Add guided multi-account setup and evaluation - #14927

Prototype overview and documentation

DO NOT MERGE. This stack is almost entirely vibe coded. These drafts are for handy review and prototyping, not merging as-is.

Expect the code to be really bad in places: highly coupled, handling edge cases that do not really matter, or doing suspiciously fragile things like parsing GraphQL request bodies.

I suspect some of the suspicious code also exposes the complexity of this feature set, because account selection, Git configuration, scope precedence, worktrees, SSH and API requests all interact.

The prototype experience is the purpose, not the code. I want to find out whether the experience is useful, not defend this implementation or treat it as a design to preserve.

Description

Follow on from #14926

Evaluating persistent authentication changes needs a clear setup path and a way to isolate the experiment. This PR adds guided setup, developer evaluation tools and debug-only keyring isolation.

Layer 8 of an eight-part prototype review stack. This is a way to review the design and implementation in smaller pieces, not a proposal to merge the current stack as-is.

How did you test this change?

I exercised the debug keyring-isolation capability and reset in disposable configuration. I read the actual global and managed files before reset, then saw the user include remain and authenticated with the retained real account.

$ cat ~/.gitconfig; cat ~/.ordinary.gitconfig
[include]
	path = /Users/Shared/gc-H8vKqb/r-5aVX8u/home/.ordinary.gitconfig
[user]
	name = Ordinary Author
	email = ordinary@example.invalid

Reset privately before returning to stable gh

Verify the debug runner prerequisite before using accounts

Key points

This layer adds evaluation and setup tooling, not a new identity storage format. The reset recording exercises cleanup already available in the stack so reviewers can see how to return to ordinary configuration.

Live evaluation tools are optional and can create GitHub resources. The offline review scripts use synthetic accounts and deny network and Keychain access. Real-account recordings are separate from those scripts.

Notes for reviewers

Start with the setup approval boundaries, then isolation and cleanup. The guided-agent workflow and full live acceptance runner were not exercised for these recordings. Reset managed configuration before removing a prerelease executable referenced by Git.

The recordings use the product code at this layer, not a later stack endpoint. The included offline replay scripts are supplementary and do not reproduce the live-account recordings.

make
env -i HOME="$HOME" PATH=/usr/bin:/bin:/opt/homebrew/bin \
  /bin/bash script/multiaccount-review/08-reset.sh --gh "$PWD/bin/gh" --stable-gh /absolute/path/to/stable/gh

Authorship and follow-up

Who wrote this:

• A human wrote it.
• An agent wrote it under close human direction.
• An agent wrote it independently, and no human has guided the implementation beyond the initial prompt.

Who answers review comments:

• @username will read and reply directly. Name the account.
• An agent will draft replies and @username will read them before they are posted.
• Nobody has explicitly committed to replying.

Acceptance Criteria

Prototype behaviour under review:

Given a developer explicitly chooses to evaluate the prototype.

When the isolated workflow is used and the experiment is reset.

Then managed identity wiring is removed while ordinary user configuration and saved accounts remain.

DO NOT MERGE