The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs standard Apple development tools such as xcodebuild, simctl, and xcresulttool for legitimate tasks including building, testing, and deploying iOS applications. These operations are conducted within the context of established developer workflows.
[EXTERNAL_DOWNLOADS]: The skill references XcodeBuildMCP, a reputable tool from a known vendor (Sentry). All other external references point to official Apple documentation, Swift community forums, or well-known developer platforms, adhering to the trust scope guidelines.
[CREDENTIALS_SAFE]: Code scaffolds for Cloudflare Worker backends correctly utilize environment variables and secret management practices (e.g., env.LLM_API_KEY, wrangler secret put) rather than hardcoding sensitive credentials.
[INDIRECT_PROMPT_INJECTION]: The skill has a standard potential attack surface as it processes external source code and tool logs. However, it implements a 'Proof-First' execution loop requiring visual verification (screenshots) and artifact-based validation, which significantly reduces the risk of accidental obedience to embedded instructions.
Ingestion points: Analyzed application source code, xcodebuild logs, and simulator UI snapshots.
Boundary markers: Explicit 'Proof-First' rules requiring the agent to verify tool reality and provide proof artifacts for every slice of work.
Capability inventory: Shell execution (xcodebuild), file system access (scripts/), and network operations (fetch in backend scaffolds).
Sanitization: Relies on system-level LLM guardrails and mandatory human/agent review of generated visual and diagnostic artifacts.

software-ios-native