RFC: migrate interactive prompts from survey to huh

[nf-matt]

I've been using this project, and it's really helpful! I like to give back on projects I use. I went to add ESC key support and found some challenges related to AlecAivazis/survey -- archived (last release June 2023 with 74 open issues that probably won't be addressed).

Footprint in this repo:

used feature	times
survey.WithStdio	17
survey.Input	15
survey.Select	14
survey.AskOne	10
survey.Confirm	9
survey.SelectQuestionTemplate	3
survey.WithValidator	2
survey.Question	2
survey.Prompt	2
survey.Required	1
survey.Password	1
survey.MinLength	1
survey.AskOpt	1
survey.Ask	1

Affected packages: assume, alias, browser/detect, cfaws/{env,creds}, granted/{credentials,sso,tokens,uninstall}, all of granted/settings/*, all of granted/registry/*, plus the pkg/testable shim.

Beyond the unmaintained risk, survey's API is awkward:

1. pkg/assume/assume.go:670 monkey-patches the package-global survey.SelectQuestionTemplate to render the profile-list header row, then restores it. A theme would be cleaner.
2. pkg/assume/assume.go:660 builds a profileNameMap (display-string → real-name) because survey.Select.Options is "stringly" typed. Generic select erases this.

There's no clean way to add an ESC-cancel keybinding to survey.Select (it has no key-handler hook).

Proposed approach — incremental, no big PR:

1. PR 1 — close the testable shim's leak (call sites that bypass the shim and call survey.AskOne/survey.Ask directly)
2. PR 2 — introduce a Prompter interface in pkg/testable, add huh alongside survey, migrate pkg/cfaws/env.go (a single Confirm).
3. PR 3+ — one package per PR, in roughly increasing complexity, ending with pkg/assume/assume.go (the only place that needs the custom template/theme work).
4. Final PR — drop the survey dep.

Each PR is independently revertible and ships behavior-equivalent UX.

huh vs. promptui: promptui last released in 2021 (also effectively dormant). huh is active (latest is March 2026), has generic-typed Select, theming as a first-class concept, and ESC-cancel built in. It's Bubble Tea–based, so the testable shim needs the interface refactor (PR 2) rather than a transparent pass-through.

Before I write any code, is this something you'd be open to? Happy to drive the work if you're open to it. Thoughts?

[nf-matt]

Ok, I have 9 PRs ready to go, or I can merge all of it locally and push as a single PR. Your call if you're interested. All tests/CI passing.

Didn't turn out to be too difficult, and I like where it landed and I'm using it locally. Should make future granted/assume UI work easier and more flexible.

[nf-matt]

I've opened a draft PR implementing this migration: #956.

Since I didn't hear back here, I consolidated the RFC's proposed per-package sequence into a single stacked branch — but the commits still follow the incremental plan step-by-step, so it can be reviewed commit-by-commit and each step is independently revertible. It ends by dropping the survey dependency entirely, and includes unit tests for the newly introduced prompt logic.

go build ./... is clean and go test ./... passes. Happy to split it back out into smaller PRs, rework anything, or close it if the direction isn't wanted — just let me know.