feat(desktop): add permission mode cycling via shift+tab

[github-actions]

Summary

Auto-generated PR for branch feature/add-permission-mode-cycling-vi

Changes

• feat: add permission mode cycling with e2e automation support
• Merge remote-tracking branch 'origin/main' into feature/add-permission-mode-cycling-vi
• Merge branch 'main' into feature/add-permission-mode-cycling-vi
• feat(desktop): add permission mode cycling via shift+tab

---

This PR was automatically created. Please update the description with relevant details.

---

Summary by cubic

Add permission mode cycling (Plan → Auto‑Accept → Ask) via Shift+Tab and a clickable badge on agent nodes. Modes persist per agent or globally and update Claude Code CLI flags.

• New Features

  • Shift+Tab cycles permission mode globally; ignores when typing or when ActionPill is expanded. Chat input also supports Shift+Tab.
  • Permission mode badge: shows on hover in the Agent Node header; click or press Enter/Space to cycle.
  • New PermissionModeStore with global default, per‑agent overrides, and localStorage persistence.
  • ClaudeCodeAdapter maps modes to CLI:
    • plan → --allowedTools "Read,Glob,Grep,LSP,WebFetch,WebSearch"
    • auto-accept → --dangerously-skip-permissions
    • ask → no flag
      AgentService passes mode to start/resume commands and supports restartSession to apply changes.
  • ClaudeCodeAgent maps UI modes to SDK values (plan/acceptEdits/default).
  • E2E support via useExpose:
    • canvas: node info and actions
    • agent node and chat: permissionMode get/set/cycle, restart session, view switching, state reads.

• Bug Fixes

  • Preserve permission-mode when resuming sessions and in fallback paths.
  • Show all SDK permission requests in ActionPill; filter PreToolUse to AskUserQuestion to avoid duplicates.

^{Written for commit d43709d. Summary will update on new commits.}

[cubic-dev-ai]

4 issues found across 18 files

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them.


<file name="apps/desktop/src/renderer/features/action-pill/ActionPill.tsx">

<violation number="1" location="apps/desktop/src/renderer/features/action-pill/ActionPill.tsx:46">
P2: Clear the global `__actionPillExpanded` flag on unmount to avoid leaving stale state if the ActionPill is removed while expanded.</violation>
</file>

<file name="apps/desktop/src/renderer/services/impl/AgentServiceImpl.ts">

<violation number="1" location="apps/desktop/src/renderer/services/impl/AgentServiceImpl.ts:527">
P2: Magic number 500ms delay should be a named constant. Consider using a descriptive constant like `TERMINAL_READY_DELAY_MS` and document why this specific value was chosen. Better yet, consider using a more robust mechanism to detect terminal readiness instead of an arbitrary timeout.</violation>
</file>

<file name="apps/desktop/src/main/services/coding-agent/query-executor/SdkQueryExecutor.ts">

<violation number="1" location="apps/desktop/src/main/services/coding-agent/query-executor/SdkQueryExecutor.ts:60">
P2: Debug `console.log` should be removed or replaced with the project's logging framework (`getLogServer().log()`). Raw console statements bypass the structured logging infrastructure documented in AGENTS.md.</violation>
</file>

<file name="apps/desktop/src/renderer/nodes/AgentNode/AgentNodePresentation.tsx">

<violation number="1" location="apps/desktop/src/renderer/nodes/AgentNode/AgentNodePresentation.tsx:113">
P2: Missing immediate state sync when `data.agentId` changes. The effect subscribes to mode changes but doesn't initialize the state for the new agent. If `agentId` changes, the permission mode UI could display stale data until the next mode change triggers a subscription callback. Add an immediate state update at the start of the effect.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}