configure: add --skip-unavailable so --agents can configure the available subset

[Edwinhe03]

Problem

--agents treats an explicit list as all-or-nothing. If any named agent isn't available on the workspace, configure_workspace_command raises before configuring any of them:

if unavailable_tools:
    _print_discovery_diagnostics(state)
    raise RuntimeError(f"Requested agent(s) not available on this workspace: {displays}.")

So on a workspace whose AI Gateway exposes no OpenAI models (codex discovery: no endpoint exposes api_type openai/v1/responses):

ucode configure --profiles DEFAULT --agents claude,codex,pi --use-pat --skip-validate --skip-upgrade

configures nothing and exits 1 — even though claude and pi are both usable there (pi needs only one of claude/codex/gemini). For CI running one command across heterogeneous workspaces, one unavailable model family takes down the whole setup.

Why not just make --agents fail open

The default stays strict, because the two directions aren't symmetric:

• Strict default: a caller who wants tolerance opts in with a flag.
• Fail-open default: a caller who wants a guarantee has no way to get it back. --agents claude,codex exiting 0 having configured only claude leaves the pipeline believing codex works — the failure resurfaces later at ucode codex launch time, far from the cause.

Naming agents explicitly is a request for a specific outcome, so it should keep meaning "all of these, or tell me why not."

Change

Add an opt-in --skip-unavailable:

ucode configure --agents claude,codex,pi --skip-unavailable

• Configures the available subset, preserving the order given in --agents.
• Skips the rest with a warning (Skipping agent(s) not available on this workspace: Codex.), on top of the existing per-source discovery diagnostics.
• Still exits non-zero when none of the requested agents are available — the pre-existing if not available_on_workspace: return 1 covers this, since that list is built only from selected_tools. The flag can't turn a completely unusable workspace into a silent success.
• Requires --agents: the interactive picker already offers only available agents, and --agent names a single agent whose absence is the whole answer.
• The strict error now points at the flag: ... Pass --skip-unavailable to configure the available ones instead.

Behavior with the flag absent is unchanged.

Tests

Added to tests/test_cli.py (the existing test_unavailable_selected_tool_errors_before_configure still guards the strict default):

• test_skip_unavailable_configures_available_subset — the azure/no-GPT case: requesting claude,codex,pi configures ["claude", "pi"], installs only those binaries, and warns about Codex.
• test_skip_unavailable_still_fails_when_none_available — returns 1, never calls configure_selected_tools.
• test_strict_error_mentions_skip_unavailable — discoverability of the flag from the error.
• test_skip_unavailable_requires_agents / _forwarded_with_agents / _absent_by_default — CLI wiring.

uv run pytest: 1690 passed, 1 failed. The one failure (test_managed_wizard.py::TestCliWiring::test_successful_apply_exits_zero) reproduces on pristine origin/main in a clean worktree with none of these changes — pre-existing and unrelated. uv run ruff check . and ruff format --check are clean.

Docs

README gets the strict-vs-tolerant distinction next to --agents, plus a quick-reference row.

[Edwinhe03]

Live-tested against a real workspace

Verified end-to-end (not just the mocked unit tests) against ai-oss-ecosystem-integration-testing with --dry-run: real OAuth, real AI Gateway verification, real model discovery. To reproduce the "GPT not allowed" workspace, GPT ids were filtered out of the model-services response and discover_codex_models was stubbed to return the real-world empty result ([], "no endpoint exposes api_type \openai/v1/responses`")` — everything else (auth, gateway, Claude/Gemini/OSS discovery from 51 live model-services ids) ran unmodified.

Case	Result	Exit
--agents claude,codex,pi --skip-unavailable, no GPT	Skipping agent(s) not available on this workspace: Codex. → Claude Code + Pi configured	0
--agents claude,codex,pi, no GPT (strict default)	ERROR Requested agent(s) not available on this workspace: Codex. Pass --skip-unavailable…, nothing configured	1
--agents codex --skip-unavailable, no GPT (none available)	ERROR No coding agents are available on this workspace., nothing configured	see note
--agents claude,codex,pi --skip-unavailable, all available (control)	all three configured, no spurious skip	0

Also confirms Pi configures correctly without codex — the if codex_models guard in agents/pi.py drops the databricks-openai provider rather than erroring.

Pre-existing bug found while testing (not fixed here)

The "none available" case prints ERROR No coding agents are available on this workspace. but exits 0. Cause: configure_workspace_command returns 1 there, but all seven call sites in configure() discard the return value — nothing converts it into typer.Exit(1). Reproduced on pristine origin/main with none of this branch's changes, so it predates this PR and affects the existing strict path too.

I've left it out to keep this PR scoped. Worth noting for CI users: with --skip-unavailable, a workspace where no requested agent is available currently exits 0 with an error on stderr. Happy to fix it here or in a follow-up — say which you'd prefer.

[Edwinhe03]

@tt-le added this fallback for oss models for codex on azure