🤖 feat: add admin policy file enforcement

[ibetitsmike]

Summary

• Adds an opt-in, admin-controlled policy file (via MUX_POLICY_FILE) that is loaded at startup and refreshed every 15 minutes.
• Enforces policy server-side for:
  • allowed providers + forced base URLs + allowed models (strict JSON allowlists)
  • MCP user-defined servers (stdio vs remote)
  • runtime allowlist (including ssh vs ssh+coder distinction)
  • minimum client version gating (blocks usage when below minimum_client_version)
• Adds renderer PolicyContext + UI gating (hide/disable disallowed options; block actions rather than auto-switch).
• Adds documentation page for policy file configuration.

Background

• Mux deployments often need centrally managed constraints so users cannot enable unsupported providers/models/MCP servers/runtimes.

Implementation

• PolicyService parses a strict JSON policy file (JSON.parse), validates it with zod, and publishes an effective (normalized) policy via ORPC.
• Backend services (ProviderService / AIService / WorkspaceService / MCP server manager + MCP endpoints) enforce restrictions.
• Frontend subscribes to policy changes and filters available UI choices; actions are blocked when current selections become disallowed.

Validation

• make static-check
• Added unit tests for policy parsing/version gating and UI helpers.

Risks

• Policy files are a high-leverage admin control surface; strict schema validation + last-known-good refresh behavior reduce accidental unlocks.

---

Generated with mux • Model: openai:gpt-5.2 • Thinking: high • Cost: $20.21

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 774cd1446a

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[ibetitsmike]

@codex review

Addressed the outstanding review threads:

• Treat policy state blocked as deny-all for backend enforcement.
• Apply runtime allowlist checks to workspace forks.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c3efccb87d

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Filter settings model dropdowns by policy

Policy filtering is only applied to the provider list used for adding custom models here, but the default/compaction model selectors still use the unfiltered getSuggestedModels(config) list later in this component. When an admin policy restricts providers/models, users can still select disallowed models in Settings, and subsequent sends fail with policy_denied. Consider driving those dropdowns from the policy-filtered list (e.g., useModelsFromSettings().models) or filtering allModels against the effective policy.

Useful? React with 👍 / 👎.