Summary

The s6 dashboard run script flipped --insecure on whenever HERMES_DASHBOARD_HOST was anything other than 127.0.0.1 / localhost. The comment ("the dashboard refuses otherwise") predates the OAuth auth gate — back then, start_server SystemExit'd on any non-loopback bind, and the run script's --insecure was the only way to make in-container deployments work at all.

The gate has since been replaced by should_require_auth(host, allow_public), which engages the OAuth flow when a DashboardAuthProvider is registered (the bundled dashboard_auth/nous provider auto-registers when HERMES_DASHBOARD_OAUTH_CLIENT_ID is set) and fails closed with a specific operator-facing error when none is. The host-derived --insecure ran upstream of all that and silently disabled the gate on every container-deployed dashboard.

Live evidence (before fix)

Test agent on the wildcard-subdomain rollout (nous-account-service PR #221), HERMES_DASHBOARD_OAUTH_CLIENT_ID correctly injected by the portal on first boot, bundled nous provider registers — and:

$ curl https://<agentId>.agents.staging-nousresearch.com/api/status
{
  ...
  "auth_required": false,
  "auth_providers": ["nous"]
}

Provider registered. Gate off. The combination is only reachable via allow_public=True, i.e. --insecure reached start_server. The dashboard SPA was served to anyone on the public internet, including /sessions, with no /login redirect.

Fix

Derive --insecure from an explicit opt-in env var, HERMES_DASHBOARD_INSECURE (truthy values matching the rest of the s6 boolean envs: 1, true, TRUE, True, yes, YES, Yes). Operators on trusted LANs behind a reverse proxy without the OAuth contract opt in explicitly; portal-managed agent deployments leave it unset and let the gate engage.

Behaviour table after the fix:

Compose-file impact

docker-compose.windows.yml already passes --insecure on the command: array directly (line 38), so it doesn't depend on the s6 auto-injection. No compose-file change required.

Tests

• tests/test_docker_home_override_scripts.py — extends the existing static-text guard with test_dashboard_run_does_not_derive_insecure_from_bind_host, asserting the legacy host-derived case-statement is gone and the new env-var opt-in is present (locks against accidental revert).
• tests/docker/test_dashboard.py — adds two Docker-in-Docker tests exercising the actual /api/status round-trip:
  • test_dashboard_oauth_gate_engages_on_non_loopback_bind — 0.0.0.0 bind + HERMES_DASHBOARD_OAUTH_CLIENT_ID → auth_required: true, "nous" in providers.
  • test_dashboard_insecure_env_var_opts_out_of_gate — 0.0.0.0 bind + HERMES_DASHBOARD_INSECURE=1 → auth_required: false.

Both new Docker tests follow the existing _poll-style pattern in test_dashboard.py and probe via venv Python's urllib.request (the image doesn't ship curl).

Docs

• website/docs/user-guide/docker.md + the matching zh-Hans i18n file — adds HERMES_DASHBOARD_INSECURE to the env-var table, replaces the previously stale prose ("the entrypoint no longer auto-enables insecure mode" — which was flat-out wrong until this PR) with an accurate description of the gate's trigger conditions and the explicit opt-out.

Local verification

• shellcheck clean (--severity=error, matches CI's docker-lint.yml).
• tests/test_docker_home_override_scripts.py — both tests pass.
• tests/docker/test_dashboard.py collects all 8 tests without errors (Docker harness not run locally — CI's docker-publish smoke test exercises it).

Validation after merge

1. Build & publish a new nousresearch/hermes-agent:latest.
2. On the next portal-side agent create, the new image is picked up automatically.
3. Hit https://<agentId>.agents.staging-nousresearch.com/api/status — expect "auth_required": true and "auth_providers": ["nous"].
4. Hit https://<agentId>.agents.staging-nousresearch.com/sessions in a browser — expect redirect to the portal's OAuth /authorize endpoint.

Lane

Solo-landable Docker/s6 territory (only docker/s6-rc.d/dashboard/run, two test files, and two doc files touched — no run_agent.py / cli.py / gateway/ / model-related code). Routine teammate review per branch protection.

cc @teknium1 for the approving review (Docker/s6 lane, but maintainer review required by branch protection).

Handover source: filed by Ben after the nous-account-service portal-side fix (commit a3347645 on the agent_domains branch / PR #221) confirmed all four dashboard env vars are now injected on first boot; the only remaining blocker for end-to-end OAuth on the wildcard subdomain was this s6 script.