The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes xcodebuild commands and a bundled Python benchmark script (scripts/benchmark_builds.py). These operations are necessary for performance verification. The Python script uses subprocess.run with argument lists, which is a secure implementation that prevents shell command injection.
[DATA_EXFILTRATION]: The skill uses git ls-remote --tags to verify version tags for remote Swift Package Manager dependencies. This is a read-only network operation used for legitimate dependency management and does not involve unauthorized data transmission.
[PROMPT_INJECTION]: The skill ingests instructions from external files like .build-benchmark/optimization-plan.md. This represents a surface for indirect prompt injection, where an attacker could theoretically embed malicious instructions in the plan to manipulate the agent's behavior during code modification or build steps.
Ingestion points: Reads optimization recommendations and approval statuses from .build-benchmark/optimization-plan.md and processes developer-supplied instructions.
Boundary markers: Relies on a checkbox-based approval system in the markdown plan but lacks explicit delimiters or instructions to ignore embedded commands within the descriptive text.
Capability inventory: The skill has the capability to modify Xcode project configuration files, shell-based build phases, and Swift source code, and can subsequently trigger builds and benchmarks.
Sanitization: The skill mitigates risks by referencing specific before/after fix patterns in references/fix-patterns.md and adhering to a predefined execution workflow.

xcode-build-fixer