running-tend

name	running-tend
description	PRQL-specific guidance for tend CI workflows. Adds a standing exception for filing issues in other repos, PR title conventions, CI structure, Dependabot-batch polling, weekly maintenance tasks, and issue-closing policy on top of the generic tend-* skills. Use when operating in CI.

Running Tend in PRQL

Tend-specific guidance for this repo. Project build commands, test strategy, error conventions, etc. are in CLAUDE.md — don't duplicate them here.

Filing issues in other repos

Standing exception granted: file directly in agent-equipped targets (per Filing Issues in Other Repos in the bundled running-in-ci skill) without asking permission here first. The default rule (open an issue here asking permission first) still applies when the target shows no agent signals.

PR conventions

• PR titles use conventional commits: feat:, fix:, docs:, chore:, refactor:, test:, ci:, internal:, devops:, web:, refine:
• No scope required (e.g., fix: resolve date parsing not fix(parser): ...)
• Dependabot PRs use chore: prefix

CI structure

• Main CI workflow: tests (watched by tend-ci-fix)
• Dependency management: Dependabot opens dependency PRs; tend-weekly reviews them and runs the tasks under Weekly maintenance below.
• tend's own action is excluded from Dependabot (max-sixty/tend is in the github-actions ignore list in .github/dependabot.yaml). Tend updates flow through the nightly tend/update-workflows regen (uvx tend init), which follows structural changes a version-only bump can't — e.g. the 0.1.7 move that split the action into claude//codex/ subdirectories broke the naive Dependabot bump #6031. Don't re-add max-sixty/tend to Dependabot.
• Automerge: not configured — pull-request-target.yaml only validates PR titles and handles pr-backport-web backports. The automerge job was removed in #5753, so bot PRs must be merged manually by a maintainer (or via repo branch-protection auto-merge if a maintainer enables it on the PR).

Verifying a rust-toolchain.toml bump

The update-rust-toolchain action opens build: Update rust toolchain version with nobody owning it, so tend usually pushes the mechanical lint fixes a new clippy demands (the repo compiles with -D warnings). Verify with the matrix's full feature set, not --features=default. The tests matrix runs clippy as --all-targets --no-default-features --features=default,test-dbs-external,lsp, and code behind test-dbs-external — prqlc/prqlc/tests/integration/dbs/ — is invisible to a default-only run.

On #6219 (1.96.1 → 1.97.1) a session fixed the 7 useless_borrows_in_formatting sites --features=default exposed and posted "clean across the workspace", but an 8th in dbs/runner.rs was still red — so the posted claim was wrong, not just the branch.

A whole-workspace --all-targets clippy on a cold cache exceeds the session budget. Scope it to the failing compilation unit instead (~9 minutes):

cargo clippy -p prqlc --test integration --target=x86_64-unknown-linux-gnu \
  --no-default-features --features=default,test-dbs-external,lsp -- -D warnings

Then scope the resulting claim to match the command: a clean run there clears that one compilation unit, not the workspace — other crates and targets stay unchecked. Name the unit that was verified rather than repeating #6219's "clean across the workspace".

CI polling during the Dependabot batch

Dependabot opens its whole batch over a couple of minutes (the 2026-08-03 batch ran 17:14:32 → 17:16:45; across 2026-06 to 2026-08 every batch has landed in 17:12–17:19 UTC), so five or six tests matrices compete for runners at once. The surviving tests run on each PR then sits in QUEUED for a long time before it starts — run 30835855220 on #6130 took 73 minutes end to end (17:14:37 → 18:27:46), far past the 9-minute cap on the poll loop in CI Monitoring in running-in-ci.

Stop after one poll round when every pending check is QUEUED. A QUEUED check has not been allocated a runner, so another round changes nothing: post the verdict, name the unverified checks, and end. If any pending check is IN_PROGRESS, keep polling — that work is advancing and may still settle.

The pending() helper in CI Monitoring returns a count without the states, so it can't tell those two cases apart. Project the states alongside it:

gh pr view <n> --json statusCheckRollup \
  --jq '[.statusCheckRollup[] | {name: (.name // .context), status: (.status // .state)}]'

Weekly maintenance

These tasks run as Step 3 of the bundled weekly skill (only when workflows.weekly is enabled in .config/tend.yaml).

• Bump pinned go-task/setup-task version. The action is invoked with a concrete version: input to avoid the intermittent unable to get latest version failure from version: 3.x (see #5836). Dependabot does not update with: inputs, so this needs a manual weekly bump. Find the latest release at https://github.com/go-task/task/releases/latest; if the current pin is older, update version: X.Y.Z in:

  • .github/actions/tend-setup/action.yaml
  • .github/workflows/build-web.yaml
  • .github/workflows/test-php.yaml
  • .github/workflows/test-prqlc-c.yaml

  Open a single chore: PR with the bump. Skip if already at the latest.

Issue management

• Close bot-opened issues once the underlying cause is resolved — don't leave them open for a maintainer. If you (prql-bot) filed an issue (e.g., a nightly "tests failed" issue, a code-quality issue, an infra/upstream bug report) and the fix has merged or the upstream problem has been addressed, close the issue with a short comment citing the resolution (e.g., "Resolved by #NNNN — closing"). Applies to any issue where author.login == prql-bot.