The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to install and use third-party tools from PyPI, specifically ghidra-ai-bridge and re-agent (by author 'Dryxio'), which are not from trusted or well-known organizations.\n- [COMMAND_EXECUTION]: The workflow involves significant use of command-line tools for installation (uv tool install), MCP configuration (claude mcp add), and running automated reverse-engineering pipelines (re-agent reverse).\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data extracted from compiled binaries, such as strings and decompiled code, which could contain adversarial instructions. \n
Ingestion points: Untrusted data enters the agent context through ghidra-bridge and re-agent output files located in the .ghidra-exports/ directory. \n
Boundary markers: The skill includes instructions to quote addresses and label inferences, which helps distinguish tool output from agent reasoning, though these are not strict security boundaries. \n
Capability inventory: The skill uses CLI tools (ghidra-bridge, re-agent) and has the ability to read and potentially write files in the project directory. \n
Sanitization: Instructions specify bounding tool output with head or jq to manage context size, but there is no specific sanitization of binary-derived strings or code.\n- [DYNAMIC_EXECUTION]: The skill assembles and executes shell commands using dynamic inputs like memory addresses and function names (e.g., ghidra-bridge decompile <addr>), which is a form of dynamic command construction.

reverse-engineer-binaries