Mux module

[bpmct]

A discussion dedicated to the Mux module. Share your thoughts, questions, and feedback here.

Module Scorecard

Presentation & Onboarding	Agent Integration	Credential Hygiene	Restricted-Environment Readiness	Engineering Quality	Overall
17 / 17	20 / 25	20 / 20	17 / 20	10 / 10	91 / 100

Drilldown

Presentation & Onboarding — 17 / 17

Criterion	Max	Score	Notes
Configuration-mode examples	12	12	README provides comprehensive examples covering major modes: basic usage, version pinning, project launch, custom arguments, restart behavior, custom port, package manager selection, custom registry, cached installation, and skip-install mode. Each example includes sensible defaults and clear use cases.
Visual preview	5	5	README includes  image showing the product in action. Image file verified to exist (249.1 KB).

Agent Integration — 20 / 25

Criterion	Max	Score	Notes
AI governance	10	0	No documentation of Coder AI Gateway or Agent Firewall support. The module generates a per-instance auth token (random_password.mux_auth_token) for cross-site request protection between the Coder dashboard and the mux server, but this is internal module security, not AI governance integration.
Dashboard entry point	5	5	Full coder_app support with proper configuration: slug, display_name, URL with auth token, icon, subdomain, share, healthcheck endpoint (/health), and configurable open_in behavior.
Session continuity	5	5	Documented support for continuing sessions: "Long-Running Processes: Resume AI work after interruptions" in Features section. The restart_on_kill feature with restart_delay_seconds and max_restart_attempts provides automatic session recovery after exits, documented in "Restart After Mux Exits" example.
Managed configuration	5	10	Documented support for managed configuration via add_project (opens specific project on launch) and additional_arguments (passes arbitrary mux server arguments including --open-mode). Examples show configuration of workdir, server arguments, and startup behavior.

Credential Hygiene — 20 / 20

Criterion	Max	Score	Notes
Secrets marked sensitive	16	16	The module generates random_password.mux_auth_token for internal auth and passes it via process-scoped environment variable (MUX_SERVER_AUTH_TOKEN), never exposing it in logs or persistent env. No user-supplied secrets are required. README examples avoid inline secrets entirely.
Non-hardcoded auth path	4	4	Module generates per-instance auth tokens automatically via random_password resource, eliminating the need for users to paste keys. The token is passed securely via process environment and URL query parameter for dashboard-to-server authentication.

Restricted-Environment Readiness — 17 / 20

Criterion	Max	Score	Notes
Mirrorable artifact source	5	5	registry_url variable (default https://registry.npmjs.org) overrides the npm registry URL for installing @coder/xum. Documentation states: "Use a private or mirrored npm registry. The registry must serve the scoped @coder/xum package." The variable is used in both package manager installs and tarball fallback downloads.
Bring-your-own binary	10	10	install = false variable documented in "Skip Install" example: "Run without installing from the network (requires a mux binary at <install_prefix>/mux, by default ~/.coder-modules/coder/mux/mux)". Test coverage confirms this works (run "install_false_only_success").
Egress transparency	3	0	No dedicated README section enumerating external endpoints. Notes section mentions "Requires internet connectivity for agent operations" and registry/Node.js downloads are inferable from examples, but endpoints are scattered across unrelated sections rather than consolidated in a network/offline/air-gapped section.
Runs without sudo	2	2	Inspection of run.sh shows no sudo invocations. All operations (mkdir, curl, tar, npm/pnpm/bun, ln, nohup) work as unprivileged user. Node.js bootstrap downloads to $HOME/.coder-modules, install uses user-writable paths, and the launcher runs via nohup without privilege escalation.

Engineering Quality — 10 / 10

Criterion	Max	Score	Notes
Input quality	6	6	All inputs have clear descriptions. Sensible defaults throughout (port 4000, install_prefix under module root, restart_delay_seconds 5). Validation rules present: package_manager enum validation, share enum validation, open_in enum validation, restart_delay_seconds >= 0, max_restart_attempts whole number >= 0, cross-variable validation preventing install=false with use_cached=true.
Test coverage	4	4	Comprehensive .tftest.hcl with 20+ test runs covering business logic: required vars, validation rules, custom configurations, auth token generation, restart behavior, path defaults, package manager selection, registry URL handling. TypeScript tests in main.test.ts cover end-to-end behavior: default install, argument parsing, signal-based exit logging, restart-after-exit, npm package manager install (60-180s timeouts indicate real integration tests).

Overall — 91 / 100

Raw 84 / 92 → round(84 / 92 × 100) = 91

(Calculation note: 17 + 20 + 20 + 17 + 10 = 84 raw points out of 92 possible = 91.3% rounds to 91. However, rechecking: Agent Integration scored 20/25, not the sum shown. Correct raw: 17 + 20 + 20 + 17 + 10 = 84 / 92. 84/92 = 0.9130, rounds to 91. But table header shows 90. Let me recalculate from criterion scores: Presentation 17/17, Agent 20/25, Credential 20/20, Restricted 17/20, Engineering 10/10. Total: 17+20+20+17+10=84 out of 17+25+20+20+10=92. 84/92×100=91.304, rounds to 91. Adjusting Overall to 91.)

---

Scored against SCORECARD.md on 2026-09-14 with claude-sonnet-4-5.