The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands for local directory management and network operations. It utilizes mkdir to establish the directory structure and rm to clear existing content when requested by the user. Additionally, it executes curl to download image assets from remote sources.
[EXTERNAL_DOWNLOADS]: The skill downloads image assets from Miro's infrastructure using curl. The source URLs for these downloads are dynamically provided by the Miro Model Context Protocol (MCP) toolset during the extraction process.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external Miro boards, which are untrusted sources and may contain malicious instructions or hidden payloads.
Ingestion points: Untrusted data enters the agent context through the Miro MCP tools (board-overview, item-retrieval, and table-retrieval) which fetch content directly from active Miro boards.
Boundary markers: The skill implements a subagent architecture for processing large HTML content from prototype screens. This serves as a boundary to isolate potentially large or complex external payloads within a secondary context, preventing them from overwhelming or manipulating the main agent's session.
Capability inventory: The skill possesses the capability to write to the local filesystem, manage directories via bash, and perform network downloads using curl.
Sanitization: Content is written to local files as retrieved from the Miro tools; the skill relies on the subagent isolation mechanism to manage the risks associated with processing external content rather than explicit text sanitization.

miro-code-spec