The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically generates C++ source code (ps2_pad.cpp) based on templates and user-provided JSON configuration. This code is intended to be compiled into the PS2Recomp runtime. A minor potential for code injection exists if the configuration file itself has a malicious name, as the filename is included in C++ comments without sanitization of newline characters, which could allow an attacker to break out of the comment and inject executable code.
[COMMAND_EXECUTION]: The skill provides the padmap.py script which performs file system operations, including writing configuration files, generating C++ code, and installing the generated code into a local repository path. It manages backups of original files and verifies that the files it overwrites contain a specific generation marker.
[INDIRECT_PROMPT_INJECTION]: The skill processes external configuration data from controls.json and input replay scripts. This represents an attack surface for processing untrusted data that could influence the agent's actions or the generated code's behavior.
Ingestion points: console/controls.json (JSON configuration) and user-specified replay scripts (plain text).
Boundary markers: None present to distinguish user configuration from system instructions.
Capability inventory: File write/overwrite via the install command, code generation via the gen command, and filesystem navigation through repo path arguments.
Sanitization: The tool employs robust JSON parsing and strict key/value validation against predefined allow-lists for controller buttons, keys, and layout names, significantly limiting the impact of malicious configuration values.

console-controller