Conversation
prql-bot
left a comment
There was a problem hiding this comment.
#6363 regenerates these same nine workflows at 0.3.1 — including the same merge: maintainer and control_plane_owner: "" inputs — while also switching the harness from claude to codex. The two PRs therefore collide on every line this one touches, and whichever lands second needs a fresh uvx tend init rather than a textual merge. #6363 is a draft, so nothing here is blocked on it, but these aren't two independent bumps: taking both is a decision about which harness runs tend.
Separately, the note in .github/actions/tend-setup/action.yaml that records the sandbox_setup: deprecation pins its claim to a specific version — ".config/tend.yaml still uses the deprecated sandbox_setup: key, which 0.2.15 accepts with a warning and emits as a run: step after this composite". Once this merges the repo pins 0.3.1, so the note stops answering the question it exists to answer: is the key still accepted at the version we run? I ran the generator at both versions against this branch's config — the deprecation warning is unchanged and both emit the same combined run: step — so the answer is still yes, and only the version needs updating. Pushing that one-line change to this branch.
Regenerates the nine
tend-*workflows with tend 0.3.1, up from 0.2.15. The diff is the action pin plus two new inputs on every job, both set to the value that preserves today's behaviour.merge: maintainer. max-sixty/tend#1172 adds an opt-inmerge: yolomode that lets the bot merge ordinary PRs itself, under a PR-only branch-protection bypass pinned to the expected head SHA.maintaineris the default and is what the generator emits here, because.config/tend.yamlsets nomerge:key — so the bot still cannot merge anything, and this PR grants it no new authority. Adopting yolo later would be a deliberate config change, not a side effect of a regen.control_plane_owner: "". The CODEOWNER that would gate workflow and.config/tend.yamlchanges under yolo. Empty undermaintainer, where it has no effect.tend checkis stricter, and that is why a new drift line appeared tonight. max-sixty/tend#1382 extends the branch-protection check fromupdatealone to the whole ref lifecycle —creation,update,deletion— and tightens credential-environment auditing. Tonight's sweep ran the 0.3.1 check and reportedbranch-protection:mainas failing for the first time; the repo's rulesets did not change. That is tracked on #6340, and clearing it needs a repository admin to runtend check --fix.Smaller upstream fixes that reach this repo's runs:
review-runscorrection rows now keep each row's author (#1374), the run listing pages to the Actions API ceiling instead of stopping early (#1377), and the CI poll loop is bounded by a sleep budget rather than a pass count (#1376). The pinned Claude Code CLI moves to 2.1.280.Full comparison: 0.2.15...0.3.1. Release mechanics, lockfile bumps and tend-internal work are omitted above.
The generated files are the generator's own output. Re-running
uvx tend@0.3.1 initagainst this branch reproduces the nine workflows byte-for-byte, so the diff is the generator's and nothing hand-edited slipped in.One hand-written file changes alongside them: the comment in
.github/actions/tend-setup/action.yamlthat records thesandbox_setup:deprecation named 0.2.15 as the version accepting the key, which this bump makes stale. Running the generator at 0.2.15 and 0.3.1 gives an unchanged deprecation warning and the same combinedrun:step, so the note now names 0.3.1 and the migration it points to is still a follow-up.