🤖 fix: refine hyper density work bundling#3411
Conversation
|
@codex review |
|
Codex Review: Didn't find any major issues. Already looking forward to the next diff. ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
ThomasK33
force-pushed
the
density-grouping-v7vb
branch
from
d8236f6 to
f308716
Compare
|
@codex review Updated after simplify/deslop cleanup: removed the now-redundant bundleable-tool status switch and inlined the all-tool-row behavior. |
|
Codex Review: Didn't find any major issues. Nice work! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
ThomasK33
force-pushed
the
density-grouping-v7vb
branch
from
f308716 to
19b1172
Compare
|
@codex review Rebased onto latest origin/main and force-pushed. |
|
Codex Review: Didn't find any major issues. Keep them coming! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
ThomasK33
force-pushed
the
density-grouping-v7vb
branch
from
19b1172 to
4302859
Compare
|
@codex review Updated the hyper-density behavior so the outer |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4302859271
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
ThomasK33
force-pushed
the
density-grouping-v7vb
branch
from
4302859 to
d970739
Compare
|
@codex review Addressed the side-question answer case: assistant side answers no longer start or finalize work bundles, and collapsed work bundles keep side-answer rows visible alongside steering user rows. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d970739e0f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Hyper transcript density now treats failed, interrupted, redacted, and partial tool rows as bundleable so noisy validation failures collapse under the same operational summaries. Assistant text rows continue to split operation bundles inside an expanded work bundle so narrative checkpoints remain visible. --- _Generated with [`mux`](https://github.com/coder/mux) • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$13.89`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=13.89 -->
ThomasK33
force-pushed
the
density-grouping-v7vb
branch
from
d970739 to
343e150
Compare
|
@codex review Addressed the interrupted-turn boundary case: work bundles now only span steering/side conversation rows when the following agent work remains in the same assistant history, so a new user prompt after an interrupted turn starts a separate bundle. |
|
Codex Review: Didn't find any major issues. Another round soon, please! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Place collapsed hyper-density work bundles immediately after the user message that triggered the turn while keeping in-turn steering messages visible below the collapsed bundle and chronological inside expanded bundles. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$43.21`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=43.21 -->
ThomasK33
changed the title
|
@codex review Updated the hyper-density layout per Chromatic feedback: the work bundle is now anchored directly after the triggering user message, collapsed steering messages stay visible below it, and expanded state keeps chronological order. |
|
Codex Review: Didn't find any major issues. Keep it up! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Keep the final assistant summary visible below collapsed hyper-density work bundles while preserving chronological rendering inside expanded bundles. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$61.40`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=61.40 -->
|
@codex review Follow-up after Chromatic feedback: collapsed hyper-density work bundles now keep the final assistant summary visible below |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c1283212b0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Restrict cross-user work-bundle spanning to partial tool rows so interrupted reasoning/text-only turns cannot absorb the next prompt's work. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$61.40`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=61.40 -->
|
@codex review Addressed the interrupted partial-turn boundary feedback by only allowing cross-user work-bundle spanning after partial tool rows, and added a regression test for partial reasoning/text interruptions before a new prompt. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 811c8e7964
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Reset cross-user work-bundle eligibility on every agent row so a partial tool cannot carry the steering gate through later partial assistant text into the next prompt. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$61.40`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=61.40 -->
|
@codex review Addressed the follow-up boundary case by resetting cross-user work-bundle eligibility on each agent row, so a partial tool cannot carry the steering gate through later partial assistant text into a new prompt. Added a regression test for mixed partial tool + text interruptions. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 00de008164
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Only record a final assistant row after an operational row has been seen so pre-tool assistant text cannot become an empty work bundle. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$61.40`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=61.40 -->
|
@codex review Addressed the pre-operation finalization feedback by only recording a final assistant row after an operational row has been seen. Added a regression test for assistant text followed by a tool with no trailing assistant final. |
|
Codex Review: Didn't find any major issues. Can't wait for the next one! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Always render the hyper-density work header for active and settled work, use a Working label for active work, and remove inner work-bundle indentation so nested content aligns with the header divider. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$74.28`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=74.28 -->
|
@codex review Follow-up layout update: hyper-density work bundles now always show the Work/Working header with divider, active work uses |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5d2867906b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Show elapsed duration in active hyper-density work headers and keep post-prompt work boundaries from using pre-prompt partial operations. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$81.55`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=81.55 -->
|
@codex review Follow-up: active work headers now show elapsed duration ( |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f12fc1af0a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Only cross user-message boundaries for active in-turn work checkpoints, keep completed partial tools from merging into the next prompt, and update density fixtures to model steering through an ask-user checkpoint. --- _Generated with `mux` • Model: `openai:gpt-5.5` • Thinking: `xhigh` • Cost: `$81.55`_ <!-- mux-attribution: model=openai:gpt-5.5 thinking=xhigh costs=81.55 -->
|
@codex review Addressed the completed-partial-tool boundary feedback by only allowing cross-user bundling for active in-turn checkpoints. Added a regression test for completed partial tools followed by a tool-using next prompt, and updated the steering fixture to model an ask-user checkpoint. |
|
Codex Review: Didn't find any major issues. 🚀 ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
1 participant
Summary
Hyper transcript density now collapses failed, interrupted, redacted, and partial tool rows into operational bundles, anchors the outer
Worked for ...bundle directly below the user message that triggered the turn, and consistently shows a work header/divider for both active and settled work. Active work now shows elapsed duration asWorking for Ns.... Collapsed mode hides detailed agent-emitted work, keeps in-turn steering user messages visible below the collapsed bundle, and keeps the final assistant summary visible below the fold; expanded mode preserves chronological order without extra right indentation inside the work bundle.Background
The previous hyper-density behavior hid successful tool noise but left failed shell commands and other non-success tool rows visible. The first revision also placed the
Worked for ...fold too low in steering-message scenarios, which made the collapsed hierarchy feel disconnected from the user prompt that started the turn. Follow-up Chromatic review showed the final assistant summary should remain visible below collapsed work, and the work header should act as a consistent divider in both active and settled states.Implementation
Working for Ns...work bundle so the header/divider is always present while the agent is working.Validation
bun test src/browser/utils/messages/transcriptRenderProjection.test.ts src/browser/features/Messages/WorkBundleMessage.test.tsxTEST_INTEGRATION=1 bun x jest tests/ui/chat/transcriptDensity.test.ts --runInBand --silentmake typecheckmake static-checkWorked for ...after the triggering user messageWorking for Ns...with the same divider and flat inner layoutRisks
Moderate UI regression risk, limited to hyper transcript density rendering. The main behavior change is intentional: detailed agent-emitted rows are hidden until the containing work/operation bundle is expanded, while user-authored steering messages and the final assistant summary stay visible.
Generated with
mux• Model:openai:gpt-5.5• Thinking:xhigh• Cost:$81.55