Problem — test/mode_block.test.ts > H4 FIRST — the session-API availability fixture > the transform hook resolves the session's agent via session.get, and session.messages' array order is ascending…(proven live) fails in CI's fast lane with a consistent ~12.2–12.3s duration (a fixed-timeout shape) on multiple UNRELATED PRs — #815, #818, #819, #827, #828 — whose diffs touch no mode_block file, while PR #824's fast run on the same base (f7c9cc4, the #814 merge) was green at 2m11s. Local runs pass on this machine (env -u OPENCODE_DB pnpm --filter amicode test: 2303/0). This reds the fast gate for every in-flight PR and blocks honest merges that weigh CI green.
Evidence trail — failed-job logs 101320734515 (#815), 101329627593 (#818), 101331411510 (#819), 101356231329 (#827 first run), 101356677833 (#828): identical test, identical duration shape. #827's re-run pending at filing.
Suspects (unverified) — the fixture is a live session-API probe (the session_api_probe fixture dir; 'proven live'): a server-boot timeout under CI load, a time-of-day or ordering dependence, or a missing CI-only prerequisite (vendored binary fetch ordering in the fast lane).
Fix — the owning surface (mode_block, #808/#814) investigates: either make the probe deterministic in the fast lane, move it to a live/env-gated lane (the telaio probe precedent), or fix the boot timeout. Until then, PR merges that weigh this red carry this issue as the recorded foreign-red pointer.
Source — discovered gating the fleet-amicissimo-forksplit campaign's PRs #824/#827 (both diffs mode_block-clean). Cross-campaign: the sibling session-lifecycle campaign recorded the same red on #815.
Problem —
test/mode_block.test.ts > H4 FIRST — the session-API availability fixture > the transform hook resolves the session's agent via session.get, and session.messages' array order is ascending…(proven live)fails in CI'sfastlane with a consistent ~12.2–12.3s duration (a fixed-timeout shape) on multiple UNRELATED PRs — #815, #818, #819, #827, #828 — whose diffs touch no mode_block file, while PR #824's fast run on the same base (f7c9cc4, the #814 merge) was green at 2m11s. Local runs pass on this machine (env -u OPENCODE_DB pnpm --filter amicode test: 2303/0). This reds the fast gate for every in-flight PR and blocks honest merges that weigh CI green.Evidence trail — failed-job logs 101320734515 (#815), 101329627593 (#818), 101331411510 (#819), 101356231329 (#827 first run), 101356677833 (#828): identical test, identical duration shape. #827's re-run pending at filing.
Suspects (unverified) — the fixture is a live session-API probe (the session_api_probe fixture dir; 'proven live'): a server-boot timeout under CI load, a time-of-day or ordering dependence, or a missing CI-only prerequisite (vendored binary fetch ordering in the fast lane).
Fix — the owning surface (mode_block, #808/#814) investigates: either make the probe deterministic in the fast lane, move it to a live/env-gated lane (the telaio probe precedent), or fix the boot timeout. Until then, PR merges that weigh this red carry this issue as the recorded foreign-red pointer.
Source — discovered gating the fleet-amicissimo-forksplit campaign's PRs #824/#827 (both diffs mode_block-clean). Cross-campaign: the sibling session-lifecycle campaign recorded the same red on #815.