The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions that direct the agent to prioritize current documentation over its internal training data. This is a common safety and accuracy pattern used to ensure the agent uses up-to-date framework APIs and patterns.
[COMMAND_EXECUTION]: The skill provides patterns for the agent to execute shell commands such as ls, cat, and grep. These are used to navigate the project structure and read documentation files or type definitions located within the node_modules/@mastra/ directory.
[REMOTE_CODE_EXECUTION]: The instructions involve standard package management operations such as npm install, npm create mastra@latest, and npx @mastra/codemod. These commands are used to initialize projects, install dependencies, and perform version migrations using official registry resources.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for processing external content by fetching documentation from https://mastra.ai/llms.txt.
Ingestion points: Documentation is fetched via remote URL in references/remote-docs.md.
Boundary markers: The instructions do not define specific delimiters for the external content.
Capability inventory: The skill allows subprocess calls for environment setup and project building (e.g., npm, npx) as described in references/create-mastra.md.
Sanitization: Content is processed as markdown documentation without explicit validation or filtering steps.

mastra