Set up Flare Actions CI for a repo end to end. Use when asked to set up, install, onboard, or wire up Flare (or CI) for a project. Probes the deployment, wires the repo, starts an executor, and verifies HEAD with a run digest.
Installs into .claude/skills of the current project.
Are you the author of Flare Setup?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/everyai-com-flare-setup)
---
name: flare-setup
description: >-
Set up Flare Actions CI for a repo end to end. Use when asked to set up,
install, onboard, or wire up Flare (or CI) for a project. Probes the
deployment, wires the repo, starts an executor, and verifies HEAD with
a run digest.
---
# Flare setup
One-command adoption: `npx flare connect` does the probing, wiring, and
verification. Drive it; don't hand-roll the steps below except to explain
what needs a human click.
## The loop
1. **Dry-run first**: `npx flare connect --dry-run` (needs
`FLARE_ACTIONS_URL`). It prints the plan and changes nothing.
2. **Human clicks** (you cannot do these — link them, don't stall on them):
- Deploy, if there is no deployment yet: `npm run setup` is fully
non-interactive and prints the Worker URL; the one-click Deploy to
Cloudflare button needs a human's Cloudflare consent.
- Dashboard admin claim (first login) and **Connect GitHub**
(GitHub App Create flow) — both are browser clicks at
`https://<worker>/dashboard`.
3. **Wire the repo**: `npx flare connect` prints the recipe for the
situation (App install URL, or the no-App webhook recipe). With
`GITHUB_TOKEN` and `FLARE_ADMIN_TOKEN` present you may pass `--wire`
to create the repo webhook — ask the human before any GitHub-side
change. If the repo has no pipeline, `connect` names the detected
stack and what `init` would scaffold; pass `--init` to scaffold it
(stack-matched starter or converted workflows) in the same flow.
4. **Executor**: if `connect` reports nothing picked the run up, start
one — `npm run runner` from the Flare checkout (BYO machine), or
managed seats (`npm run setup` on a docker machine). Runner mode
(`runs-on: flare`) needs `npm run runner -- --github` instead.
5. **Verify**: `connect` dispatches HEAD and reports the digest itself.
On failure, hand off to the verify loop (failing step + tail), not a
blind re-run.
## Rules
- Never invent a deployment URL — ask for `FLARE_ACTIONS_URL` or deploy.
- Never claim "CI passed" without a run digest (see flare-verify).
- Workflows run unchanged: don't rewrite `.github/workflows` to adopt;
`npx flare init` is opt-in scaffolding, not a requirement.
- Runner mode is a human decision (it asks the GitHub App for
`administration:write`): propose it, link
`docs/GITHUB-RUNNERS.md`, and let them enable it in Settings.
## Reporting back
Deployment URL, wiring state (App vs webhook), executor in use, and the
verification digest (run id, status, failing step + one tail line).