GO subscription hit a monthly usage limit (resets in 9 days). Add LOOP_TIER
env (default 'zen') to launch-upgrader.py + launch-report.py so the opencode
backend can use either subscription: zen→opencode/glm-5.2, go→opencode-go/glm-5.2.
The tier selects the default model, the usage-limit probe endpoint, and the
auth.json key. The systemd timer reads LOOP_TIER from /srv/cc-ci/upgrader.env.
Mirrors cc-ci-loops.service for the SEPARATE p-lichen-orchestrator (lichen.page testing). On boot,
resumes its Remote Control session + watchdog + pipeline via engine/agents.py up. Added after the
2026-07-08 reboot (Hetzner rollback of a bad staging nixos-rebuild) left this orchestrator down while
cc-ci-loops auto-recovered. Points at the /home path for now; re-point to /srv at that migration.
Verified: diff-closures shows only this unit added; sshd/dhcpcd/tailscaled/network unchanged.
A run can finish the recipe work + summary + report yet never print a clean
sign-off marker (it may wedge afterwards, as on 2026-07-03). Without a second
completion signal the supervisor would try to 'resume' the already-finished run
every hour until the 96h window closed. Now the gate also treats the run as done
when a weekly summary file (upgrade-all-<date>.md, the Step-5 output) exists with
an mtime after the session began.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WxbpH3DquKzoSTSwGvGuET
The 2026-07-03 finish run wedged because _completed() returned True while the run
was still mid-work — so the watchdog exited early and nothing recovered the wedge.
Cause: it scanned part.get('text') across ALL message parts, so DONE_MARKER inside
a TOOL part (a subagent `task` prompt / bash command that referenced 'print
UPGRADE RUN COMPLETE') matched. Now: require the marker in the LAST assistant
TEXT (prose) message — the genuine sign-off — ignoring tool-call args and any
mid-run echo of the instruction (work after the echo disqualifies it).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WxbpH3DquKzoSTSwGvGuET
More fixes from live-running the finish of the 2026-07-03 run:
- _session_idle_min(): measure staleness across the whole session TREE (top-level
run + all descendant subagents), via the opencode server time.updated, NOT the
tmux log mtime (which freezes when a headless run doesn't stream to the pane).
A per-recipe subagent deploy runs 20-40min during which the PARENT session's
updated time is stale — reading the parent alone looked 'idle' and would false-
resume, killing the productive run. Renamed from _log_idle_min (kept as alias).
- watchdog(): wrap each poll in try/except so a transient server blip/race can
never kill the watchdog (a dead watchdog silently abandons the run).
- watchdog + supervisor now read the tree-aware session idle instead of log mtime.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WxbpH3DquKzoSTSwGvGuET
Live-testing the resume path surfaced two gaps: (1) an `opencode run` proc
EXITS when the model ends its turn, so a long /upgrade-all run's process dies
repeatedly before the whole run completes — and the log mtime freezes on death,
so the watchdog's log-idle>15min signal is both too slow and unreliable. (2) A
resumed run had no watchdog, so nothing re-continued it.
- watchdog(): detect PROC-DEATH (no live `opencode run` proc for the session +
not completed) and resume promptly, in addition to log-idle. Guarded by
MAX_RESUMES (default 20) so a no-progress loop (e.g. disk-full) eventually hands
off to the supervisor/operator instead of spinning forever.
- resume(): auto-spawn a watchdog if none is alive (skips when the watchdog itself
called resume — it lives in {SESSION}-watchdog — so no duplicate).
- launch-supervisor.py gate: defer while the per-run watchdog is alive (it is the
single writer for prompt-recovery). The supervisor only takes over once the
watchdog gives up (MAX_RESUMES) — i.e. a wedge a bare resume can't fix. Removes
the supervisor/watchdog double-resume race.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WxbpH3DquKzoSTSwGvGuET
Root-cause fix for the 2026-07-03 run stalling: the cc-ci host disk filled to
100% (ENOSPC) mid-run (Wave 6, lasuite-drive), the agent stopped to reclaim
space, and nothing resumed it — the log-idle/429 watchdog only covers opencode-go
usage-limit stalls, not an environmental wedge.
- launch-upgrader.py: step-0 prereclaim_cc_ci() prunes STALE cc-ci docker images
(unused AND older than a week, so this week's likely-reused images stay) before
each weekly run. Best-effort; env-tunable (UPGRADER_PRERECLAIM*).
- launch-supervisor.py (new): hourly glm-5.2 orchestrator wake-up. Cheap
deterministic gate — no-ops (zero tokens) when the run is complete or
progressing; only when a run stalled/died before completing does it launch a
short-lived glm-5.2 agent to diagnose + drive it to a clean DONE. Progress is
judged by live run-proc + log mtime (session_busy() is claude-tuned and misreads
a headless opencode run as idle).
- configuration.nix: cc-ci-upgrade-supervisor service + hourly timer (:07).
- upgrade-all SKILL §0: note the stale-image reclaim for manual runs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WxbpH3DquKzoSTSwGvGuET
Two gaps for the scheduled Thursday glm-5.2 run:
1. Survival: the watchdog was a Popen child of the Type=oneshot service, which
systemd's cgroup cleanup kills on exit. Spawn it under the persistent tmux
server instead (_spawn_watchdog), like the run sessions — survives the oneshot.
2. The report runs on glm-5.2 sharing the same opencode-go budget the upgrade run
drains, so it can 429-stall with no recovery. launch-report.py now spawns the
SAME watchdog pointed at the cc-ci-report session (generic via UPGRADER_SESSION/
_MODEL/_DONE_MARKER/_RESUME_FILE), with a report-specific resume prompt.
Also: _run_pids() is now scoped to the managed session (title or -s <sid>) so the
report watchdog can't kill the idle upgrader process and vice-versa; resume() adds
--dir and honors a custom resume prompt file.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
_completed() grepped the log for UPGRADE RUN COMPLETE, but the kickoff/resume
PROMPT (a user message) contains that string verbatim, so it false-positived
'done' while the run was still going. Check the model's ASSISTANT message output
via the web server API instead (log grep only as an offline, prompt-excluding
fallback).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Subcommand, function, env (UPGRADER_WATCHDOG), and log file renamed; behavior
unchanged. Only the opencode upgrader 'start' auto-spawns it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The opencode-go subscription's rolling usage-limit (429) ends the 'opencode run'
agent loop mid-run; it does NOT self-resume. Add:
- resume: continue the SAME session (context preserved) via 'opencode run -s <id>
--continue' — finds the session from the web server, kills the idle proc safely
(via /proc scan, never pkill -f self-match), relaunches in the tmux session.
- babysit: poll the session log; on a stall (>15min idle) wait out any 429
retry-after then auto-resume. Spawned automatically by an opencode 'start'.
So a usage-limit pause now self-heals instead of needing a manual nudge.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Make explicit that ALL formatting/HTML is owned by recipe-report.py render() and
the model's only artifact is the spec JSON — never hand-write/edit HTML. Matters
now that glm-5.2 drives the report. Also fix stale 'default opus' refs (report
now defaults to opencode-go/glm-5.2, overridable via REPORT_BACKEND/REPORT_MODEL).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Report launcher now defaults backend=opencode, model=opencode-go/glm-5.2 (claude
override → opus). Replaced the broken opencode 'attach + send-keys' path with the
same 'opencode run -m … --share --attach --title' pattern as the upgrader (the
old path passed no model and injected the prompt via keystrokes).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Weekly upgrade run now defaults backend=opencode, model=opencode-go/glm-5.2 with
no env set. Model default tracks backend (claude override → sonnet). Override via
LOOP_BACKEND/LOOP_MODEL or /srv/cc-ci/upgrader.env.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
cc-ci-upgrade-all now reads an optional EnvironmentFile so the weekly run can
switch backend/model (e.g. LOOP_BACKEND=opencode LOOP_MODEL=opencode-go/glm-5.2)
without a rebuild. Absent file → claude/sonnet (unchanged). Built+switched on
cc-ci-orchestrator-hetzner, host verified healthy.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The opencode backend emitted 'opencode --model X run ...' but -m/--model is a
flag on the run subcommand, so the model was being ignored. Move it after run.
Add OPENCODE_SHARE (default on): attach the session to the shared opencode web
server (oc.commoninternet.net) AND create a public --share link for monitoring.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Trust abra fully for any image whose tag it can read — a normal semver/calver
tag with no newer version is genuinely up-to-date. Only cross-check upstream for
images abra physically can't parse (tag+digest pins, digest-only pins), which is
the actual immich blind spot. Avoids redundant upstream checks on every recipe.
abra recipe upgrade is the first approach, but it silently contributes no
candidate for tag+digest pins (FATA: tag and digest not supported), digest-only
pins, and non-semver tags. immich kept getting skipped this way. Before
concluding SKIPPED — up-to-date, do a direct upstream tag check for every image
abra could not cleanly evaluate; only skip when BOTH agree nothing is newer.
Operator 2026-06-17. The recipe-test runtime env is declared 2-3x today
(harness.nix cc-ci-run runtimeInputs; nightly-sweep.nix duplicate pyEnv + a
divergent list + DEFECT-3 host-PATH patch; host systemPackages git-lfs for the
Drone runner) -> they drift (DEFECT-3 = missing bash/git-lfs in the timer).
Factor ONE shared env referenced by cc-ci-run, the sweep timer, and host
systemPackages; sweep invokes cc-ci-run so env is identical by construction.
Queued last (after settings).
Operator 2026-06-17: (1) no-canonical upgrade base should prefer the most recent
release TAG < head (a real published predecessor, reusing samever's helper),
with raw main-tip only as a last resort, then skip — always-on, improves this
server too. (2) SKIP_CANONICALS_FOR_UPGRADE=true feeds that same release-tag-first
fallback (so it swaps canonical->latest-release base without losing upgrade
coverage). (3) model bumped sonnet->opus.
Operator 2026-06-17. Introduce a minimal, extensible server-level settings.toml
for the cc-ci server; first value SKIP_CANONICALS_FOR_UPGRADE (bool, default
false, false on this server). When true, resolve_upgrade_base bypasses the
canonical and uses the main-tip predecessor — codifying that canonicals are an
optional optimization for the upgrade base. Scoped to the upgrade base only
(promote/--quick separate). Runs after canon/dash (builds on the final resolver).
Operator 2026-06-17: UPGRADE_BASE_VERSION is still used (plausible pins 3.0.1+
v2.0.0 to dodge the broken 3.0.0 base; bluesky-pds references it as a future
re-enable). Once canon establishes plausible's canonical at 3.0.1, the dynamic
base resolves correctly without the pin -> strip the key (meta/resolver/docs/
tests) + migrate plausible + update bluesky-pds note. GATED: keep it if
plausible genuinely still needs the escape-hatch (never drop upgrade coverage).
Operator 2026-06-17. /recipe/<name> shows only the latest run for most recipes
because history is built from a single page of the latest 100 Drone builds,
while 362 runs exist on the host. Source per-recipe history from the local
/var/lib/cc-ci-runs artifacts (already bind-mounted read-only) instead — full,
durable history. Deploy + verify live on bluesky-pds.
Operator 2026-06-17. (1) should_promote_canonical also requires the tested
version to have a release tag -> canonical is always a real published release,
never an untagged main commit. (2) sweep trigger is version/tag-keyed not
commit-keyed: skip a recipe unless its latest release tag is newer than its
canonical (skip new-untagged-commits-without-a-version). This makes the sweep
and samever ORTHOGONAL (samever never fires in the sweep; it's the PR-path
same-version guard). Updated gates/DoD accordingly.
Operator 2026-06-17: all recipes WARM_CANONICAL (watch warm-volume disk),
weekly timer, and an explicit M2 requirement to prove the sweep<->samever
interaction — skip uses COMMIT equality, samever uses VERSION equality; M2
must demonstrate all 3 sweep paths (skip / version-bump upgrade / same-version-
label step-back) and the commit-vs-version boundary.
