The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill exposes the run_script tool, enabling the execution of arbitrary IDAPython code. This allows for dynamic code execution within the context of the IDA Pro process, which can be leveraged for unauthorized operations if the agent receives malicious instructions.
[EXTERNAL_DOWNLOADS]: The installation process involves downloading software from third-party, non-standard repositories, specifically a Homebrew tap and a Scoop bucket hosted on GitHub by an external maintainer ('blacktop').
[DATA_EXFILTRATION]: The skill provides tools like open_idb and read_bytes that can access any file or memory region accessible to the user running IDA Pro. Combined with the run_script capability, this facilitates the reading and potential exfiltration of sensitive data.
[PROMPT_INJECTION]: As the skill's primary function is to ingest and analyze untrusted binary data (malware, executables), it is susceptible to indirect prompt injection.
Ingestion points: Files are ingested via open_idb and processed by tools such as strings, list_symbols, and decompile.
Boundary markers: Documentation does not specify the use of delimiters or warnings to ignore instructions embedded in the analyzed binaries.
Capability inventory: The skill possesses significant capabilities including file system access (open_idb, read_bytes) and script execution (run_script).
Sanitization: No explicit sanitization or validation of the analyzed content is described before it is returned to the agent's context.

ida-mcp-headless-reverse-engineering