The unit (linux) job fails identically on local/amicode and on every PR — 8 tests, all in the pasqal credential surface:
pasqal submit — validator spawn contract (169 AC1)pasqal submit — token-only persistence + device metadata (169 AC2) (×2)pasqal submit — null token → session-only (169 AC4)pasqal revalidate — freshness + keychain silent re-auth (×2)mtime staleness — hand-edited credential file (170 AC1)opencode run — unknown model exits nonzero (regression #27371)
Signature: spawned-validator and keychain paths on a headless runner. These fail on the base branch, so PRs currently merge with red unit (typecheck green is the effective gate). Either env-gate them (the telaio convention: skip loudly when the host lacks the dependency) or provision the validator in CI. Until then, e2e never runs (it waits on unit), which is the real cost — e2e coverage is silently zero.
The
unit (linux)job fails identically onlocal/amicodeand on every PR — 8 tests, all in the pasqal credential surface:pasqal submit — validator spawn contract (169 AC1)pasqal submit — token-only persistence + device metadata (169 AC2)(×2)pasqal submit — null token → session-only (169 AC4)pasqal revalidate — freshness + keychain silent re-auth(×2)mtime staleness — hand-edited credential file (170 AC1)opencode run — unknown model exits nonzero (regression #27371)Signature: spawned-validator and keychain paths on a headless runner. These fail on the base branch, so PRs currently merge with red unit (typecheck green is the effective gate). Either env-gate them (the telaio convention: skip loudly when the host lacks the dependency) or provision the validator in CI. Until then, e2e never runs (it waits on unit), which is the real cost — e2e coverage is silently zero.