routine-env

name	routine-env
model	sonnet
effort	medium
allowed-tools	Read, Write, Edit, Bash(claude:*), Bash(npx:*), Bash(jq:*), Bash(yq:*), mcp__tilth__*
description	Render or apply a "routine env" — a named bundle of Claude Code plugins + MCP servers + skill packs defined in `envs/<name>.yaml` — into a setup-script body for Claude Cloud (Claude Code on the web), or apply it to the local machine. Use when the user says "set up a routine env", "provision my Claude Cloud environment", "install my cheese env", "render the setup script for <env>", "update my env", "scaffold a new env", or invokes "/routine-env". Do NOT use for cloud cron/scheduled agents that open PRs (that is `routine-scaffold`) or for a personal CLAUDE.md preferences overlay in a contributed repo (that is `claude-local`) — routine-env is about plugin/MCP/skill provisioning only.

routine-env

Turns an envs/<name>.yaml bundle definition into the commands that actually install it — either as a setup-script body to paste into the Claude Cloud UI, or run directly on the local machine.

What a routine env is

A routine env = a named, versioned bundle of:

• marketplaces — claude plugin marketplace add <ref> sources
• plugins — claude plugin install <id> targets (<plugin>@<marketplace-id>)
• mcp — npx-launched MCP servers (not plugins)
• skills — skill packs installed via the npx skills@latest CLI

Definitions live in envs/<name>.yaml at the repo root (see envs/README.md for the full schema and field-by-field rationale). This skill is the only consumer of envs/ — the directory is standalone data, not wired into ap, profiles/base, the agents/ registries, or dots sync.

The Claude Cloud reality

"Claude Cloud" here means Claude Code on the web (claude.ai/code), a different product from Managed Agents (ant beta:environments). Ground facts from .cheese/research/routine-env-provisioning/routine-env-provisioning.md (addendum, A1/A2/A5):

• The environment's setup script is UI-configured plain bash — typed into the Claude Cloud environment dialog's "setup script" field. It is not a repo-committed file, and <certain> there is no CLI/API to push it from git. This skill cannot write directly to that dialog — it can only render the bash body for the user to paste in.
• The setup script runs once per build cache-miss (first run, or after an invalidating change) and then persists into every session reusing that cached container — it is not re-run per session.
• A3 caveat — the setup-script install path is unverified. Whether claude/node/npx are even available inside the setup-script shell, and whether a plugin or skill installed there survives into the booted session, is graded <don't know> in the research (A3): every official setup-script example is plain OS commands (apt install, docker compose); none invoke claude, npx, or node. Spike this empirically on a real Claude Cloud environment before relying on the rendered setup script for plugins/skills.
• The confirmed (<certain>) path for plugins/skills-only envs is git-native, not the setup script. A repo's committed .claude/settings.json (enabledPlugins/extraKnownMarketplaces) auto-installs at cloud session start with zero setup-script involvement, and .claude/skills/ loads with the clone — no install step, no A3 uncertainty. When an env only needs plugins/skills (no OS packages, no services), committing to .claude/ is the confirmed default, and rendering a setup-script body is the fallback for what that path can't cover (OS packages, background services) — not the default over a confirmed path. Mention the .claude/ alternative to the user before reaching for render on a plugins/skills-only env.

Modes

render <env>

1. Read envs/<env>.yaml.
2. Emit one idempotent bash setup-script body, in this order:
   • claude plugin marketplace add <ref> for each marketplaces[] entry
   • claude plugin install <id> for each plugins[] entry
   • claude mcp add <name> -- npx -y <package> for each mcp[] entry (e.g. claude mcp add tilth -- npx -y @paulnsorensen/tilth-nightly)
   • npx skills@latest add <pack> <args> for each skills[] entry
3. Every command must be safe to re-run (idempotent) — rely on each CLI's own already-installed/already-added no-op behavior; do not add custom existence checks that duplicate what the CLI already handles.
4. Print the rendered body as a single fenced bash block the user can paste verbatim into the Claude Cloud environment dialog's setup-script field.

apply-local <env>

Run the same rendered commands directly on the current machine via Bash, in the same order as render. Report each command's outcome.

update <env>

Bump plugin/skill versions or ids in envs/<env>.yaml (e.g. after confirming a new marketplace id or skill pack revision), then re-run render <env> so the user has the refreshed setup-script body to re-paste.

new <name>

Scaffold a fresh envs/<name>.yaml from the locked schema in envs/README.md (name, description, marketplaces, plugins, mcp, skills), with the same inline confirm-comment on every plugins[].id. Leave placeholder values for the user to fill in — do not fabricate marketplace ids or skill packs.

Verification

• Confirm every plugins[].id marketplace segment against claude plugin marketplace list before treating it as final — a marketplace's actual .name in its marketplace.json is not assumed by envs/*.yaml.
• Reuse plugin coordinates from agents/plugins/registry.yaml (the cross-harness plugin registry) — that file is the source of truth for which plugin lives at which marketplace; do not invent new coordinates in envs/*.yaml.

What it never does

• Not cloud cron/scheduled agents. A routine env provisions plugins/MCP/ skills for a session to boot with; it never authors a recurring watcher that opens PRs — that is routine-scaffold.
• Not a CLAUDE.md preferences overlay. It does not distill or write a CLAUDE.local.md — that is claude-local.
• Does not wire into ap, profiles/base, the agents/ registries, or dots sync. envs/ is standalone data consumed only by this skill.
• Does not push the rendered setup script into the Claude Cloud UI programmatically — no such CLI/API exists (see "The Claude Cloud reality" above); the user pastes the rendered body in manually.
• Does not fabricate marketplace ids or skill packs — every id traces to agents/plugins/registry.yaml or an explicit user-confirmed source, with the claude plugin marketplace list confirmation step called out above.