Repository navigation
fix(mobile): show a clear message when a new session cannot start - #8124
Conversation
A new-session create rejection showed the server's message verbatim, so an internal configuration failure could reach the composer. The create failure copy is now classified by the tRPC error code: known user-actionable refusals (billing, access, repository problems) keep the server reason, everything else shows localized generic copy that asks the reader to try again later. The inline composer error, the hook's toast fallback, and the clone toast follow the same rule.
iscekic
added
drain
|
bot: End-to-end proof on Android (emulator, local dev stack). Drove the real new-session composer: typed a prompt, picked a GitHub repository, tapped "Start session" against a local backend whose create path fails with an internal configuration error. The composer shows the generic localized line under "Failed to create session" — the raw internal server string never appears. A second Start attempt shows the same copy and the form recovers for editing.
|
![kilo-code-bot[bot]](https://avatars.githubusercontent.com/in/2193792?s=60&v=4){kind=small}
Code Review SummaryStatus: No Issues Found | Recommendation: Merge Executive SummaryThe residual delta maps a prepare-time Assumptions: Files Reviewed (2 files)
Note: the previous Previous Review Summaries (4 snapshots, latest commit 28f2154)Current summary above is authoritative. Previous snapshots are kept for context only. Previous review (commit 28f2154)Status: 1 Issue Found | Recommendation: Address before merge Executive SummaryThe new Assumptions: Fix these issues in Kilo Cloud Overview
Issue Details (click to expand)WARNING
Files Reviewed (3 files)
Previous review (commit 29342de)Status: No Issues Found | Recommendation: Merge Executive SummaryThe incremental change adds the model-unavailable refusal as its own case, matching the exact server-authored message ( Files Reviewed (2 files, incremental)
Previous review (commit 2d2768f)Status: 1 Issue Found | Recommendation: Address before merge Executive SummaryThe incremental change fixes the strict-mode type failure in the prepare-failure classifier, but the previously reported regression remains: an actionable model-unavailability Overview
Issue Details (click to expand)WARNING
Files Reviewed (1 file, incremental)
Fix these issues in Kilo Cloud Previous review (commit a7f2710)Status: 2 Issues Found | Recommendation: Address before merge Executive SummaryThe new create-failure classifier indexes a const object with a Overview
Issue Details (click to expand)CRITICAL
WARNING
Files Reviewed (92 files)
Reviewed by deepseek-flash-latest · Input: 83.2K · Output: 8.3K · Cached: 2M Review guidance: REVIEW.md from base branch |
The readonly list satisfies the code contract while staying indexable, which the previous const-table form was not under strict tsgo checks.
Model retirement between pick and create rides a generic request-level code; matching its stable server-authored constant keeps that actionable reason visible while every other internal request-level string stays behind the generic line.
iscekic
left a comment
iscekic
left a comment
There was a problem hiding this comment.
bot: Fixed in 29342de. The model-unavailable refusal is now surfaced as its own case: its message is a fixed server-authored constant (MODEL_UNAVAILABLE_MESSAGE), so the classifier matches that exact string and shows it even though the code is a generic request-level one. Every other internal request-level string (including repository-authorization and mode-slug failures) stays behind the generic line.
…error-copy # Conflicts: # apps/mobile/src/components/agents/new-session-screen-body.tsx # apps/mobile/src/components/agents/use-new-session-creator.test.ts # apps/mobile/src/i18n/locales/en.json # apps/mobile/src/lib/app-actions/prepare-agent-session.ts
iscekic
removed
the
merge:human
|
bot: End-to-end proof on Android (emulator, local dev stack, owned fake account). Opened the new-session composer with a seeded GitHub repository and a default environment profile, removed that profile row while the composer held the selection, then tapped Start session. The composer shows the localized missing-selection message under "Failed to create session" and does not echo the server's profile detail.
|
iscekic
added
the
merge:bot
Change
A new-session create resolves the requested environment profile before it registers anything, so a
NOT_FOUNDrejection fromprepareSessionmeans the profile the composer picked no longer exists. The composer showed the generic server-failure line for that case; it now shows the localized removed-or-unavailable line, so the reader refreshes the selection instead of retrying a request that cannot succeed.apps/mobile/src/components/agents/cloud-prepare-error-display.ts:cloudPrepareFailureMessagemaps aNOT_FOUNDrejection (the tRPC client shape and the server-shaped form) toqueryError.notFoundDescription, ahead of the generic fallback. Every other rejection keeps its current copy, and no raw server string reaches the UI.apps/mobile/src/components/agents/cloud-prepare-error-display.test.ts: covers a missing profile record and the server-shaped form, and asserts the profile id is not echoed.The branch merged main first: main carries the shared copy helper for a failed create, so the diff here is only the missing-selection copy on top of it.
Verification
Android emulator (
omp36), local dev stack, owned fake account:Drain default.Failed to create sessionwithThis item may have been removed or is no longer available.— and not the server's profile detail.pnpm --filter kilo-app exec vitest run src/components/agents/cloud-prepare-error-display.test.ts src/components/agents/use-new-session-creator.test.ts src/components/agents/new-session-cloud-create-error.test.ts src/i18n/catalog-parity.test.ts→ 159 passed.pnpm --filter kilo-app exec tsc --noEmit→ clean; oxlint on the two changed files → 0 findings.Not covered: no simulator run on iOS. The change is shared JavaScript copy selection with no platform branch.
Proxy answers
The owner's proxy gave these answers in the owner's place.
Question: Code set: Which tRPC error codes count as user-actionable (server message shown verbatim), with everything else getting generic copy? The credential-proxy failure arrives as BAD_REQUEST.
Answer: Narrow, no BAD_REQUEST (Recommended)
Confidence: 0.70 · Precedent: Fix the root cause and the class of the issue: the internal config strings arrive as BAD_REQUEST, so showing that code verbatim would leak them again (session 7f3470e0ba6f, "fix class of issue preventing progress")
Question: Toast path: Should the toast fallback path (when no form owns the failure feedback) apply the same rule?
Answer: Yes, same rule (Recommended)
Confidence: 0.85 · Precedent: Fix the class of the issue on every affected surface (session 75f8833fe8c1, "fix this for all shared services model pickers on all surfaces")
Question: Generic copy: What should the generic copy say? (Shown under 'Failed to create session' for internal failures.)
Answer: Try again later (Recommended)
Confidence: 0.70 · Precedent: none
Question: Other callers: Scope: apply the classifier to the new-session composer only, or also to the Continue clone and OS action path that share prepareAgentSession?
Answer: All creators
Confidence: 0.60 · Precedent: Fix the root cause and the class of the issue on every affected surface (session 75f8833fe8c1, "fix this for all shared services model pickers on all surfaces (web, app)")
Question: Clone copy: The Continue clone (use-continue-cloud-create) has its own error UI with its own copy. Apply the same classification there with its own localized headline + generic key?
Answer: Same rule, new key (Recommended)
Confidence: 0.80 · Precedent: Fix the root cause and the class on every affected surface: session 75f8833fe8c1, "fix this for all shared services model pickers on all surfaces"
Question: Internal retry: For internal-kind failures, should the Retry control behavior change?
Answer: Keep existing retryability (Recommended)
Confidence: 0.75 · Precedent: Do the least work possible with no drop in quality: session 7f3470e0ba6f, "do the least amount of work possible, with no drop in quality"
Question: main already landed the same class of fix in commit a6f0b73 (fix(mobile): show localized error details on session start and comment edit #8090): it adds the shared prepare-failure copy helper, routes the composer, the shared create core and the toast fallback through it, and bans the raw server string. This branch's earlier classifier is fully subsumed. How should this pull request end?
Answer: Reduce it to a small residual delta
Confidence: 0.55 · Precedent: Autonomy: "if you saw a gap, fix it"; speed: "do not redo work that was already done". Keep this pull request for the missing profile or repository record on top of the shared helper, drop the subsumed classifier, keep the no-raw-string design, prove it on Android, and update the description to match the reduced diff.
Question: If this pull request ends up closed, should the remaining gap be filed as a separate work item?
Answer: File a separate work item
Confidence: 0.60 · Precedent: The gap should go through triage rather than be dropped. With the delta kept here, no separate item is needed.