What the agent does
The agent reads the workflow and the tools it calls before it changes a version, because in a Go CLI the action pins and the Go matrix fail together. In a project like spf13/cobra-cli, .github/workflows/test.yml has three jobs: golangci-lint on actions/setup-go@v2 with Go 1.17 and golangci/golangci-lint-action@v2 at version: latest; test-unix on a matrix of ubuntu and macOS across Go 1.14 to 1.17 with actions/checkout@v3.1.0 and actions/cache@v2 keyed on go.sum, installing richgo and gox with go install before make test build; and test-win on msys2/setup-msys2@v2 with MinGW's Go. go.mod says go 1.15 with cobra v1.6.1 and viper v1.12.0. .golangci.yml enables deadcode, golint, interfacer, maligned, structcheck, varcheck and gas, and .github/dependabot.yml already asks for weekly github-actions updates, which means the pins it has not moved are the ones it cannot move alone.
The agent resolves each action with gh api to its latest release and the commit that tag points at: actions/checkout v7.0.1, actions/setup-go v7.0.0, actions/cache v6.1.0, golangci/golangci-lint-action v9.3.0 and msys2/setup-msys2 v2.33.0 as of the day it runs, and writes each as uses: owner/action@<sha> # vX.Y.Z so the version is readable and the pin is immutable. It then reads what each major removed. setup-go from v4 on caches the module and build directories itself when go.sum is present, so the separate actions/cache@v2 step and its restore-keys go away rather than get bumped. golangci-lint-action v9 wants golangci-lint v2, so version: latest would run a linter that refuses the v1-style .golangci.yml with its deleted linters; the agent pins version: v2.15.0 and rewrites the config with golangci-lint migrate, dropping deadcode, golint, interfacer, maligned, structcheck and varcheck (all removed upstream, their checks now inside unused and staticcheck) and renaming gas to gosec.
The Go matrix is the other half. Go 1.14 to 1.17 are out of support and actions/setup-go@v7 still installs them, so the job would pass while testing versions nobody ships. The agent changes the matrix to the two supported releases, 1.26 and 1.27 as go.dev/dl lists them when it runs, and the lint job to go-version-file: go.mod with check-latest: true. It does not raise the go 1.15 directive in go.mod, because that is a compatibility promise to people who install the generator with an older toolchain, and it notes the choice. richgo and gox are still installed with go install ...@latest, since the Makefile calls both; the case statement that chose between @latest and no version for Go 1.14 and 1.15 is deleted along with those versions.
The agent runs what it can in its sandbox before it opens the pull request: actionlint on the workflow, golangci-lint run -v at v2.15.0 against the migrated config, make fmt, make test and make build on a current Go, and zizmor to check the pins and permissions. It cannot run the Windows job locally, so the pull request says so and points at the checks that run on the pull request itself. The description lists each action with its old tag, new tag and SHA, and names the two behaviour changes, the dropped cache step and the linter migration, separately from the plain bumps.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Update the GitHub Actions in this Cobra CLI repository. Read .github/workflows/test.yml, .github/dependabot.yml, .golangci.yml, Makefile and go.mod first. Resolve each action to its latest release with gh api and pin it by commit SHA with a version comment. Read each major release's notes: drop the separate cache step if setup-go now caches modules itself, and if the golangci-lint action now requires golangci-lint v2, pin that version and migrate .golangci.yml, removing linters v2 deleted. Replace the Go matrix with the currently supported Go releases and use go-version-file for the lint job, but do not raise the go directive in go.mod. Run actionlint, golangci-lint at the pinned version, make fmt, make test and make build in the sandbox, run zizmor on the workflow, then open a pull request that lists every action's old tag, new tag and SHA and names each behaviour change separately from the plain bumps.",
"resources": [
{
"url": "https://github.com/your-org/your-app"
}
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
What comes back is a pull request that changes .github/workflows/test.yml and .golangci.yml and nothing else. The description has a table of each action with its old tag, new tag and the SHA it is now pinned to, a paragraph on the removed cache step, a paragraph on the linter migration listing the six linters dropped and the one renamed, the new Go matrix, and the golangci-lint, make test and make build output from the sandbox.
The run's record keeps every gh api call that resolved a tag to a commit, the golangci-lint migrate run and its diff, and the test and build runs, so a reviewer can check a SHA against the release it claims without leaving the record.
Before merging, a reviewer checks that the Go versions the matrix names are the ones the project means to support, decides whether go.mod's directive should move too, and watches the Windows job on the pull request, which the sandbox could not run. main.go, cmd/ and the golden files under cmd/testdata are untouched.
Run it on every Monday morning
.github/dependabot.yml already proposes action bumps weekly, but a bump that needs a config migration or removes a step sits unmerged until a person reads the release notes. A scheduled job on the reader's server sends this request each Monday with the repository in resources and a request_id that is a UUID v5 of the repository's full name and the ISO week, so a second send in the same week returns the run already started. When every pin is already current and the linter config is already v2, the agent reports that and opens nothing.
A team with several Go CLIs sends one request per repository; each run reads its own workflow, so a repository already on setup-go@v7 with a v2 .golangci.yml gets only the SHA refresh, and one with a goreleaser action gets that resolved and pinned the same way.
The use case this job follows, with its automation and trigger: Upgrade dependencies weekly.
Why run it on Rush?
The hard part of an actions update is not editing a tag but knowing what each major release removed: a cache step that setup-go made redundant, a lint action whose new default refuses the old config, a Go matrix that quietly tests versions nobody ships. In a Rush sandbox the agent resolves every tag with gh, reads the release notes, migrates the linter config with the linter's own tool and runs make test and make build on the result, so the pull request reflects what the workflow will do and not what the diff looks like.
The gh api answers, the migration diff and the test run stay with the run and the pull request cites them. Doing this across a dozen Go repositories at once costs the same attention as one, because each run has its own machine, its own record and its own pull request.
Frequently asked questions
Why pin by SHA rather than keep @v7?
A tag can be moved; a commit cannot. Pinning owner/action@<sha> # v7.0.1 makes the workflow run exactly the code the reviewer looked at, and the comment keeps the version readable for people and for Dependabot, which updates SHA pins and their comments together.
Why drop the cache step instead of bumping it?
actions/setup-go has cached ~/go/pkg/mod and the build cache itself, keyed on go.sum, since v4, so a second actions/cache step keyed on the same file saves nothing and doubles the restore time. Bumping it to v6 would be a correct-looking change that makes the job slower.
What happens to the .golangci.yml linters?
The action's new major runs golangci-lint v2, whose config format changed and which no longer ships deadcode, golint, interfacer, maligned, structcheck or varcheck; their checks live in unused and staticcheck now, and gas became gosec. The agent runs golangci-lint migrate, keeps every linter that still exists, and shows the before-and-after lint output so a reviewer can see no new warning appeared.
Does it change go.mod?
No. The go 1.15 directive is what lets people on older toolchains build the generator, and raising it is a separate decision with a different audience. The matrix tests current Go releases because those are what the runners and the module proxy support today; the directive says the minimum, the matrix says what is checked.