design-agent-deploy
GitHub指导部署 Design Agent,支持 Managed Deep Agents 托管或 Docker Compose 自托管。涵盖环境检查、密钥配置、沙箱发布及 Slack 集成验证,确保团队安全运行代理。
Trigger Scenarios
Install
npx skills add langchain-ai/design-agent --skill design-agent-deploy -g -y
SKILL.md
Frontmatter
{
"name": "design-agent-deploy",
"description": "Put Design Agent in a team's Slack. Deploy it to Managed Deep Agents (one command, managed Slack app) or self-host it with Docker Compose, verify it with a real request, ship updates, and keep a separate test copy. Use when someone wants the agent running for their team."
}
Deploy Design Agent
A deployment is what the team uses. Ask before mda deploy, before connecting Slack, and before
anything that bills (publishing a snapshot, hosting). Never print a key; name missing settings.
Setup comes first: uv run design-agent doctor should show no FAIL line and the brand pack should
validate (design-agent-onboarding, brand-onboarding). A first deploy can use the sample brand
and needs no local Docker; doctor's mda-deploy line lists what the deploy still needs.
| Managed Deep Agents (recommended) | Self-hosted | |
|---|---|---|
| Who runs it | LangSmith: runtime, sandboxes, Slack app, schedules | the person: API, Postgres, Docker host |
| Needs | a LangSmith Cloud account (MDA is in public beta) and a deployment-capable LANGSMITH_API_KEY |
Docker with Compose v2.24+ |
| Cost | hosting is paid, plus model and media spend | their machine, plus model and media spend |
Managed Deep Agents
Ask first, in one message: do they have a LangSmith Cloud account with Managed Deep Agents (US, public beta), which model (Anthropic by default, or the LangSmith Gateway), whether to start with the sample brand (fine for a first deploy) or theirs, and whether to deploy a test copy first. Say what bills: the snapshot build, hosting and model spend (pricing).
Keys never pass through you. Name the keys (LANGSMITH_API_KEY, needs deployment permission,
and the model's key) and have the person run uv run design-agent config set LANGSMITH_API_KEY ANTHROPIC_API_KEY in their own terminal: each value is typed hidden. Without a terminal the command
refuses rather than asking you for the value. Never ask for a key in chat.
Long and interactive steps. sandbox publish takes 10 to 15 minutes and asks to confirm: after
their yes, run sandbox publish --yes in the background. The first mda deploy sets up the Slack
app and waits for a person: a browser sign-in (no LANGSMITH_API_KEY in .env), an authorization
link, and Enter in a real terminal. Have the person run it in their own terminal. Later deploys
need no Slack step; run them after a yes (--yes overwrites same-name resources).
- Publish the computer once.
uv run design-agent sandbox build && uv run design-agent sandbox testproves the recipe locally;uv run design-agent sandbox publishbuilds the 16 GiB snapshot in their workspace (sandbox time is billed). Publish again wheneversandbox/setup.shchanges. - Check the keys.
mda deployforwards every.envkey except LangSmith's as a deployment secret. List the key names in.envand confirm they are only what the deployed agent uses (model, media,DESIGN_AGENT_*,PAPER_*); self-hosting keys (SLACK_*,DATABASE_URL,DESIGN_AGENT_API_TOKENS) must not be there. Renders inworkspace/out/ship too and count toward MDA's 200 MB archive limit.uv run design-agent doctornames all of this (mda-deploy), anduv run mda buildcompiles what would ship into.mda/build/. - Try a copy first.
uv run mda deploy --name design-agent-testis a separate deployment; the team's agent is untouched. - Deploy with
uv run mda deployafter a yes. The first deploy provisions the Slack app (its name isDESIGN_AGENT_SLACK_NAMEin.env). If LangSmith isn't authorized for their Slack workspace, the CLI prints an HTTPS link: they open it, pick the workspace and approve (a workspace admin may have to), then press Enter. Rerun the deploy if the link was missed. Without aLANGSMITH_API_KEY,mda deployoffers a browser sign-in. The Slack app is managed: there is no manifest orSLACK_*key to set, whatever a generic MDA skill says. - Verify with real requests. Once live, the agent DMs the person who deployed; they reply,
or
/inviteit to a channel and mention it. Start with a quick still, then a short storyboarded teaser. In Slack you should see live progress, a check-in while it renders, and files attached at the end. Without Slack, call the deployment's API (docs/deploy-mda.md has the snippet); files attach to the run's LangSmith trace.uv run mda logstails the deployment. - Updates: change, deploy the test copy, then
uv run mda deploy. A redeploy stops running helpers; each affected thread offers to resume from its saved transcript.
Self-hosted
uv run design-agent config set DESIGN_AGENT_MODEL=…(the person types the key or sets it in.env), thenuv run design-agent config generate DESIGN_AGENT_API_TOKENS POSTGRES_PASSWORD.docker compose -f deploy/self-host/compose.yaml up -d --build, thencurl -s http://127.0.0.1:8080/health.- Slack (optional): create an app from
config/slack-manifest.example.yaml, setSLACK_BOT_TOKENandSLACK_APP_TOKEN, then add--profile slackto the compose command. - Verify:
POST /threads/{id}/messageswithAuthorization: Bearer <token>, then pollGET /threads/{id}/updatesfor background work. - Security: the API starts sandboxes through the Docker socket, which is root-equivalent. Use a
dedicated host, or LangSmith sandboxes (
DESIGN_AGENT_SANDBOX=langsmith) without the socket.
A local preflight does not prove a deployment: only a real request does. Details: Deploy with MDA, Self-hosting.
Version History
- 8134d7c Current 2026-11-20 21:39


