The Agent Skills Directory

[PERSISTENCE]: The skill includes scripts/install-autostart.ps1, which registers a Windows Scheduled Task named reverse-skill-ida-mcp. This task is configured to execute a watchdog script at logon and every minute thereafter to ensure the MCP server remains active, which is a standard persistence technique.
[DYNAMIC_EXECUTION]: The skill explicitly enables and promotes the use of the idapro_py_eval tool, which allows the execution of arbitrary Python code in the context of IDA Pro. The server is intentionally started with the --unsafe flag to permit this dynamic code execution.
[EXTERNAL_DOWNLOADS]: The skill instructions and scripts (scripts/start.ps1) facilitate the installation of the ida-pro-mcp package directly from a third-party GitHub repository (github.com/mrexodia/ida-pro-mcp) using pip install. This repository is not part of the trusted vendor list.
[COMMAND_EXECUTION]: Multiple PowerShell scripts (start.ps1, open.ps1, watchdog.ps1) perform system-level operations, including process management via Win32_Process, command execution, and file manipulation within sensitive system directories like System32.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze untrusted binary files (EXE, DLL, SO, etc.), which serves as an ingestion point for potentially malicious data. The instructions lack boundary markers or sanitization logic to prevent the agent from interpreting strings or metadata within these binaries as instructions.
Ingestion points: Binary files processed by scripts/open.ps1 and idapro_survey_binary.
Boundary markers: Absent.
Capability inventory: Includes file modification (idapro_patch), process management (scripts/start.ps1), and arbitrary code execution (idapro_py_eval).
Sanitization: None provided for extracted binary content.

ida-reverse