Menu
Route a new Trellis user through the Cloudflare-first path without making them understand the whole stack first.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Build or modify a Trellis agent by editing the agent blueprint, skills, knowledge, state map, MCP surfaces, and verification commands.
Connect Claude Code, Codex, Cursor, or OpenCode to a local or deployed Trellis Cloudflare app over MCP.
Connect Trellis providers as Cloudflare Worker secrets one provider at a time.
Scaffold a new Cloudflare-first Trellis app and hand the user to readiness or deploy.
Take a first-time Trellis user from local proof to the simplest credible Cloudflare production deployment.
Diagnose what is missing before a Trellis Cloudflare app can boot, deploy, or verify live.
| name | trellis-onboarding |
| description | Route a new Trellis user through the Cloudflare-first path without making them understand the whole stack first. |
Use this skill when a user is new to Trellis and wants the shortest path to a working Trellis demo or app.
Trellis is a vertical GTM agent stack. Flue is hidden behind Trellis. Cloudflare is the default runtime and deploy target.
Keep the explanation simple:
src/agent.ts is the Trellis app blueprintknowledge/ is contextskills/ is behaviorwrangler.jsonc is the Cloudflare binding map/mcp/trellis is the agent control and inspection surface/mcp/operator is the runtime control-plane surface when the app exposes a role-shaped MCP like trellis-sdrDo one thing at a time:
Prefer:
Next: create the Cloudflare appNext: run cf:loginNext: run doctorNext: run smokeNext: deployNext: verify cloudflareNext: connect remote MCPBlocked: Cloudflare auth is missingBlocked: TRELLIS_API_KEY is missing for protected remote routesUnless the user explicitly wants a different path, route them through this order:
Say the next step in these exact verbs when possible:
Next: create appNext: use Node 22Next: installNext: authenticate CloudflareNext: run doctorNext: run smokeNext: deployNext: verifyNext: connect remote MCPNext: run one safe demoUse these skills based on the user's next need:
trellis-create-app
trellis-readiness-check
trellis-setup-database
trellis-stack-explainer
trellis-build-agent
src/agent.ts, src/state/*.map.ts, knowledge/, skills/, or MCP tool surfacestrellis-connect-providers
trellis-connect-mcp
trellis-first-deploy
Treat the Trellis CLI as the execution contract. Prefer machine-readable commands:
npm run trellis -- init <target-dir> --name <app-name> --json
npm run trellis -- doctor --json
npm run trellis -- smoke --json
npm run trellis -- deploy --json
npm run trellis -- verify cloudflare --json
npm run trellis -- verify cloudflare --live --url <worker-url> --api-key "$TRELLIS_API_KEY"
npx wrangler d1 execute <database-name> --remote --command "SELECT name FROM sqlite_master WHERE type = 'table' AND name LIKE 'trellis_%' ORDER BY name"
npm run trellis -- connect <provider> --json
npm run trellis -- docs add <path> --json
npm run trellis -- mcp claude-code --remote --write --json --url "${APP_URL}/mcp/trellis" --token "$TRELLIS_API_KEY"
Do not route new users through Convex, Vercel, Railway, or legacy AI SDR commands. Do not scrape pretty terminal output if JSON is available.
When the user asks to build an agent, use trellis-build-agent and make these concepts explicit:
src/agent.ts defines the Trellis workflow.src/state/*.map.ts defines business state projections.trellis-setup-database handles D1 creation, binding, schema proof, seed data, and safe demo cleanup.skills/**/SKILL.md files are bounded methods, not the entire agent.knowledge/**/*.md files are fetched into runtime context through R2 packs./mcp/trellis should be the role-facing surface such as trellis-sdr./mcp/operator should be the platform/operator surface such as trellis-operator.