The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses eval on the output of xcodebuild -showBuildSettings to dynamically set shell variables (DIR and NAME) in both run-simulator/SKILL.md and run-device/SKILL.md. This pattern of executing strings generated from project metadata could be exploited if a project target or path contains malicious shell characters.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading and auditing user-provided Swift and SwiftUI code files using tools like Glob, Grep, and Read. It then performs actions such as building and running the code via Bash. This creates a surface for indirect prompt injection attacks:
Ingestion points: Local project files and build configurations (detected via Glob, Grep, and Read tools in run-simulator/SKILL.md, run-device/SKILL.md, and ui-review/SKILL.md).
Boundary markers: None identified in the prompt instructions to delimit untrusted code content from instructions.
Capability inventory: Significant command execution capabilities via Bash (building with xcodebuild, managing simulators with simctl, and hardware with devicectl) and network capabilities via WebFetch.
Sanitization: No explicit sanitization or validation of the content read from the user's project files before it is processed or used in shell commands.
[COMMAND_EXECUTION]: The skill uses Bash to execute several local development tools including xcodebuild, xcrun simctl, and devicectl. While these are necessary for its stated purpose of building and running iOS applications, they provide control over the local system processes.
[EXTERNAL_DOWNLOADS]: The skill documentation suggests installing libimobiledevice via Homebrew (brew install) and fetches documentation from official developer domains. These references target well-known services relevant to the skill's domain.

ios-development