fix(web): intercept approval text input in chat

[henrypark133]

Summary

• When a tool requires approval in the web UI, typing "yes"/"no"/"always" (or aliases like "y", "n", "approve", "deny", "ok") in the chat input now resolves the approval card directly instead of sending a regular message
• Previously, text input went through the backend message pipeline, causing duplicate approval prompts and "No pending approval for this thread" errors
• The frontend intercepts keywords in sendMessage() and routes them through sendApprovalAction() — same code path as clicking the card buttons
• Skips already-resolved cards (during the 1.5s removal animation), picks the most recent card when multiple are visible

Test plan

• 7 new Playwright E2E tests in test_tool_approval.py:
  • test_text_yes_intercepts_approval — verifies "yes" resolves card, no user bubble, input cleared
  • test_text_no_intercepts_denial — verifies "no" shows "Denied"
  • test_text_always_intercepts_always — verifies "always" shows "Always approved"
  • test_text_aliases_intercepted — tests "y", "n", "approve", "deny"
  • test_text_approval_case_insensitive — tests "Yes", "YES", "No", "ALWAYS"
  • test_normal_text_not_intercepted_with_approval_card — regular text passes through normally
  • test_text_approval_resolves_real_tool_call — full E2E with real mock LLM tool call
• All 12 approval tests pass (5 existing + 7 new)
• cargo fmt clean, cargo clippy zero warnings

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces a 'DB MIGRATION' label for automated PR categorization and implements a new frontend feature to intercept chat keywords (such as 'yes', 'no', and 'always') for resolving tool call approvals. Comprehensive E2E tests were added to verify this interception logic. Feedback was provided to improve the robustness of the approval card selection in the frontend by ensuring the logic targets the most recent unresolved card, preventing potential issues during the resolution animation phase.

[Copilot]

Pull request overview

Updates the web chat UI to treat simple approval keywords typed into the chat input as direct approval/deny actions when an approval card is pending, avoiding backend-side duplicate approval handling. Also extends E2E coverage for this behavior and tweaks GitHub label automation to ensure a migration label exists and is applied based on changed paths.

Changes:

• Intercept approval/deny keywords in sendMessage() and route them through sendApprovalAction() instead of posting a normal chat message.
• Add Playwright E2E tests covering keyword interception and passthrough behavior.
• Update PR label automation to create/apply a “DB MIGRATION” label based on migration-related file changes.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/channels/web/static/app.js	Adds frontend keyword interception for pending approval cards before normal message sending.
tests/e2e/scenarios/test_tool_approval.py	Adds new E2E tests validating text-based approval interception and non-interception for normal text.
.github/workflows/pr-label-scope.yml	Ensures “DB MIGRATION” label exists, pins labeler action, and updates permissions.
.github/scripts/create-labels.sh	Adds creation of the “DB MIGRATION” label in the label bootstrap script.
.github/labeler.yml	Adds path-based mapping for “DB MIGRATION” label and updates header comments.

Comments suppressed due to low confidence (2)

src/channels/web/static/app.js:763

• The interception logic only inspects the last .approval-card. If the last card was just resolved (it stays in the DOM for the 1.5s removal timeout) while an older card is still pending, keyword input won’t be intercepted and will instead be sent as a normal message. To match the intended behavior (“skip already-resolved cards”), walk the cards from newest→oldest and select the first card that is unresolved (and has a request id) before mapping keywords to sendApprovalAction().

    }
    if (lastAssistant) lastAssistant.setAttribute('data-streaming', 'true');

    // Mark turn as having received content so the Done safety net
    // does not trigger a spurious loadHistory() for streaming responses.

src/channels/web/static/app.js:771

• The PR description lists aliases like y, n, approve, deny, ok, but the code also intercepts additional keywords (e.g., single-letter a, reject, cancel, yes always, approve always, etc.). Intercepting a in particular is very broad and will prevent users from sending a normal one-letter message whenever an approval card is visible. Either narrow the accepted keyword set to what’s intended, or update the PR description/tests to reflect and justify the expanded behavior.

    // Accumulate chunks and debounce rendering at 50ms intervals
    _streamBuffer += data.content;
    // Force flush when buffer exceeds 10K chars to prevent memory buildup
    if (_streamBuffer.length > 10000) {
      appendToLastAssistant(_streamBuffer);
      _streamBuffer = '';

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.