Don't override a hub-managed agent's real token with dev auth on localhost

[dea-rour]

Summary

createHubClient() in pkg/hubsync/sync.go (and a duplicated copy,
getHubClient(), in cmd/hub.go, plus the display-only getAuthInfo())
prefers dev auth over a non-dev scion-token whenever the Hub endpoint is
localhost, specifically to avoid a stale agent token left over on a
developer's machine from a previous remote hub connection.

That exception fires unconditionally — including inside a container the
Runtime Broker itself started. Workstation mode (Hub + Broker + Dashboard on
one machine, per the docs) always runs the Hub on localhost, so every
hub-managed agent in workstation mode hits this branch. Inside that
container the scion-token is freshly minted for this exact Hub, not stale
at all — but it gets silently swapped for dev auth anyway.

Dev auth resolves to a generic/dev identity, not a specific agent. Any Hub
endpoint that requires the caller to be a particular agent breaks under
it — concretely, POST /api/v1/agents/{id}/outbound-message (used by
scion message user:<x>, an agent's only way to reach a human's inbox
directly) 401s with "Agents can only send outbound messages as themselves",
surfaced to the CLI user as the generic authentication failed, login to hub with 'scion hub auth login'.

Net effect: no hub-managed agent running against a workstation-mode Hub
could ever message a user directly via scion message — a full feature
gap for any agent/harness that doesn't have its own out-of-band relay to
the message broker (discovered running an OpenCode-harnessed agent, which
has no such native relay, alongside a Claude Code agent that does).

Fix

Gate the localhost dev-auth-preference exception behind
!config.IsHubManagedAgent() — an existing helper keyed off
SCION_AGENT_ID, which the Runtime Broker sets when it starts an agent
container — in both createHubClient and getHubClient, and apply the
same condition in getAuthInfo so scion hub auth status reports
accurately from inside an agent container too.

The original developer-workstation scenario this exception was written for
is untouched: it only ever applied when SCION_AGENT_ID isn't set, which
is true for any normal CLI invocation outside an agent container.

Test plan

• go build ./... — clean
• go vet ./cmd/... ./pkg/hubsync/... ./pkg/config/... — clean
• go test ./cmd/... ./pkg/hubsync/... — all passing
• Added TestGetAuthInfo_HubManagedAgentUsesRealTokenOnLocalhost and
  TestCreateHubClient_HubManagedAgentUsesRealTokenOnLocalhost —
  regression coverage for the fixed case (agent token used, not dev
  auth, when SCION_AGENT_ID is set on a localhost endpoint)
• Fixed TestGetAuthInfo_DevAuthPreferredOverStaleAgentTokenOnLocalhost
  to explicitly clear SCION_AGENT_ID — a latent test-hermeticity gap
  the fix exposed (the existing suite happened to always run outside
  any agent context, so nothing had previously needed this cleared)
• Verified live against a real workstation-mode deployment: an agent
  container that previously 401'd running scion message user:<name> --channel telegram "..." from inside itself

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[gemini-code-assist]

Code Review

This pull request ensures that hub-managed agents (where SCION_AGENT_ID is set) use their freshly minted scion-token when communicating with a localhost Hub, rather than being incorrectly overridden by dev auth. This resolves a bug where agents on workstation-mode Hubs received 401 Unauthorized errors on self-only endpoints. The feedback recommends handling errors explicitly with t.Fatal(err) instead of ignoring them with _ = in the new test setup, and resolving or removing the <TBD> placeholder issue URL in the code comments.

[gemini-code-assist]

In test setup, ignoring errors with _ = can lead to silent failures or confusing test errors if the directory cannot be created or the token file cannot be written. It is highly recommended to handle these errors using t.Fatal(err) to fail the test immediately with a clear error message.

Suggested change

	_ = os.MkdirAll(scionDir, 0700)
	// Real per-agent token, freshly issued by the broker for this Hub.
	_ = os.WriteFile(filepath.Join(scionDir, "scion-token"), []byte("fresh-agent-jwt"), 0600)
	if err := os.MkdirAll(scionDir, 0700); err != nil {
	t.Fatal(err)
	}
	// Real per-agent token, freshly issued by the broker for this Hub.
	if err := os.WriteFile(filepath.Join(scionDir, "scion-token"), []byte("fresh-agent-jwt"), 0600); err != nil {
	t.Fatal(err)
	}

[gemini-code-assist]

The comment contains a placeholder <TBD> in the issue URL. Please update this with the actual issue number or remove the placeholder reference if it is not yet available.

Suggested change

	// self-only endpoints. See https://github.com/GoogleCloudPlatform/scion/issues/<TBD>.
	// self-only endpoints.

[ptone]

Thanks for the contribution - this looks correct, can you click through and sign the CLA?