Repository navigation
fix(dashboard): backport queued/steered text shown as typed to release/0.8.0 - #15577
Conversation
Local test evidence on the release/0.8.0 tree (head
|
First Principles Review (Fable 5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS With no shell, byte-identity to main's 1b5f87e is unverifiable here — a human should confirm the cherry-pick carries nothing beyond #15197. What this change shipsInventory (8 items) — 8 justifiedIntent: on the stable 0.8.0 channel, show a user their own mid-turn (queued/steered) message exactly as typed — a FIX, backporting merged #15197.
[FIRST-PRINCIPLES-REVIEWED] 86eac22 |
Design Review (Fable 5) — ✅ PASSDesign-level review of Design-Verdict: PASS Backend CI skips release-branch PRs, so confirm the posted local test-run comment (526 passed; 9/12 fail with src reverted) exists on the PR. [DESIGN-REVIEWED] 86eac22 |
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
Byte-identity and combined-tree checks (the review lanes asked for these; they have no shell)
Combined check across all eight 0.8.0 backports (#15574 to #15581). They octopus-merge cleanly onto release/0.8.0 @ 556bf8e (merge
|
Problem / Motivation
Goal: On a 0.8.0.x build, a message the user sends while a turn is running is shown to them exactly as they typed it, as an ordinary send already is.
This backports #15197 to
release/0.8.0; it is already merged onmain(1b5f87e). On 0.8.0, a send that is steered into the running turn or queued for the next one shows[REDACTED: suspicious URL to <host>]in place of a link the user just pasted. Cancelling a queued message refills the composer with the placeholder, so the link is lost from the user's own draft. Root cause and design are in #15197.Why it matters
0.8.0.x is on the stable channel. Anyone who pastes a link mid-turn sees their own words replaced and concludes the agent never got the link.
Not a goal
main.CHANGELOG.md.What changed (motivation → approach → change)
One commit,
git cherry-pick -x 1b5f87ed8a(Ray Xu's commit, authorship kept), in the #11339 / #11645 / #15314 backport shape. It applied without conflict, and every line matches the source.dashboard/chat_delivery.py,chat_handlers.pyandchat_runner.pycarry the entry's origin and show user-authored queued/steered text as typed;security_posture.pyandlearn-cron-dashboard.mddocument it.Dependencies: none; no other
main-only commit is needed.Backwards compatibility
Compatible: only the display of the user's own queued/steered text changes. Model-facing delivery and redaction of agent output are unchanged.
Tests
test/test_queued_user_text_display.py(new, from #15197). Run on this branch (output in the PR comment):test_queued_user_text_display.py,test_chat_runner_coverage.py,test_chat_runner_early_recovery_actor.py,test_chat_runner_folder_steering.py,test_dashboard_chat_handlers_coverage.py,test_security_posture.py).src/reverted torelease/0.8.0and the test kept: 9 of 12 fail.Backend CI (
ci.yml) does not run on release-branch PRs, so these local runs are the backend evidence.Manual verification
N/A beyond #15197: the code is byte-identical to
main's fix.Related Issues
Backport of #15197 to
release/0.8.0.Pattern harvest
Rule candidate: review-prompt
Pattern: an egress that redacts text by path instead of by author, so user-authored text gets output-side redaction on one path but not its sibling. (Harvested in #15197.)
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement