The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override the agent's internal knowledge and training data. Specifically, SKILL.md states 'Critical: Do not trust internal knowledge', 'Everything you know about Mastra is likely outdated or wrong', and 'Never rely on memory. Always verify against current documentation.' While intended to ensure technical accuracy for a fast-evolving framework, these instructions follow the pattern of overriding previous system state or knowledge.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and process content from external sources, creating a vulnerability surface where malicious instructions could be embedded in the documentation.
Ingestion points: The agent is instructed to fetch documentation from https://mastra.ai/llms.txt and various subpages using the WebFetch tool, as described in references/remote-docs.md.
Boundary markers: Absent. There are no instructions to the agent to treat the fetched content purely as data or to ignore any embedded instructions within the remote documentation.
Capability inventory: The skill includes capabilities to read local files (cat, ls, grep), install packages (npm install), and execute local development servers (pnpm dev, mastra dev).
Sanitization: Absent. There is no mention of sanitizing or validating the content retrieved from remote URLs before processing.
[COMMAND_EXECUTION]: The skill references several command-line operations for project setup and maintenance, including npm install, npx @mastra/codemod, and starting local servers (pnpm dev). These are standard development workflows but represent a capability to execute code and modify the local environment.

mastra-framework-guide