The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_ios_component.sh utilizes the eval command to assign variables based on user input in the prompt_input and prompt_select functions. This implementation is vulnerable to shell command injection; if the agent provides an input containing command separators (e.g., ;, &&) or backticks, arbitrary code can be executed in the environment.
[DYNAMIC_EXECUTION]: The skill provides a component generation utility (scripts/generate_ios_component.sh) that dynamically constructs SwiftUI and UIKit source files. It uses sed to replace placeholders with user-provided names and feature lists without escaping special characters. This allows for 'sed injection,' which can lead to corrupted source files or the injection of unexpected code into the generated Swift components.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it reads and analyzes external Swift source files.
Ingestion points: scripts/audit_accessibility.sh and scripts/validate_design.sh read content from files or directories provided at runtime.
Boundary markers: No clear delimiters or instructions are present to prevent the agent from following instructions potentially embedded in the files being audited.
Capability inventory: The skill can write new files to the filesystem and execute shell scripts.
Sanitization: The auditing scripts perform simple grep operations and return raw matches without escaping.

apple-hig-designer