Add prompt_progress stream chunk + prefill progress on the status line - #841
jonathanKingston wants to merge 1 commit into
Conversation
E2E screenshot updatesRefreshed (1)Clearly-different shots re-rendered by the CI e2e run and committed: Ignored as render noise (2)These re-rendered below the variance threshold (font hinting / anti-alias wobble) and were not committed. To accept them anyway, add the
These PNGs live under |
438ca23 to
19c6b38
Compare
Merge conflict needs a human
|
Surfaces prompt-processing (prefill / time-to-first-token) progress on the live agent status line. Adds a `prompt_progress` StreamChunk carrying an optional 0–1 `fraction`, forwards it through the agent loop and the agent:chunk IPC, and renders "Processing prompt… N%" (or an indeterminate "Processing prompt…" when no fraction is reported) on the current step. The chunk is provider-agnostic: any provider that can report prefill progress (e.g. a future LM Studio native-API provider via its prompt_processing.progress events) lights up the indicator. Cloud OpenAI/Anthropic don't expose this, so they simply never emit the chunk. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A4AStYh3tVN4kxG9o4aoc4
6eede3c to
dbe2f90
Compare
There was a problem hiding this comment.
PR health — leaving draft
CI is CLEAN, but the PR body marks this as intentional dormant scaffolding (prompt_progress with no provider emitter yet). Not marking ready / not enabling auto-merge until a producer lands or you want the groundwork merged alone.
Sent by Cursor Automation: Check in with agent-pane prs


Recovered from the previously-unmerged branch
claude/surface-percentages-ui-ug0792(no PR had been opened), rebased onto currentmain.What
Adds a
prompt_progressstream chunk and surfaces prefill progress on the status line: aprompt_progressmember onProviderStreamChunk, forwarding through the agent run loop, a renderercase 'prompt_progress', and a testedpromptProgressLabel()helper.Notes
prompt_progressyet, so nothing renders until an emitter is added (e.g. an LM Studio prefill producer). Adopt as groundwork or close until the producer exists.StreamChunkaddition to main's refactor by placing the member onProviderStreamChunk(its correct home, matching howtext/tool_call/reasoningare forwarded).npm run typecheck(node+web) passes clean.🤖 Generated with Claude Code
Generated by Claude Code