fix(frontend): request_input renders its form again after the transcript replay - #5859
Merged
Conversation
…ipt replay
The client-tool branch of the durable-record replay adapter was never
written ('v1 scope: HITL approvals only'), so a parked request_input
replayed as a bare tool part with no data-render sibling; the dispatcher
matched neither axis and fell through to UnhandledClientTool, which
auto-settled {status: not_handled}. The LIVE render was fine - the
watermark reset at turn end made hydration adopt the server transcript
seconds later, which is why the form flipped to 'Not handled' after
parking.
- transcriptToMessages (oss + agenta-chat parity copies): replay a parked
client_tool interaction as the tool part plus its render hint.
- registry: request_input resolves by tool NAME as the second dispatch
axis (same as request_connection), covering transcripts persisted
before the hint existed.
- agentApprovalResume: request_input joins CLIENT_TOOL_NAMES.
Proven against the real 17 records of the failing session: renderKind
undefined -> elicitation, and each half of the fix independently restores
dispatch. 158/244/215/338 tests pass across the four packages; tsc,
eslint, prettier clean.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
dosubot
Bot
added
size:L
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Repository YAML (base), Organization UI (inherited) Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (10)
📝 WalkthroughSummary by CodeRabbit
WalkthroughParked ChangesClient-tool replay and resume
Estimated code review effort: 3 (Moderate) | ~25 minutes Sequence Diagram(s)sequenceDiagram
participant DurableTranscript
participant transcriptToMessages
participant replayClientTool
participant MessageParts
DurableTranscript->>transcriptToMessages: client_tool interaction_request
transcriptToMessages->>replayClientTool: client-tool payload
replayClientTool->>MessageParts: update or synthesize tool part
replayClientTool->>MessageParts: append data-render when hinted
Possibly related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches 💡 1🛠️ Fix failing CI checks 💡
📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
Railway Preview Environment
Updated at 2026-08-10T10:52:50.373Z |
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
1 participant
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.
Mahmoud's regression: an agent calls
request_inputwith a schema, and instead of the question form the chat shows "Not handled by this client."Root cause — not a broken renderer, a missing replay branch
The live render worked. The durable-record replay adapter (
transcriptToMessages) had never implemented the client-tool branch (// v1 scope: HITL approvals only), so when the record watermark reset at turn end and hydration adopted the server transcript, the parkedrequest_inputcame back as a bare tool part without itsdata-renderhint. The dispatcher matched neither axis and fell through to the auto-settling "not handled" fallback. That is why the form appeared to work, then flipped seconds after the turn parked.The fix
request_inputresolves by tool name as the second dispatch axis (the same axisrequest_connectionalready uses) — covers transcripts persisted before hints existed.request_inputjoinsCLIENT_TOOL_NAMESin the approval-resume path.No parser change needed —
parseElicitationPayloadalready handles theoneOfoptions andx-ag-stepper; pinned with the verbatim payload from the failing session.Verification
Replayed the REAL 17 records of the failing session through the actual code: before →
renderKind: undefined(unhandled), after →elicitation+ElicitationWidget. Each fix half independently restores dispatch. 158 + 244 + 215 + 338 tests pass across the four touched packages; new tests pin the replay, both dispatch axes, and the verbatim stepper payload.Live repro
New chat on a build-kit agent → ask it to build a PR reviewer → the stepper form must appear AND still be a form 5-10s after the turn parks (the adoption tick that used to break it).