Build with a coding agent and the official skill

Separate the layers

An API calls the model. A skill supplies instructions and implementation patterns. MCP connects tools to compatible clients. Installing a skill does not grant API access or guarantee every agent can execute it.

Official installation options

For supported agent environments:

npx skills add typesafe-ai/skills --skill typesafe-ai

Claude Code users can instead choose the plugin route:

claude plugin marketplace add typesafe-ai/skills
claude plugin install typesafe@typesafe-ai

Inspect installation contents and scope before confirming. Do not label a community MCP server official or claim unverified client compatibility.

A task brief to copy

Use the official TypeSafe skill to build a local email classifier.
First define inputs, questions, output schema and execution boundaries.
Read only my redacted samples. Read the key from the environment, never source code.
Pin a model version that I approve. Run offline tests before I approve live API calls.
Return categories, usage and review records. Do not send emails, edit orders or issue refunds.
Errors, timeouts, uncertainty and missing evidence must go to human review.
Include a README, dependencies and test commands.

Instructions are not enforcement

A sentence asking an agent to follow every step is not an execution guarantee. Implement state transitions, required-field validation, authorization, approvals and stop conditions in code. The skill can explain those gates; it cannot replace them.