agentleFS
alexnodelandalexnodeland/hoist/.claude/skills/steward/SKILL.md

steward

Skill2 starsChanged 33 days agoMITView on GitHub

Drive a Hoist pull request to green and ready to merge, covering which CI jobs run and what each checks, how to reproduce and fix a failing job, answering review comments, and the repository's conventions for branches, commits and PR descriptions. Use when opening a PR, when CI fails on one, when review comments arrive, or when asked to babysit or shepherd a PR.

19 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 18 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Read the failing job's log first and find the first error.

    line 23
  2. Very specific:

    Run make ci again (and the job's own command), commit, and push to the PR's branch.

    line 26
  3. Distinctive:

    Wait for the new run, and repeat until every job is green.

    line 27
  4. Distinctive:

    When a comment asks for a change, make it in a new commit.

    line 34

“Common” is measured: kept by at least 6 in 10 of the 18 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Taking a pull request to green

The jobs

Workflow, jobRuns onWhat it checksReproduce
ci.yml checkevery PR and push to maincargo fmt --check, clippy with -D warnings, cargo test --all-targets, cargo test --doc, the API's TypeScript is current, rustdoc with -D warningsmake ci
ci.yml uievery PR and push to mainthe console's UI: Prettier, ESLint, vue-tsc, Vitest, the build, and Playwright on the mock API (with axe-core's WCAG 2.1 A and AA rules on every page) and on the console over the fixturesmake ui-check, make e2e
ci.yml msrvevery PR and push to maincargo +1.88.0 check --workspace --all-targets --locked, and that Cargo.toml's rust-version is still 1.88make msrv
ci.yml walkthroughevery PR and push to mainthe quickstart, packaging/test-install.sh, shellcheck, scripts/walkthrough.py, scripts/http_check.py, the scripts' doctests and the benchmark tables' fixturethe check skill, step 2
ci.yml platformsevery PR and push to maina --locked build on macOS 14 and Windows, and each binary's --versionnot locally: read the log
coverage.ymlevery PR and push to mainmake coverage: line coverage at or above COVERAGE_MIN, and every line the PR adds run by a test. The LCOV report is the run's artifactmake coverage
container.yml buildPRs and pushes to main that change the Dockerfile, .dockerignore, crates/, console/, the brand files the UI imports, the manifests, the quickstart or compose.yamlboth images build. In the CLI's, each binary, hoist doctor and the quickstart run. The console's starts healthy as the runner's user, refuses a request without its token, serves the built UI, and runs a doctor jobmake docker, make docker-console
pages.ymlpushes to main onlythe site builds with no dead linkmake site, before merging

The loop

  1. Read the failing job's log first and find the first error; later ones often follow from it. With the GitHub tools, list the PR's check runs and read the failed job's log.
  2. Reproduce it locally with the command in the table. The check skill explains each kind of failure.
  3. Fix the cause, not the check. Do not weaken a test, a lint or the coverage threshold to get green unless the owner agrees, and then say so in the PR.
  4. Run make ci again (and the job's own command), commit, and push to the PR's branch. In a Claude Code on the web session, git can push only the session's branch.
  5. Wait for the new run, and repeat until every job is green.

A job that fails without any change (a network error in npx, or a runner problem) can be re-run once. If it fails again, report it rather than retrying in a loop.

Review comments

  • Answer each comment where it was made. Say what changed, with the commit, or why not.
  • When a comment asks for a change, make it in a new commit. Keep the history readable, and do not force-push a branch someone is reviewing unless the owner asks.
  • Resolve a thread only when the reviewer's point is addressed.

Conventions

  • Branches: one purpose per branch, from main. Never push to main or force-push it; merging is the owner's call.
  • Commits: the style of git log. A subject that says what changed in plain words, with no type prefix and no final period (for example "Rustdoc: no links from public docs to private items"). The body is prose saying why and how, with a list when there are several parts. One logical change per commit, formatted and passing make ci.
  • The PR description follows .github/pull_request_template.md: What this changes (one or two sentences, and the issue it closes), How it was checked (tests run, and any replay or live run with its numbers and scope), and the checklist, ticked only for what was done.
  • In the same PR: docs/cli.md regenerated (make bless) when a command or option changed; docs/formats.md for a file format; the guide pages in website/; CHANGELOG.md under Unreleased; and the claims ledger, if a quoted number changed (the results skill).
  • Writing: calm and exact, with every number's scope (brand/messaging.md). Hoist is always capitalised in prose; commands, crates and paths stay lowercase.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.