fix(ci-runner): annotate CLAUDE_CODE_VERSION for Renovate, and bump it #94
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "fix/ci-runner-claude-code-and-renovate-annotation"
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?
Merging this cuts ci-runner 1.2.3 — which is what unblocks the rest of the queue.
The bug
The comment above this ARG said "Renovate bumps this ARG". It did not, and could not.
renovate.json's Dockerfile custom manager matches a# renovate:annotation immediately followed by the ARG line.CLAUDE_CODE_VERSIONhad no annotation — so it was invisible, pinned at2.1.214while upstream reached2.1.223. Nine patch releases, silently, on a pin everyone believed was automated. The comment was the only thing maintaining it, and comments don't open PRs.Verified, not assumed
I ran
renovate.json's own regex over the Dockerfile. Seven ARGs are now visible to it:CLAUDE_CODE_VERSION@anthropic-ai/claude-codeCOSIGN_VERSIONsigstore/cosignCVE_GATE_VERSIONharbor.webgrip.dev/webgrip/cve-gateEXTERNAL_NODE20_VERSIONnodeHELM_VERSIONhelm/helmSYFT_VERSIONanchore/syftYQ_VERSIONmikefarah/yqThree remain unannotated, deliberately:
DOTNET_CHANNEL=9.0,PHP_VERSION=8.3,NODE_MAJOR=24are major-channel selectors. Moving those is a migration with a blast radius, not a patch bump — they should stay under human control.Still open, and worth your attention
yq carries a correct annotation and was nine minor versions stale anyway (4.44.3 vs 4.53.3, fixed by hand in #88). Renovate can see it, so something else is dropping it — a schedule, the shared
webgrip/renovate-configpreset, or a PR that was closed unnoticed.That needs the Renovate dashboard rather than another guess from here, and it matters more than yq: the same mechanism is what's supposed to be moving
cosign,syftand thecve-gatedigest — all three security-relevant.Why this PR exists at all
Run 240's release died on Harbor 502s. No tag was cut, and
determine-changed-directoriesis push-diff, so a failed release never retries itself — it needs either a re-run click or a newops/dockercommit. This is a real change that also happens to be that commit.The companion PR adds a manual trigger so the next failure doesn't need one.