Wizard scopes a manual procedure, maps each stage’s journey, and authors a bash script from a fixed template. The script opens each URL, says what to click and copy, captures values, writes them to .env or GitHub secrets, confirms at every stage and shows how many stages remain.
It is meant for provisioning, credentials, CI secrets, unfamiliar dashboards and one-off migrations, and not for steps the agent can perform itself.
What You Get
- A runnable bash script with stage-by-stage progress and confirmation gates
- Hidden entry for secrets and idempotent .env upserts
- GitHub secret and variable writes through gh
- A closing summary of what was captured and where it went
Common Use Cases
- Walk a teammate through setting up third-party service credentials
- Set up CI secrets that every workflow reference needs
- Run a one-off migration or cutover with confirmation before each irreversible action
What Makes It Unique
The interaction design is solved once in the template, so every wizard behaves the same way and the agent only authors stages. It requires a procedure scoped from the repo’s env files and workflows, instructs the agent to say so when it does not know a dashboard path, and verifies the script statically since it cannot be run end to end without a human.
Under The Hood
A SKILL.md plus template.sh, a bash library with helpers such as stage, ask_secret, open_url, write_env, set_secret and confirm. The agent copies the template, reads the repo’s env files, docker-compose files and GitHub workflows, writes stages, then runs bash -n and shellcheck and sets the executable bit. The finished script opens browsers, writes .env and calls gh secret and gh variable.