The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install a package directly from a third-party GitHub repository using pip install git+https://github.com/mrexodia/ida-pro-mcp.git. This practice results in the execution of remote code during the installation phase from an unverified source.
[COMMAND_EXECUTION]: The instructions include the execution of local system commands ida-pro-mcp --install and ida-pro-mcp --config to set up and verify the IDA Pro bridge, which allows for direct interaction with the host operating system.
[DYNAMIC_EXECUTION]: The skill provides a py_eval tool (detailed in cheatsheet.md) that enables the agent to execute arbitrary Python scripts within the IDA Pro process, bypassing standard tool constraints.
[DATA_EXFILTRATION]: The skill encourages the access and manipulation of sensitive system paths, specifically advising the user to copy files from C:\Windows\System32 for analysis, which exposes critical system-level data.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect prompt injection due to its core function of processing untrusted binary data.
Ingestion points: Loading PE/ELF/Mach-O/DLL/SYS binaries into the agent's context via idb_open and subsequent analysis tools described in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious embedded strings or function names within the processed binaries.
Capability inventory: The skill combines high-privilege capabilities including arbitrary Python execution (py_eval), binary patching (patch_asm), and system command execution (pip install, ida-pro-mcp --install).
Sanitization: While the skill suggests using an isolated VM for untrusted binaries, this is a procedural recommendation and does not programmatically sanitize the data processed by the agent.

reverse-eng-ida