The Agent Skills Directory

[COMMAND_EXECUTION]: The toolkit automates game builds and security verification (e.g., APK signing, macOS codesigning) by executing standard developer tools such as godot, apksigner, xcodebuild, and codesign via Python's subprocess module. These calls are parameterized using list-based arguments to prevent command injection and are restricted to local development tools.
[DYNAMIC_EXECUTION]: The skill uses dynamic execution for benchmarking and interoperability. In perf_bench.gd, hardcoded GDScript templates are compiled at runtime to create non-intrusive performance sentinels. Additionally, web_bridge.gd leverages JavaScriptBridge to facilitate communication between the game and the browser environment, using JSON.stringify to ensure data is passed safely.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it ingests and processes user-controlled project and scene files to generate audit reports.
Ingestion points: The toolkit reads project.godot and export_presets.cfg in gd_perf.py, and recursively audits scene tree structures in perf_audit.gd.
Boundary markers: Data is parsed into structured Dictionaries and JSON, though specific natural language boundary markers are not used when reporting findings to the agent.
Capability inventory: The skill has capabilities to execute shell commands, write to the project directory, and modify project settings.
Sanitization: All external data is parsed into engine-native types or sanitized via JSON encoding before being processed, maintaining the integrity of the agent's reporting loop.

godot-performance-export