The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute npx -y xcodebuildmcp@latest mcp. This command downloads and runs the latest version of an external NPM package from the public registry at runtime. Because it is unversioned and from a non-whitelisted source, it introduces a supply chain risk where malicious updates to the package could compromise the agent's environment.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to https://mobbin.com/llms.txt using curl to fetch UI design flows and patterns. While this appears to be for legitimate design reference, it constitutes an external dependency on a non-whitelisted domain.
[COMMAND_EXECUTION]: The skill relies extensively on local shell commands to manage the development lifecycle, including xcodebuild for compilation, simctl for simulator automation, and devicectl for physical device interaction. It also involves piping raw JSON-RPC strings to xcrun mcpbridge to interact with Xcode's internal MCP bridge.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by ingesting and processing data from external sources:
Ingestion points: Processes UI tree snapshots from axe describe-ui and xcodebuildmcp, and parses HTML design mockups (mock.html) from the DesignSync MCP.
Boundary markers: None specified for the parsed mockup data or UI dumps.
Capability inventory: Extensive shell execution capabilities including project building, app installation, and simulator manipulation.
Sanitization: The skill uses grep and regex to extract specific values from mockups, but does not explicitly sanitize the larger context of the data it processes.

solo-ios-dev