Public: anyone can read this; nobody can change it.

seed-release workflow has never succeeded — releases are being hand-made #263

Closed opened this view on origin

Closed without the crew

opened this view on origin

Evidence: before 2026-09-01 this repo had ZERO releases and ZERO tags, though seed packages r405…r829 exist — so seed-release.yml (create the seed-r<N> tag + Release + attach the tarball, link the package) has failed on every firing since it shipped. It failed again on the r829 and r873 merges (runs on runs-on: isolated; Forgejo has no run-log API so the step-level cause needs the web UI’s log view). Hand-made so far: seed-r829, seed-r873 (tag + release + attachments + package link — every API call succeeds when run by hand as plat, so the break is environmental to the isolated pool: suspects are github.token scope on the raw fetch, or the pool’s egress netpol vs the public host used for the API base). Suggested next steps: add workflow_dispatch to seed-release.yml (ride it with the next mitosis change — every solo merge costs a reseed cycle), read the failing step in the Actions web UI, and if it’s the raw fetch, use the in-cluster REGISTRY host with PLAT_PAT instead of github.token. The package lane is unaffected (r873 published under open-platform/seed by reseed-run + publish_seed).

commented

Root cause found (task log read from the pod: /data/actions_log/.../28177.log.zst): jq: command not found on the isolated runner image. Step 1 completes fully — the github.token raw fetch and the in-cluster BASE derivation both work — the job dies at step 2's first jq -n. Every firing since birth died here (run 18368 fix/seed-release-jq was an earlier skirmish). Second defect: the tag fallback mints seed-r20260902045124 — manifest seedVersion is a timestamp stamp, not the reseed number; the deterministic source is genesis/upstream.json .release. Fix incoming: jq-free workflow + tag keyed to upstream.json.

commented

Fixed and proven. #266 (jq-free, tag from upstream.json .release, in-cluster API) merged; workflow_dispatch run 28190 → success, release seed-r886 created through the 201 path — the first successful firing in this workflow's life. The push path gets its proof on the next reseed merge. Residual (non-blocking): the isolated runner image ships no jq — worth adding in ci-runner someday, but seed-release no longer cares.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs