feat(demo): Project auto-escalation wiring, live-progress fixes, vision-click hardening#210
Merged
Conversation
…ling, harden vision-click screenshots
Demo-hardening from the 2026-07-02 dry-run:
- Step 10 Q11 finally wired: after a task-shaped chat reply (regex
prefilter: >=60 chars + action verb, so casual messages never pay the
classifier cost), _should_escalate_to_project runs post-reply and the
stream emits an escalate_project frame; the UI renders a suggestion
chip ("Start as Project" pre-fills Project mode — user still hits
send; "No thanks" → new POST /api/chat/escalate_silence, Q11
session silence). Emits agent_auto_escalated_from_chat audit event.
- Agent live progress: _startAgentPoller now auto-resumes after a page
reload while the bound agent is active (piggybacked on the existing
lifecycle tick, idempotent); schedulePollAgentTimeline's type filter
included only legacy types (agent_reply/agent_status) — the runner
posts agent_update/agent_blocked/agent_question/agent_done/_aborted,
so live updates were filtered out on the reply path.
- mouse_control: screenshot capture 5s→10s timeout + one retry. Stress
test (6 runs): UI-TARS coords stable within 2px, but 2/6 runs died on
screencapture timeout — the screenshot, not the model, is the flaky
part. Manifest regenerated.
Tests: 4 new (escalation prefilter/verdict/never-raises). Full suite
2440 passed (4 pre-existing pilot-e2e failures, known-issues).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Three demo-hardening fixes from the 2026-07-02 dry-run:
1. Project auto-escalation (Step 10 Q11) — implemented but never called; now wired.
Post-reply (never adds latency to the answer), gated by a cheap regex prefilter so the Qwen classifier only runs on task-shaped messages. UI shows a suggestion chip; "Start as Project" flips to Project mode with the text pre-filled (user still presses send), "No thanks" hits the new
POST /api/chat/escalate_silence(Q11 session silence). Audit:agent_auto_escalated_from_chat.2. Agent live progress actually shows.
_startAgentPoller(approve-path, correct) now also auto-resumes after page reload while the agent is active. AndschedulePollAgentTimelinefiltered for message types the runner never posts (agent_reply/agent_status) — added the real vocabulary (agent_update/agent_blocked/agent_question/agent_done/agent_aborted).3. Vision-click (demo step 7) screenshot hardening.
Stress test showed UI-TARS coordinates stable within 2px across runs — the flakiness is
screencaptureblowing the 5s cap under load (2/6 runs). Now 10s + one retry. Manifest regenerated for the skill edit.Tests
4 new escalation tests; full suite 2,440 passed.
🤖 Generated with Claude Code