Conversation
Co-authored-by: Charles Huang <77707222+GeeseGoo@users.noreply.github.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using high effort and found 2 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 7549ca4. Configure here.
| from sqlalchemy.dialects import postgresql | ||
|
|
||
| revision: str = "task_browse_summary_001" | ||
| down_revision: Union[str, Sequence[str], None] = "verdict_state_001" |
There was a problem hiding this comment.
Stale alembic head assertion
High Severity
test_verdict_state_migration_is_the_only_head still expects verdict_state_001 as the sole Alembic head, but restoring task_browse_summary_001 chains onto that revision and becomes the real head. The assertion is now false and will fail anywhere this test runs.
Reviewed by Cursor Bugbot for commit 7549ca4. Configure here.
| out["steps"] = kept | ||
| return out | ||
|
|
||
|
|
There was a problem hiding this comment.
Clip before inert-step drop
Medium Severity
Overflow retries in generate still clip by raw step count, while drop_inert_steps runs only later inside _fmt_trajectory. On long runs where the kept head and tail are mostly contentless, the retry prompt can collapse to little more than the omission marker after the contentful middle was discarded.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 7549ca4. Configure here.
There was a problem hiding this comment.
Risk: high. Left a non-blocking comment — Cursor Bugbot finished as skipping with 2 unresolved findings (including a high-severity Alembic head assertion), and this staging→main promotion exceeds the medium approval threshold. Assigned reviewers for human review.
Sent by Cursor Approval Agent: Pull Request Router and Approver
|
/promote |
|
Promoted. main now points at |




Caution
DO NOT USE THE MERGE BUTTON ON THIS PULL REQUEST.
THE BUTTON CREATES A NEW COMMIT AND BREAKS THE RELEASE MODEL.
COMMENT
/promoteTO COMPLETE THE PROMOTION.Summary
Ships trajectory-taxonomy improvements (grouped label vocabulary,
writing_report, contentless-step filtering before summarisation), live transcripts on public experiment share links, and a task-browser perf revert plus the migration-revision repair that followed it.Commits in this promotion
Note:
aa54cd62(#1152) is reverted by241419bc(#1156) within this range, andbd139668(#1157) restores the task-browse alembic revision the revert dropped. Net effect of that trio is no task-browser perf change, with migration state consistent.Validation
mainis a clean ancestor ofstaging(0 commits onmainnot instaging), so this fast-forwards.Staging Deploy on the pinned target
7549ca42must be green before/promote.Note
Medium Risk
Public live endpoint expands unauthenticated surface area but stays token-scoped; trajectory prompt/taxonomy changes affect all new summaries; migration runs DDL and heavy backfill on deploy.
Overview
This promotion bundles trajectory-analysis improvements, public live viewing on share links, and migration housekeeping after a reverted task-browser perf change.
Trajectory summarization restructures component labels into explore vs implement groups with per-label definitions in the prompt (
writing_plan/plan_correctionmove to explore; newwriting_reporton implement). The summarizer strips contentless steps before building the model prompt (step IDs and validation still use the full trajectory). Frontend types, labels, and--tc-writing-reportstyling support the new kind.Public sharing adds
GET /public/experiments/{token}/trials/{id}/live(core + Next.js proxy) and documents it.trial-detail-panelshows the Live tab for running/retrying trials without requiring the analysis view—relevant for public share pages.Database:
task_browse_summary_001migration addstask_version_browse_summaries, indexes, and backfill SQL—restoring revision consistency after the task-browser perf revert (#1156/#1157); net perf behavior from that experiment is unchanged per the promotion notes.Reviewed by Cursor Bugbot for commit 7549ca4. Bugbot is set up for automated code reviews on this repo. Configure here.