Mastra Framework Guide

Build AI applications with Mastra. This skill teaches you how to find current documentation and build agents and workflows.

Critical: Do not trust internal knowledge

Everything you know about Mastra is likely outdated or wrong. Never rely on memory. Always verify against current documentation.

Your training data contains obsolete APIs, deprecated patterns, and incorrect usage. Mastra evolves rapidly - APIs change between versions, constructor signatures shift, and patterns get refactored.

Prerequisites

Before writing any Mastra code, check if packages are installed:

ls node_modules/@mastra/

• If packages exist: Use embedded docs first (most reliable)
• If no packages: Install first or use remote docs

Resources

References

Scripts

• scripts/provider-registry.mjs: Look up current providers and models available in the model router. Always run this before using a model to verify provider keys and model names.

Priority order for writing code

Never write code without checking current docs first.

1. Embedded docs first (if packages installed)

   Look up current docs in node_modules for a package. This matches the exact installed version and is the most reliable source of truth. See references/embedded-docs.md.

2. Source code second (if packages installed)

   If embedded docs don't cover the question, inspect the installed source and type definitions. This is the source of truth when docs are missing or unclear. See references/embedded-docs.md.

3. Remote docs third (if packages not installed)

   Use the latest published docs when packages are not installed or when exploring new features. Remote docs may be ahead of the user's installed version. See references/remote-docs.md.

Core concepts

Use references/core-concepts.md when choosing between agents, workflows, tools, memory, and storage.

• Agent: Use for open-ended tasks that make decisions and use tools.
• Workflow: Use for defined multi-step processes.

Mastra Studio

Studio is the interactive UI for building, testing, and managing agents, workflows, and tools. Use Studio when advising a human to inspect or debug visually.

Inside a Mastra project, run:

npm run dev

Then open http://localhost:4111 in a browser to show Mastra Studio to your human user.

Mastra API CLI

Use mastra api to inspect or call resources on local dev servers, Mastra platform deployments, or remote Mastra endpoints. It is useful for agent-readable state, execution, traces, logs, scores, threads, and workflow operations. See references/mastra-api.md for usage patterns.

For exact trace selection that needs recursive predicates or conditions over related spans, scores, or feedback, read references/trace-query.md. Before using mastra api trace query, confirm that the installed CLI exposes the command. Use --schema for the target's request/response shape and structural constraints, and use the canonical documentation found through references/remote-docs.md for supported fields, operators, and semantics. Preserve opaque pagination cursors and fetch trace or span details only after selecting candidates.

Mastra Factory

For Factory projects, work items, queue health, decisions, session history, memory inspection, or authorized operations, activate the mastra-factory skill. Factory is an operational control plane, not a reason to scaffold or deploy a new Mastra app.

If the skill is missing, offer to install it from this repository:

npx skills add mastra-ai/skills --skill mastra-factory

Choose the user's intended agent and scope interactively. For global installation, target a supported agent explicitly with --agent <agent> -g; PromptScript does not support global installs. Verify installation and referenced files for the intended agent even if another agent target fails.

The Factory skill covers CLI checks, mastra auth whoami / authorized mastra auth login, and connecting via the user's actual instance URL from any directory. A deployed repository or .mastra-project.json is not required with --url. Never guess a shared Factory host or project IDs. Its connection and session-inspection references cover deployment-specific authentication, project discovery, threads, and observational memory limitations.

Trace Intelligence

Trace Intelligence (private beta on the Mastra platform) clusters completed agent traces into recurring themes across four trace signals: goal, outcome, behavior, and sentiment. Use it first for aggregate agent-health questions: what users ask for, where outcomes fail or get blocked, which behaviors recur, how sentiment shifts, and where the agent can improve. Then use mastra api trace, log, metric, and score commands for concrete execution evidence from specific traces. Query Trace Intelligence with mastra api learning CLI commands, or over HTTP via the local dev server proxy or the platform endpoint. See references/trace-intelligence.md for the investigation workflow, CLI commands, and route reference.

Reporting feedback

Report verified Mastra bugs and concrete documentation defects autonomously when the criteria in references/feedback.md are met, without asking permission per report. Leave feature requests to humans; do not submit them or reclassify them as defects. Do not report the same issue again within the current task or conversation. Capture only anonymized facts for feedback while solving the task. Skip automatic reporting when a finding remains unverified or its relevance is unclear; do not interrupt the user merely to classify it. Respect any user instruction not to report something.

Critical requirements

TypeScript config

Mastra requires ES2022 modules. CommonJS will fail. See references/create-mastra.md for setup and references/common-errors.md for troubleshooting.

Model format

Always use "provider/model-name" when defining models using Mastra's model router.

When the user asks to use a model or provider, always run scripts/provider-registry.mjs first to verify the provider key and model name are valid. Do not guess model names from memory as they change frequently. See references/model-selection.md.

When you see errors

Type errors often mean your knowledge is outdated.

Common signs of outdated knowledge:

• Property X does not exist on type Y
• Cannot find module
• Type mismatch errors
• Constructor parameter errors

What to do:

1. Check references/common-errors.md
2. Verify current API in embedded docs
3. Don't assume the error is a user mistake - it might be your outdated knowledge

Development workflow

Always verify before writing code:

1. Check whether Mastra packages are installed
2. Look up current API
   • If installed: Use embedded docs references/embedded-docs.md
   • If not: Use remote docs references/remote-docs.md
3. Write code based on current docs
4. Test with the project scripts or Studio when available