What does this skill do, and when should you use it?
cmux-backend is one of the 25 skills bundled in the manaflow-ai/cmux repository, located at skills/cmux-backend/SKILL.md. It is not a tool script but a set of backend development rules for AI coding agents: it mandates the Effect framework under web/app/api and web/services, establishes Postgres as the single source of truth for VM lifecycle, and constrains how migrations and secrets are handled. When an agent is asked to modify backend APIs, Cloud VM lifecycle code, or provider integrations, these rules steer it toward changes that match the project's architecture.
- Enforces Effect as the default framework when editing web/app/api/, web/services/, and backend scripts touching providers, databases, auth, rate limits, retries, timeouts, or telemetry
- Requires Next route handlers to stay thin: parse the request, run one Effect program at the boundary, map typed errors to HTTP responses
- Establishes Postgres as the source of truth for VM lifecycle, active VM limits, idempotency, and usage events
- Specifies production/staging connections to the PlanetScale PostgreSQL database cmux-prod, with the runtime reading DATABASE_URL and CMUX_DB_DRIVER=url
- Mandates migrations run via bun run cloud-vm:migrate -- staging then -- production, never from Vercel build or route startup
- Guides loading provider secrets like FREESTYLE_API_KEY and R2 upload vars from ~/.secrets/cmux.env
- A developer maintaining cmux's backend who needs an AI agent to follow project conventions when editing Effect services under web/services
- Someone building new Cloud VM control-plane features who wants the agent to use Postgres for VM lifecycle and limits rather than reintroducing Rivet or a raw actor protocol
- A maintainer needing to run migrations against staging/production PlanetScale databases in the correct order with the correct commands
- Configuring Stack Auth team payment items as Cloud VM create pricing gates and wanting rule-guided changes
- Preparing a local secret environment for provider image build scripts that require sourcing ~/.secrets/cmux.env
- Not suited for cmux frontend React code or trivial TypeScript data shapes — the skill explicitly limits where Effect applies
- Not for projects outside the cmux repository — rules are deeply tied to cmux's specific paths, database, and secret file layout
How do you install this skill?
- The web/ directory the skill targets is under BUSL-1.1, requiring a commercial license for production use; the skill does not disclose this.
- This is a conventions-only skill that runs no scripts; rule-to-codebase consistency is unverified by execution.
- The backend stack depends on overseas services (PlanetScale, Vercel, Stack Auth); mainland-China reachability of core functions is not declared.
- Repository license metadata is NOASSERTION and the publisher is unverified by FollowSkills; identity is unknown, not suspicious.
- Shell / CLI
- Network access
- Local filesystem
BunDocker (local Postgres via bun dev)PlanetScale PostgreSQL
This skill is part of the 25-skill collection bundled in the manaflow-ai/cmux repository; the source docs give no standalone install command. Clone the repo and place the skills/cmux-backend/ folder into your Agent Skills-compatible client.
git clone https://github.com/manaflow-ai/cmux.git
cp -r cmux/skills/cmux-backend ~/.claude/skills/How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Add a Cloud VM rate-limiting service under web/services using Effect per cmux-backend rules, mapping errors to HTTP responses
- Run a migration against the staging PlanetScale database for me, making sure it is not triggered from a Vercel build
- I want to add a Stack Auth team payment gate to Cloud VM creation — consult effect-boundaries.md and cloud-vm-control-plane.md first, then implement
The skill triggers via its YAML frontmatter description: when a task involves web/app/api, web/services, backend scripts, Cloud VM lifecycle, provider integrations, Postgres, Stack Auth pricing gates, migrations, or provider image build scripts, the agent should load it and follow its rules. The core rules reference two detailed documents: references/effect-boundaries.md (route handlers, services, typed errors, retries, dependency injection) and references/cloud-vm-control-plane.md (VM lifecycle, migrations, Postgres, provider idempotency, pricing gates). For local development, bun dev sources ~/.secrets/cmux.env first, then ~/.secrets/cmuxterm-dev.env.
What are this skill's strengths and limitations?
- Highly specific rules bound to real paths, commands, and environment variables — immediately actionable
- Includes two detailed reference documents covering Effect boundaries and the Cloud VM control plane in depth
- Explicitly prohibits anti-patterns (running migrations in Vercel builds, reintroducing Rivet)
- Only valuable within the cmux repository itself; the rules do not transfer to other projects
- No executable scripts or test suite — adherence depends on the agent's self-checking
- The repo's License field is NOASSERTION; actual licensing is GPL-3.0 plus BSPL 1.1 for web/ (Business Source License 1.1), so commercial use of the server needs care
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| cmux Backend Development Rules Skill this page | 51 · Use with care | ★ 28k | 1d ago | NOASSERTION |
| Effect JSDoc Authoring Skill | 77 · Strongly recommended | ★ 26k | 3d ago | MIT |
| Anti-Slop: Low-Evidence Pattern Rules for Oxlint | 62 · Recommended | ★ 5.3k | 1mo ago | MIT |
| PinMe Worker Auth API Integration | 59 · Recommended | ★ 3.7k | 3mo ago | MIT |
| Supabase Development Guide | 55 · Use with care | ★ 2.7k | 9d ago | MIT |
Compared to generic Effect-TS or Next.js best-practice guides, cmux-backend's value is that its rules map precisely onto cmux's actual codebase (specific paths, PlanetScale database, secret file locations) rather than generic framework advice; for developers not working on cmux, the official Effect documentation is likely a better fit.
How did FollowSkills review this skill?
The skill executes no code and requests no broad permissions; its secrets guidance (~/.secrets separation, AWS credentials not treated as DB credentials, protected migration secret) shows least-privilege and data-flow awareness. Deducted for: no user confirmation or rollback mechanism, no self-disclosure of data flows, attribution limited to repo level with unverified publisher, and NOASSERTION license metadata.
Instructions are internally consistent (SKILL.md and both references align, no contradictions) with complete progressive-reference structure. Deducted for: purely rules-based content with no runnable scripts, no tests or failure-feedback path; static calibration caps at 10 and key paths (migrations, Effect boundaries) were not executed.
Trigger conditions are explicit and specific (web/app/api, web/services, Cloud VM lifecycle, migrations, pricing gates); fit/non-fit boundaries (Effect vs plain TS) are well drawn. Deducted for: core dependencies (PlanetScale, Vercel, Stack Auth, AWS) are all overseas services with no declared Chinese-language support or mainland-China reachability consideration; no explicit non-fit list.
Good layering (overview + detailed references), stable naming, mapped to repo structure. Deducted for: no per-skill version, changelog, or maintenance-owner statement; the repo is dual-licensed GPL+BUSL and the skill's target directory web/ falls under BUSL, which the skill does not disclose; license metadata is NOASSERTION.
As a coding-conventions skill, its value claim (uniform Effect architecture, migration discipline, idempotency checklist) is concrete and actionable, giving real marginal value to cmux backend contributors. Deducted for: static review cannot verify correctness of rule-driven output, and comparative benefit vs. reading architecture docs directly has limited evidence.
Referenced files exist and are mutually consistent; repository has genuine CI workflows corroborating engineering practice. Deducted for: no independent tests or third-party execution evidence tied to the skill's rules, and consistency of rules with the actual codebase is only inferable; static calibration caps at 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →