The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script assets/enemy-lab/tools/check_enemy.py performs runtime compilation and execution of C code on the host system.
It uses the subprocess module to invoke a C compiler (e.g., clang, cc, or gcc) on the source files enemy.c and test_enemy.c.
The resulting binary is then executed to perform work-bound and behavior checks for the AI engine.
[EXTERNAL_DOWNLOADS]: Build instructions in assets/enemy-lab/README.md require pulling the Docker image ewrogers/megadev:latest from a public registry.
This image is provided by the author of the skill and is used for building and mastering the game discs.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates an environment where the agent generates C code that is subsequently executed locally, creating a vulnerability surface.
Ingestion points: The agent ingests game logic, enemy rules, and navigation requirements to implement AI states in C code, as described in SKILL.md.
Boundary markers: No explicit markers or warnings are used to instruct the agent to disregard instructions potentially embedded in the game data or rule definitions it processes.
Capability inventory: The skill provides a script (check_enemy.py) with the capability to compile and run the generated C code via subprocess calls.
Sanitization: There are no sanitization or validation steps to ensure that the generated C code is safe before it is compiled and executed on the host.

megadev-ai