The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze untrusted data from game binaries, decompiled source code, and memory strings. This creates an attack surface where malicious instructions hidden in game data could influence the agent's behavior.
Ingestion points: Decompiled C#, Java, and C++ code; hex dumps of asset files; strings extracted from live memory and binaries.
Boundary markers: No specific delimiters or instructions to ignore embedded prompts in the analyzed data are provided in the skill instructions.
Capability inventory: The agent is expected to use various MCP servers for Ghidra, IDA, and Cheat Engine, which likely have capabilities for file access, tool interaction, and script execution.
Sanitization: The instructions do not mention sanitizing or escaping the content retrieved from decompilers or memory scans.
[COMMAND_EXECUTION]: The skill recommends installing global command-line tools for reverse engineering.
Evidence: Recommends the execution of dotnet tool install -g ilspycmd for decompiling .NET binaries.
[DYNAMIC_EXECUTION]: The skill facilitates the use of tools that execute scripts during the reverse-engineering process.
Evidence: Mentions IDA Pro's "code mode" which executes IDAPython scripts and Frida for hooking functions using JavaScript.
[EXTERNAL_DOWNLOADS]: The skill suggests downloading and using various third-party modding and reverse-engineering tools.
Evidence: References a wide array of community tools such as UABEA, AssetRipper, FModel, and GhidraMCP. These are standard in the modding community but involve external dependencies.

reverse-engineering