The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to use npm exec to download and execute the latest versions of the mastra, @mastra/core, and typescript packages if the local CLI is unavailable. These are vendor-owned or standard technology packages. Additionally, it uses npx skills to fetch skill content from the vendor's GitHub repository.
[COMMAND_EXECUTION]: The skill relies on the mastra CLI for its primary operations. It also employs common system utilities like jq for processing JSON data, uuidgen for generating request identifiers, and open or xdg-open for opening URLs in the user's browser.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from external sources, including API responses and thread transcripts. It incorporates several defensive measures:
Ingestion points: Data enters the agent's context through CLI outputs such as work-item list and thread messages (found in references/session-inspection.md).
Boundary markers: The instructions emphasize using narrow jq filters to extract only necessary fields and limiting the amount of text displayed from message histories.
Capability inventory: The agent can execute shell commands, specifically the mastra CLI and local utilities.
Sanitization: The skill advises the agent to treat content from messages as unverified claims and to use structured JSON parsing to avoid misinterpreting truncated or malformed data.
[SAFE]: A dedicated 'Safety boundary' section in SKILL.md explicitly prohibits the agent from reading or exposing sensitive information, such as .env files, bearer tokens, or saved login credentials. It correctly directs the agent to use the CLI's built-in auth login flow for authentication.

mastra-factory