feat(cli): structured error for non-interactive workspace selection

[ItamarZand88]

Summary

When something automated runs alien login — an agent like Claude, a script, anything without a real terminal — it can't use the arrow-key menu to pick a workspace. Before, it hit a dead end ("needs a real terminal"). Now it gets the actual list of workspaces back, so it can pick one with a follow-up command.

What an agent sees when it runs alien login:

1. The browser opens and you log in — same as always.
2. It needs a workspace but can't show the menu, so it returns WORKSPACE_SELECTION_REQUIRED with the names under context.workspaces. ← the new bit
3. It reads the names and runs alien workspaces set <name>.

So instead of a dead-end string, you get back something a machine can act on. I also removed alien login --json — an agent can get the same list from alien workspaces ls --json, so login didn't need its own JSON mode.

How I tested it

On a local stack, logged in fresh, on an account with 3 workspaces:

• Ran alien login, finished in the browser — it ended on WORKSPACE_SELECTION_REQUIRED with the 3 names and a hint pointing at alien workspaces ls / set.
• alien workspaces ls --json gave the list, alien workspaces set <name> saved it, alien workspaces current confirmed it.
• alien login --json now says "unexpected argument".
• cargo test -p alien-cli --features platform is green, including a test that checks the names actually land in context.workspaces.

[greptile-apps]

Greptile Summary

This PR replaces the opaque "requires a real terminal" string error in prompt_workspace with a machine-readable WORKSPACE_SELECTION_REQUIRED error that carries the available workspace names in context.workspaces, and removes the now-redundant --json flag from alien login.

• error.rs: New WorkspaceSelectionRequired { workspaces: Vec<String> } variant exposes workspace names in the structured error context; accompanied by a unit test that validates the context.workspaces array via AlienErrorData.
• workspace.rs: prompt_workspace now calls InteractionMode::current(json_mode).is_machine() instead of require_prompt, returning the new typed error when running non-interactively; login.rs hardcodes json_mode = false (correct since --json is removed from login).
• lib.rs / login.rs: LoginArgs is reduced to a fieldless struct, wants_json_output returns false for Login, and the JSON output branch is deleted — a deliberate breaking change for callers that used alien login --json.

Confidence Score: 5/5

The change is a tightly scoped error-type upgrade with no mutations to auth logic and no new async paths; safe to merge.

All changed paths are narrow: the new structured error is only returned when multiple workspaces exist and the caller is non-interactive, the single-workspace fast path is untouched, and the removal of --json from alien login is intentional and self-consistent. The test directly exercises the context.workspaces payload. No auth state is modified before the error is returned.

No files require special attention.

Important Files Changed

Filename	Overview
crates/alien-cli/src/commands/platform/login.rs	Removes the --json flag and LoginOutput struct from alien login; simplifies login_task to always emit human-readable output and forwards json_mode=false to prompt_workspace.
crates/alien-cli/src/commands/platform/workspace.rs	Replaces the generic require_prompt ConfigurationError with a structured WorkspaceSelectionRequired error when InteractionMode::current detects machine mode; drops can_prompt import since it is now called internally by InteractionMode::current.
crates/alien-cli/src/error.rs	Adds WorkspaceSelectionRequired error variant carrying Vec workspaces and a unit test that asserts the variant serializes context.workspaces correctly via AlienErrorData.
crates/alien-cli/src/lib.rs	Updates wants_json_output to return false for Login commands (destructuring the now-fieldless LoginArgs with _); consistent with the removal of the --json flag.
crates/alien-cli/src/ui.rs	Renames the test module from tests to command_event_tests to avoid a name collision, no functional change.

Sequence Diagram

sequenceDiagram
    participant Agent
    participant CLI as alien CLI
    participant Auth as force_login
    participant API as Platform API
    participant Browser

    Note over Agent,Browser: New non-interactive workspace-selection flow

    Agent->>CLI: alien login (no TTY)
    CLI->>Auth: force_login()
    Auth->>Browser: open OAuth URL
    Browser-->>Auth: token callback
    Auth-->>CLI: AuthHttp

    CLI->>API: list_workspaces()
    API-->>CLI: ["ws-a", "ws-b", "ws-c"]

    Note over CLI: workspaces.len() > 1 AND is_machine() == true
    CLI-->>Agent: WORKSPACE_SELECTION_REQUIRED context.workspaces: ["ws-a","ws-b","ws-c"]

    Agent->>CLI: alien workspaces set ws-a
    CLI->>API: validate_workspace_name("ws-a")
    API-->>CLI: ok
    CLI->>CLI: save_workspace("ws-a")
    CLI-->>Agent: workspace saved

    Note over Agent,Browser: Single-workspace or interactive flow (unchanged)
    Agent->>CLI: alien login (single workspace)
    CLI->>Auth: force_login()
    Auth-->>CLI: AuthHttp
    CLI->>API: list_workspaces()
    API-->>CLI: ["ws-a"]
    Note over CLI: workspaces.len() == 1 - auto-select
    CLI->>CLI: save_workspace("ws-a")
    CLI-->>Agent: human-readable success output

Loading

_{Reviews (1): Last reviewed commit: "feat(cli): structured error for non-inte..." | Re-trigger Greptile}