The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of local binaries and scripts through the ida-cli tool, as well as IDA Pro's run_script and dbg_start_process capabilities. These are used for the intended purpose of analyzing and debugging binaries.\n- [DATA_EXFILTRATION]: Example scripts in the reference documentation demonstrate writing analysis artifacts, such as function lists and decompiled code, to the /tmp directory (e.g., /tmp/functions.json and /tmp/decompiled). This is documented as a standard practice for temporary storage of analysis results.\n- [PROMPT_INJECTION]: The skill contains instructional rules for the agent, such as 'F5 first, disasm second' and 'Rename-as-you-go', which are intended to ensure analysis quality and do not attempt to bypass agent safety filters.\n- [SAFE]: The skill processes untrusted binary data (ELF, PE, Mach-O) ingested via tools like open_idb and open_sbpf. While this creates a surface for indirect prompt injection, no malicious exploitation patterns were found. The skill's inventory of capabilities, including script execution and byte patching, are standard requirements for its stated reverse engineering purpose. Boundary markers for ingested binary data are absent but are consistent with the tool's specialized use case.

ida