The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands to manage build environments and run test artifacts.
Evidence: assets/test-lab/tools/build.py invokes docker run to build SGDK projects using the vendor-provided ewrogers/sgdk:latest image.
Evidence: assets/test-lab/tools/qa.py uses subprocess.run to call local C compilers (such as clang or gcc) and to execute the resulting test runner binaries.
[DYNAMIC_EXECUTION]: The skill generates and executes code at runtime to verify game logic consistency across different platforms.
Evidence: assets/test-lab/tools/qa.py compiles C source files (host.c, lab.c) into a native host binary for logic verification.
Evidence: assets/test-lab/tools/m68k.py utilizes the unicorn engine to emulate a M68000 processor, executing binary data from Sega Mega Drive ROMs for target-specific logic traces.
[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON data describing game input sequences, which constitutes a potential attack surface, though it is mitigated by strict validation.
Ingestion points: Replay tapes and trace results stored as JSON files (e.g., assets/test-lab/replays/courier.json).
Boundary markers: Data ingestion is strictly delimited by the JSON format and structural validation logic.
Capability inventory: The skill can execute shell commands (via subprocess.run) and emulated code (via unicorn).
Sanitization: assets/test-lab/tools/qa.py performs rigorous type and range checking on all incoming JSON fields before they are used in simulation or compilation headers.

sgdk-testing