You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
bench canary: the chat door waits for a status line dev no longer spells, so every chat cell records noframe #1034
bench/canary/lib/chat.sh waits for the literal ' · idle' on the v3 status line before it types the brief, and chat_asked_frame looks for [a] accept on a card. Neither string exists on dev (f08201024) any more. Every chat cell therefore records noframe after 61 s having spent nothing, and a grid run against dev reports a wall of empty cells that look like model failures.
Compare with tmux capture-pane of a live bin/aforge chat: the idle status line no longer contains · idle.
Acceptance, end to end first
A chat cell against dev reaches the model: cell.json carries a non-zero cost and calls.jsonl has at least one priced row, on the same command above. The needles the canary waits for come out of the same table the e2e suite reads (internal/e2e/tuiwords_test.go) or a sibling that an untagged test checks against internal/tui3 sources, so the next respelling names the canary instead of silently emptying the grid.
Found by the 2026-09-14 PlanDB A/B grid; the lane patched chat.sh locally to run at all (spark:~/af-plandb-evidence/rig-local-patch.diff). Related: #1028 (fix-branch leak in fresh_clone).
Drafted with CodeAF · reviewed and owned by the author
Activity