Summary

Describe the change.

What Changed

Key files and behaviors.

Validation

How you tested.

Risks

Anything to watch.

  Open in ADE  ·  ade/chat-20260610-105420-b2e74a31 branch  ·  PR #551

Greptile Summary

This PR makes two primary changes: (1) it extracts lane-name fallback logic into a new shared module (laneNameFallback.ts) and wires both the main process and renderer to use deterministic, prompt-derived fallback names instead of timestamp-based ones; and (2) it changes the remote runtime bootstrap to use an explicit socket path (--socket $ADE_HOME/sock/ade.sock) and removes the pre-connection runtime stop call.

• Shared laneNameFallback module (apps/desktop/src/shared/laneNameFallback.ts): extracts deriveDeterministicLaneNameFromPrompt, GENERIC_LANE_FALLBACK_NAME, genericLaneFallbackName, and genericSuffixFromLaneFallbackName, resolving the duplicate-stopword-set drift risk flagged in previous reviews. Both agentChatService.ts (main) and AgentChatPane.tsx (renderer) now import from the single source of truth.
• reasoningEffort propagation: titleReasoningEffort and summaryReasoningEffort are added to ResolvedChatConfig, plumbed through runSessionIntelligenceTask, and respected in ptyService.ts for CLI session title and summary generation. The legacy ai.chat.autoTitleReasoningEffort field is migrated into sessionIntelligence.titles.reasoningEffort on load.
• Remote runtime bootstrap: stopRemoteRuntimeDaemon is removed; the RPC command now always includes --socket <channel-specific-socket> so a freshly uploaded helper CLI doesn't mismatch-recycle the desktop-owned brain.

Confidence Score: 5/5

Safe to merge; changes are well-scoped and thoroughly tested. The shared laneNameFallback module resolves the previously flagged stopword drift, and the remote bootstrap socket change has full test coverage across all channel variants.

All core paths (lane-name derivation, remote bootstrap, reasoning-effort plumbing, null-model-clear) have matching test updates. The only open question is timing of Claude's native vs. ADE-generated title for CLI sessions, but the worst outcome is a brief title flicker, not data loss or a broken flow.

apps/desktop/src/main/services/pty/ptyService.ts — the dual-title-generator path for Claude CLI sessions is new and worth watching in production.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User submits prompt] --> B[AgentChatPane: compute genericSuffix timestamp]
    B --> C{titleSettings.enabled?}
    C -- false --> D[createDeterministicAutoLaneName\nprompt + genericSuffix]
    C -- true / undefined --> E[suggestAutoLaneName\nwith 10s timeout]
    E --> F{AI model\nsucceeds?}
    F -- yes --> G[Use AI-generated name]
    F -- timeout / error --> H[onFallback warning message]
    H --> D
    D --> I[fallbackName = slug or\nparallel-task-TIMESTAMP]
    G --> J[suggestLaneName IPC call\nto main process]
    I --> J
    J --> K[agentChatService:\nsuggestLaneNameFromPrompt]
    K --> L{AI model\nsucceeds?}
    L -- yes --> M[normalizeSuggestedLaneName]
    L -- no --> N[uniquePromptFallbackLaneName\npromptFallback vs explicitFallback]
    N --> O{promptFallback ==\nGENERIC_LANE_FALLBACK_NAME?}
    O -- no --> P[Return prompt slug]
    O -- yes --> Q[genericSuffixFromLaneFallbackName\nextract timestamp from fallbackName]
    Q --> R[genericLaneFallbackName\nparallel-task-TIMESTAMP]
    M --> S[Final lane name]
    P --> S
    R --> S

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
apps/desktop/src/main/services/pty/ptyService.ts:1417-1445
**Claude sessions may see competing title writes**

The old guard `if (isClaudeTrackedCliToolType(entry.toolTypeHint)) return` was deliberately keeping ADE's prompt summariser out of the Claude path so that Claude's own `ai-title` event wins. Removing it means both ADE's `tryCliUserTitleFromWrite` (fires immediately on first prompt) and Claude's native title pipeline run concurrently. Whichever `sessionService.updateMeta` call lands last determines the displayed title, so users may see the ADE-generated title briefly replaced by Claude's title (or vice-versa if the network is slow). The test passes because it only checks that `summarizeTerminal` was called — it does not assert a stable final title after both pipelines complete. If the intent is "ADE title first, Claude title overwrites when ready", the session update path for Claude's `ai-title` event should take precedence (e.g. set `manuallyNamed: true` or a higher-priority flag when applying Claude's title), otherwise the outcome is non-deterministic depending on network timing.

### Issue 2 of 2
apps/desktop/src/renderer/components/onboarding/AiRuntimesBand.tsx:1348-1354
**Previously-cleared `null` overrides are silently dropped during the rebuild loop**

The loop `for (const [k, v] of Object.entries(aiConfig?.featureModelOverrides ?? {})) { if (typeof v === "string" && v) overrides[k] = v; }` only re-emits non-null string values from the current local config snapshot. If `aiConfig.featureModelOverrides` already contains `{ terminal_summaries: null }` from a prior clear, that entry is skipped and the outgoing update omits it. Because `updateConfig` performs a merge, the persisted `null` survives in storage and the effective config remains correct — but the in-memory `overrides` object sent to the handler no longer reflects the full intended state. This is harmless today only because the merge is additive, but any future handler that treats the payload as a replacement would silently re-enable cleared features.

_{Reviews (2): Last reviewed commit: "Address lane fallback review" | Re-trigger Greptile}