
mirroring TasksIslandViewModel.Regroup). Children render indented with
the same indent-track visual as the main task list; ticking a parent
cascades selection to its children, while a child can still be
deselected on its own. A child whose parent got filtered out (e.g.
already Done) renders top-level with an explicit hint instead of
disappearing. Confirm keeps returning parents ahead of their children.
git add -A ran against the shared target checkout, sweeping in
whatever untracked or unrelated modified files a concurrent session
left there and committing them into an unrelated merge. Stage exactly
the paths that were reported conflicted and resolved instead.
A hand-driven task produces no streamed agent output, so the Output tab
opened empty. Bind now selects git for manual/interactive rows and resets
to output for autonomous ones.
QueueServiceSlotFailureTests's throwing-slot test broke its poll loop the
instant the DB read observed Status==Failed, but TaskStateService.FailAsync
commits the status flip before calling the broadcaster's TaskUpdated, so the
assertion could race ahead of the broadcast landing in hub.Proxy.Calls
(~1-in-5 failures in isolation). Wait for both signals before breaking.
RunInSlotAsync only logged an unexpected exception, leaving a task stuck
Running in the DB forever with the UI never notified (the raw-SQL queue
claim that put it there never broadcasts). Cancellation is handled
separately and left alone, since the cancel path already wrote the
terminal status.
Every tool description now leads with what the tool does AND when to reach for
it, since MCP clients rank tools by that text. Per-parameter prose moved onto
the parameters as [Description], exhaustive result-shape enumerations and
design/history rationale dropped, and the repeated boilerplate clauses
(lean-task-ref, batch cap, refused-while-Running) pulled into McpToolDocs,
which also documents the style for future tools.
Tool-level description text: 20494 -> 13605 chars (-34%); combined with the new
parameter descriptions 18517 (-10%).
Closes gaps that caused wrong calls rather than just verbose ones:
- list_task_attachments returns metadata only, no file content
- run_task_now shares continue_task's single override slot and throws when busy
- list_runs is ordered oldest-first and feeds get_run
- workingDir on create_list/update_list is an existing local git repo path,
unvalidated until the first task run
- get_task_worktree's behind=0 also means the main ref was unreachable
Removes get_task_status_values: a whole tool entry for static reference text.
GetTask's description is now the canonical place for status meanings.
Audited all 12 prompt kinds against the code they drive. Every real defect sat on
the boundary between prompt text and the --allowedTools the launcher passes.
- Planning: "Use nothing else" after a six-tool list forbade the brainstorming
Skill the same prompt demands two paragraphs earlier. WindowsTerminalLauncher
allowlists mcp__claudedo__*,Read,Grep,Glob,WebFetch,WebSearch,Skill -- name them,
and tell the planner to ground subtasks in the repo with Read/Grep/Glob.
- System: SuggestImprovement is only allowlisted when ParentTaskId is null and
PlanningPhase is None, and TaskRunMcpService throws for any child, but this
prompt reaches every run. Planning children were told to use a tool they lack.
- MergeHelperExecute: derived "effective max-turns" from task/list/preset by hand,
which misses TaskRunner's MaxTurnsCeiling clamp. Call get_effective_run_config
instead -- built for exactly this and reports the clamp.
- MergeHelperExecute: quoted the override-slot error as the raw lowercase throw
rather than the string ExternalMcpService actually surfaces.
- Refine: listed Read/Grep/Glob unconditionally though RefinePrompt.BuildArgs only
appends them when a repo is available.
Two findings deliberately left open, both needing a code decision rather than a
prompt edit: the System prompt's worktree claim is false for a list without a
WorkingDir (task runs in a plain sandbox dir), and 'fable' is missing from both the
prompt's cost ordering and ModelRegistry.ByCostAscending.
The execute prompt told the handler to wait with timeoutSeconds up to 170 -- a
leftover from the retired MCP_TOOL_TIMEOUT=200000ms era. The real server-side
clamp is TaskWaitMcpTools.MaxTimeoutSeconds = 900 and every launcher sets
930000ms, so the handler was making ~5x the wait_for_task_change calls it needed
and burning turns on re-waiting.
It also never passed treatWaitingForChildrenAsBusy, and only waited on ids that
were Queued or Running. A task with children reports "changed" the moment it
reaches WaitingForChildren, so such a task both dropped out of the wait set and
signalled completion early -- the handler could reach review/merge while
children were still running.
RenderTemplate replaced tokens one key at a time over a StringBuilder, so a
token appearing inside an already-substituted value got substituted again on a
later pass. The prompt briefs only escaped this because their callers happen to
pass "tasks" last -- reordering the dictionary or adding a fourth token would
have started rewriting user-authored task descriptions, which after the enhance
phase carry file paths and config snippets.
Single-pass regex over the template instead; unknown tokens still pass through.
Both merge-helper ConPTY sessions loaded PromptKind.MergeHelper, so the
post-handoff session received the phase 0-2 dedupe/enhance instructions and was
told to ignore them by its brief alone.
Split into MergeHelperTriage (phases 0-2 + handoff) and MergeHelperExecute
(phases 3-5), so each session carries only its own phases. Consolidated the
generic ask-the-user rule to one place per prompt, scoped Phase 5's summary to
what the execute session actually knows, and moved the dedupe/enhance bilanz to
the triage handoff. Regression guards assert neither prompt carries the other's
phase headings and that the shared-checkout git rule stays in execute.
The ConPTY host flattens LaunchSpec.Args into one Windows command line and quotes
each token, so a list working dir stored as "C:\repo\" produced the token
"C:\repo\" -- whose trailing backslash escapes its own closing quote. Everything
after it collapsed into --add-dir's variadic list, including
--append-system-prompt-file and the positional kickoff, so "Let Claude handle it"
opened a session with no prompt at all and the CLI warned that brief.md is not a
directory. Only user-supplied working dirs can carry a trailing separator; the
session dirs the worker builds never do.
Register(taskId, cts) already refuses (and logs) a double-registration, but
both call sites discarded the bool and dispatched anyway under an
unregistered CTS. If the loser then unregistered the winner's CTS during
its own cleanup, TryCancel could silently no-op against a live process.
- OverrideSlotService.StartInSlot now fails RunNow/ContinueTask loudly
(throws) when it loses the registration race instead of registering
over — or silently proceeding despite losing to — the queue picker.
- QueueService's picker loop retries registration briefly (the loser's own
claim-attempt-then-unregister resolves fast) before dispatching; if
registration never resolves it marks the already-claimed row Failed
instead of running it unregistered.
- RunCancellationRegistry.Unregister already had compare-and-remove
semantics (TryRemove(KeyValuePair)), so a loser's cleanup could not have
removed the winner's CTS once registration correctly failed.
Added regression tests exercising the real registry through both dispatch
paths: RunNow losing the registration race throws without disturbing the
winner, the picker's retry succeeds and TryCancel reaches the live run when
the loser unregisters in time, and the picker fails the task instead of
running unregistered when it never does.
InteractiveSessionId is persisted before the ConPTY process spawns (1a988ff).
If `claude --session-id <guid>` exits immediately after launch (auth/network
hiccup, crash before the TUI starts), OnSessionProcessExited set HasExited
but never StartError, so ConPtyPaneViewModel.CanRetry (which requires
StartError) never offered Retry — every reopen just resumed the same dead
session id, permanently poisoning that task's interactive sessions.
Now a nonzero exit within 5s of the session becoming "running" is treated
as a died-at-startup failure and routed through the same StartError path
as a launch-time exception, so the pane shows the error banner + Retry.
Retry re-fetches the LaunchSpec via BuildForTaskAsync, which reuses the
same persisted InteractiveSessionId — correct for a transient failure
(fresh process, same id), but does not help a genuinely dead session id.
Clearing a dead session id server-side is a separate design question,
left out of scope here.
TasksIslandViewModel.CancelReviewAsync swallowed the HubException the
worker raises when a task's unit merge is draining (79f90a9), unlike
the details-pane version fixed in the same commit. Mirror that fix:
report the rejection via ErrorReported instead of a bare catch.
Reset & Retry discarded the branch and queued an autonomous run even while the
user had an interactive ConPTY pane open on the task, and finalizing a plan
queued every child unconditionally (the hub has no notion of a UI-hosted
session) — both bypassed the HasInteractiveSession gate added for
CanSendToQueue. CanResetAndRetry now checks it too, with a subscription on the
bound task so the command re-evaluates when the flag flips without Task
itself changing; SendToQueueAsync now blocks queuing the whole plan and
surfaces the affected child titles when any child has an open session.
mika kuns