Templates

Overview

Templates are opt-in showcase agents the scaffold can drop into your project alongside the default assistant-agent. Each one demonstrates a specific Mastra pattern — supervisor delegation, durable scheduling, workspace + skills — wired all the way through MastraKit's auth, multi-tenancy, and observability so you can see the pattern running end-to-end before adapting it.

You pick templates during npx mastrakit (each is a separate prompt, default off). They install into apps/mastra/src/mastra/ next to your other agents and tools — no separate runtime, no extra deploy.

Available templates

More templates ship over time. The list above tracks what's wired into the current CLI release.

Daily Briefing demo agent

A supervisor that decomposes a briefing request — "summarise these sources for me each morning" — into two specialised subagents and produces a polished markdown digest archived in the user's workspace.

What gets installed

When you answer yes to the "Include the Daily Briefing demo agent?" prompt, the scaffold adds:

apps/mastra/src/mastra/
├── agents/
│   ├── briefing-supervisor.ts   # orchestrator
│   ├── research-agent.ts        # subagent: URL → bullet-point facts
│   └── writer-agent.ts          # subagent: bullets → markdown digest
├── lib/
│   └── workspace.ts             # per-user AgentFS over Turso
└── tools/
    ├── http-fetch.ts            # SSRF-hardened HTTP fetch
    └── approve-draft.ts         # promotes a draft into /briefings/

The supervisor registers as briefing-supervisor in the Mastra agents map, so it shows up in the agent picker at http://localhost:4111 and in your web chat UI.

Patterns it demonstrates

• Supervisor delegation — the supervisor exposes the two subagents via the agents map and decides which to call. Subagents have no overlapping tools, so the routing decision stays clean.
• Background tasks — research-agent is opted into background dispatch. The supervisor's first response ("I'm gathering your briefing now…") streams immediately while research fans out across multiple http-fetch calls.
• Per-user workspace — every user gets an isolated filesystem under users/{userId}/ backed by Turso, exposed to the agent as /skills/, /drafts/, and /briefings/. See workspace.ts for the per-user prefix logic.
• Skills — when the user expresses an ongoing preference ("write in Dutch", "skip crypto"), the supervisor persists it as a SKILL.md file under /skills/. Skills auto-load on every future run, even from a fresh chat.
• BM25 search — the supervisor can search over indexed /briefings/ to answer "what did you tell me about X last week?" with a grounded summary.
• Drafts approval flow — for sensitive deliveries, the supervisor writes to /drafts/ first, asks for confirmation, and only promotes to /briefings/ after the user approves via the approve-draft tool.
• Guardrails — PromptInjectionDetector on URL content, PIIDetector redaction on the final digest, TokenLimiterProcessor to cap cost.
• SSRF hardening — http-fetch blocks loopback, private, and cloud-metadata addresses (full fe80::/10 and fc00::/7 ranges) before issuing the request.

Usage

After scaffolding, start the stack with pnpm dev, open the web chat, and pick the briefing-supervisor agent.

> Summarise https://news.ycombinator.com and https://simonwillison.net for me

The supervisor will:

1. Confirm the sources
2. Dispatch research-agent in the background — first chunk arrives in ~1s
3. Wait for research to complete (~30s–2min depending on source count)
4. Delegate to writer-agent for the final markdown digest
5. Archive the digest at /briefings/{date}T{time}Z.md in the user's workspace
6. Stream the result back

Skills flow

Mid-conversation:

> From now on, write briefings in Dutch and keep them under 5 bullets.

The agent will skill_search for an existing match → no result → write_file two skills (/skills/language-dutch/SKILL.md and /skills/short-format/SKILL.md) → confirm. The next briefing applies both preferences automatically.

Search flow

> What did you tell me about Anthropic last week?

The agent calls search over the indexed /briefings/, reads the matching files, and answers with citations to the actual archived briefings.

Scheduled briefings

The supervisor exposes schedule_briefing and list_scheduled_briefings tools so users can ask "send me this digest every weekday at 9am". Scheduled tasks are gated by plan (see Scheduled Tasks).

Note: scheduled ticks currently route through the default assistant-agent, not briefing-supervisor. Tracked for V1 polish — see MAK-242.

Customization

The most useful starting points for tweaks:

• Model selection — swap providers in apps/mastra/src/mastra/lib/models.ts
• Guardrail thresholds — raise the PII threshold or change GUARDRAIL_MODEL_ID in apps/mastra/src/mastra/lib/guardrails.ts
• Source tools — http-fetch is intentionally minimal. Add a Tavily search tool, RSS reader, or Browserbase-driven browser tool to give research-agent richer inputs
• Delivery — the digest streams to chat. Wire it to email/Slack by intercepting the supervisor's final response, or by adding a workflow that calls a delivery tool after archiving (V1 polish, MAK-244)

Standalone vanilla template

If you want the briefing pattern without MastraKit's auth/billing/observability — for example, to publish your own variant or run it standalone — there's a vanilla extract at templates/briefing-agent/ in the MastraKit repo. It's published to mastra.ai/templates and runs as a single Mastra dev server with a local SQLite workspace, no auth, no scheduling.

The vanilla template is not a drop-in equivalent of the MastraKit-installed variant — it deliberately ships a smaller surface area. The two differ on purpose:

If you only need the supervisor-with-subagents pattern, the vanilla template is the right starting point. If you need the full drafts/skills/scheduling experience documented above, scaffold with the MastraKit CLI and answer "yes" to the prompt instead.

Skipping a template later

If you scaffolded with a template and want to remove it, delete the agent files, drop the briefing-supervisor entry from the agents map in apps/mastra/src/mastra/index.ts, and remove any related env vars from .env. There's no programmatic uninstall — templates are plain TypeScript once they're in your repo.