ci: add .solstice/job.sh — native dispatch has failed 8/8 on a missing file, and it parked the concurrency series #254
Loading…
Reference in a new issue
No description provided.
Delete branch "claude/solstice-native-job"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Why
Anima's native dispatch falls back to
bash .solstice/job.sh. This repo has no such file, so all 8 native CI dispatches to date failed instantly — and every failure fed Anima's CI-red corrective loop, which auto-parks a work item at 3 strikes. That is precisely what parked the session-concurrency series (WI 1 SessionContext / WI 2 peer dispatch / WI 3 session registry, now at fdc 3–4): a phantom verdict from a file that didn't exist, not from the code.(The Anima-side half — the server should not treat "job script absent" as a code failure — is WI-349, in progress on the claude-code lane. This PR fixes the repo side so akh-medu has an honest native lane either way.)
Scope — deliberately identical to the Actions lane
CARGO_BUILD_JOBS=2 cargo check -p seshat, default features — mirroringcheck-seshatin.forgejo/workflows/ci.yml, including its reasoning: the full workspace release build OOM-kills rustc on the Solstice runner VMs. The two lanes must stay at the same scope: a native red for work the Actions lane calls green re-opens the phantom-parking hole. Widen both together or neither.Distro protoc 3.12 is sufficient (no proto3
optionalin this repo's protos; Actions is green on the same protoc) — the header comment documents what to do if that changes.After merge
UPDATE projects SET default_runs_on='rust' WHERE name='Akh-Medu';so the orchestrator picks the proven rust image explicitly.ready,failed_dispatch_count=0) — parked on false evidence.🤖 Generated with Claude Code