Flip the in-code default OpenCode tier from zen to go in both launchers:
cc-ci-plan/launch-upgrader.py: UPGRADER_TIER default + _TIER_CFG fallback → go
cc-ci-plan/launch-report.py: REPORT_TIER default + _TIER_MODEL fallback → go
nix/modules/cc-ci.nix: refresh the stale 2026-08-16 tinfoil/zen comment block to document the current go + glm-5.3-flash config
Why
The ZEN endpoint (opencode/glm-5.2) is erroring server-side (UnknownError: Unexpected server error — probed 2026-09-18 and again 2026-09-21), while the GO tier works. The weekly report agent for the completed 2026-09-18 run died on ZEN: it was launched from inside the resumed run's tmux chain where LOOP_TIER=go did not survive, fell back to the in-code zen default, produced zero output, and the watchdog's one resume hit the same error. No report has been published for week 2026-09-18.
upgrader.env already sets LOOP_TIER=go for the systemd-fired runs — but the Sep 18 incident shows env can get lost in the tmux→agent→launch-report chain, and the in-code default is the last line of defense. The supervisor launcher already defaults to the go tier.
Evidence
With a deliberately empty env, both launchers now resolve: Tier: go · Model: opencode-go/glm-5.2.
opencode/glm-5.2 (zen) probe → UnknownError (2026-09-18 07:09 and 2026-09-21 15:58); opencode-go/glm-5.2 (go) probe → OK; the resumed 2026-09-18 run completed its remaining 5 recipes green on the go tier.
python3 -m py_compile clean on both launchers.
Deployment note: the host units exec these scripts straight from /srv/cc-ci (→ this worktree), so merging + pulling is the whole deploy — no rebuild required (the nix change is comment-only).
## What
Flip the in-code default OpenCode tier from `zen` to `go` in both launchers:
- `cc-ci-plan/launch-upgrader.py`: `UPGRADER_TIER` default + `_TIER_CFG` fallback → `go`
- `cc-ci-plan/launch-report.py`: `REPORT_TIER` default + `_TIER_MODEL` fallback → `go`
- `nix/modules/cc-ci.nix`: refresh the stale 2026-08-16 tinfoil/zen comment block to document the current go + glm-5.3-flash config
## Why
The ZEN endpoint (`opencode/glm-5.2`) is erroring server-side (`UnknownError: Unexpected server error` — probed 2026-09-18 and again 2026-09-21), while the GO tier works. The weekly **report** agent for the completed 2026-09-18 run died on ZEN: it was launched from inside the resumed run's tmux chain where `LOOP_TIER=go` did not survive, fell back to the in-code `zen` default, produced zero output, and the watchdog's one resume hit the same error. No report has been published for week 2026-09-18.
`upgrader.env` already sets `LOOP_TIER=go` for the systemd-fired runs — but the Sep 18 incident shows env can get lost in the tmux→agent→launch-report chain, and the in-code default is the last line of defense. The supervisor launcher already defaults to the go tier.
## Evidence
- With a deliberately empty env, both launchers now resolve: `Tier: go · Model: opencode-go/glm-5.2`.
- `opencode/glm-5.2` (zen) probe → `UnknownError` (2026-09-18 07:09 and 2026-09-21 15:58); `opencode-go/glm-5.2` (go) probe → `OK`; the resumed 2026-09-18 run completed its remaining 5 recipes green on the go tier.
- `python3 -m py_compile` clean on both launchers.
Deployment note: the host units exec these scripts straight from `/srv/cc-ci` (→ this worktree), so merging + pulling is the whole deploy — no rebuild required (the nix change is comment-only).
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
What
Flip the in-code default OpenCode tier from
zentogoin both launchers:cc-ci-plan/launch-upgrader.py:UPGRADER_TIERdefault +_TIER_CFGfallback →gocc-ci-plan/launch-report.py:REPORT_TIERdefault +_TIER_MODELfallback →gonix/modules/cc-ci.nix: refresh the stale 2026-08-16 tinfoil/zen comment block to document the current go + glm-5.3-flash configWhy
The ZEN endpoint (
opencode/glm-5.2) is erroring server-side (UnknownError: Unexpected server error— probed 2026-09-18 and again 2026-09-21), while the GO tier works. The weekly report agent for the completed 2026-09-18 run died on ZEN: it was launched from inside the resumed run's tmux chain whereLOOP_TIER=godid not survive, fell back to the in-codezendefault, produced zero output, and the watchdog's one resume hit the same error. No report has been published for week 2026-09-18.upgrader.envalready setsLOOP_TIER=gofor the systemd-fired runs — but the Sep 18 incident shows env can get lost in the tmux→agent→launch-report chain, and the in-code default is the last line of defense. The supervisor launcher already defaults to the go tier.Evidence
Tier: go · Model: opencode-go/glm-5.2.opencode/glm-5.2(zen) probe →UnknownError(2026-09-18 07:09 and 2026-09-21 15:58);opencode-go/glm-5.2(go) probe →OK; the resumed 2026-09-18 run completed its remaining 5 recipes green on the go tier.python3 -m py_compileclean on both launchers.Deployment note: the host units exec these scripts straight from
/srv/cc-ci(→ this worktree), so merging + pulling is the whole deploy — no rebuild required (the nix change is comment-only).