Repository navigation
Conversation
A title derived from an agent's schedule_task prompt is cut at 80 UTF-16 code units, which can split a surrogate pair and leave a lone high surrogate that renders as a replacement character. Drop that dangling half after the cut, keeping the 80-unit limit, explicit titles and the "Scheduled task" fallback. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
github-actions
Bot
added
vouch:unvouched
ApprovabilityVerdict: Needs Human Approval Macroscope's review found this PR requires human approval — This prevents broken characters in automatically generated scheduled-task titles without changing task execution or explicit titles. Because it changes the default title generated when no title is supplied, human review is required. You can add or adjust custom eligibility rules. Learn more. |
|
No actionable comments were generated in the recent review.
 |
| Layer / File(s) | Summary |
|---|---|
Title derivation and scheduled-task tests apps/server/src/mcp/OrchestratorMcpService.ts, apps/server/src/mcp/OrchestratorMcpService.test.ts |
Prompt-derived titles use the trimmed first line, are limited to 80 code units, and avoid ending on a high surrogate. Explicit titles remain unchanged, and an empty derived title falls back to "Scheduled task". Tests record complete upsert inputs and check title and workspace-strategy values. |
Estimated code review effort: 2 (Simple) | ~10 minutes
Change: Bug fix
Suggested reviewers: juliusmarminge
Merge Risk: 
b1932
-
The scheduled-task title fix is mergeable with no identified correctness, security, integrity, or availability risk.
Pre-merge checks: 
-
Check name Status Explanation Title check The title clearly and concisely describes the primary fix: preventing long scheduled-task titles from ending with a broken character. Description check The description covers the problem, expected behavior, reproduction steps, implementation, scope justification, focused tests, broader verification, observed results, and known limitations. Linked Issues check Check skipped because no linked issues were found for this pull request. Out of Scope Changes check Check skipped because no linked issues were found for this pull request.
Finishing Touches
-
Generate unit tests (beta)
-
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Problem
When an agent schedules a task over MCP without a title, the server names it after the prompt's first line, cut to 80 UTF-16 code units. If an emoji or other character outside the Basic Multilingual Plane straddles unit 80, the cut keeps only its high surrogate. The saved title then ends in
�, and that character is stored, so it shows on every client that lists the task.To reproduce: have an agent call
schedule_taskwith no title and a prompt whose first line is 79 ASCII characters followed by🚀. Settings → Scheduled tasks shows the title ending in�.Change
After the 80-unit cut, drop a trailing lone high surrogate. This is the same idiom
GitVcsDriverCorealready uses for worktree directory names. The 80-unit limit, explicit titles and theScheduled taskfallback for a blank first line are unchanged. A first line that is nothing but a lone surrogate now falls back toScheduled taskrather than saving the surrogate.Scope and approval
There is no issue or discussion. This is a very small, focused fix for an obvious bug: a truncation that splits a surrogate pair and saves a broken character. It changes one expression in the MCP
scheduleTaskdefault-title path. It does not change title limits or add grapheme-aware truncation.Verification
vp test run apps/server/src/mcp/OrchestratorMcpService.test.ts -t "titles": a new table covering a long ASCII line, an emoji split by the limit, an emoji ending exactly at the limit, a blank first line and an explicit title. With the source change reverted, only "an emoji split by the limit" fails, receiving a title ending in\uD83D. With the change, all pass.vp run check:changed 1d626d02d95f9a2343553f6498037ee82290d304: format, lint, knip, typecheck and tests all pass.schedule_taskonce with no title. Its prompt's first line puts🚀at units 79–80. Each client opened the task from Settings → Scheduled tasks (the list on web and desktop, the task editor on Android and iOS), once on the base server and once on this change. A read-only SQLite check found the base stored title ending in U+FFFD and this change's title ending insummary.Web, base server: the title ends in a replacement character.
Web, this change: the title ends after "summary".
Desktop (Electron dev app), base server: the title ends in a replacement character.
Desktop, this change: the title ends after "summary".
Android emulator, task editor scrolled to the end of Name, base server: the name ends in a replacement character.
Android, this change: the name ends after "summary".
iOS simulator, task editor with the Name caret at the end, base server: the name ends in a replacement character.
iOS, this change: the name ends after "summary".
Not checked: other browsers; tunnel and relay connection modes, though the title is server data and travels unchanged; iPad layout. A cut that lands just after a space still leaves that trailing space, as it did before; this change does not alter that.
Prior PRs
Generated with Claude Opus 5.5 and GPT-6.1 Terra and GPT-6.1 Sol in T3 Code Nightly