The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's common script scripts/_xcodebuildmcp_common.sh attempts to resolve the xcodebuildmcp tool by calling npx -y xcodebuildmcp@latest. This fetches the latest version of the tool from the npm registry at runtime if it is not installed locally. Additionally, data/sources.json and SKILL.md reference external developer utilities including xcbeautify (GitHub), DocSetQuery (GitHub), and sosumi.ai (a documentation search service).
[REMOTE_CODE_EXECUTION]: The resolution of xcodebuildmcp via npx constitutes remote code execution, as the package is downloaded and run immediately. However, xcodebuildmcp is an official tool from Sentry, a well-known technology company, and its use here is consistent with standard developer workflows for simulator automation.
[COMMAND_EXECUTION]: The skill provides several shell scripts (scripts/build-ios.sh, scripts/run-ios.sh, scripts/test-ios.sh, scripts/capture-screenshot.sh) that execute xcodebuildmcp commands to perform builds, launch applications, run tests, and capture simulator screenshots. These commands are integral to the skill's stated purpose of design verification.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core workflow of processing external project data.
Ingestion points: The agent is instructed to ingest SwiftUI source code, build logs, and simulator UI snapshots for auditing (documented in SKILL.md and references/ai-design-review.md).
Boundary markers: There are no specific delimiters or instructions provided to the agent to distinguish between its own system instructions and potentially malicious commands embedded in the code or logs it audits.
Capability inventory: The skill possesses the capability to execute shell commands, compile code, and perform network requests via the integrated developer tools.
Sanitization: No sanitization or filtering logic is present for the ingested project data.

software-ios-design