The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents the use of PRAGMA idasql.enable_idapython = 1 and the idapython_snippet function to execute arbitrary Python code within the IDA Pro environment. This is a core feature for advanced analysis but allows for unrestricted code execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted binary data, which serves as a primary attack surface for indirect prompt injection. Malicious strings or comments embedded in a binary could potentially override agent instructions during analysis.
Ingestion points: Data is pulled from the strings, funcs, comments, and pseudocode tables, as well as the decompile() function.
Boundary markers: While output is structured as JSON, there are no explicit instructions for the agent to ignore natural language commands found within the binary data.
Capability inventory: The skill has access to the Bash tool and can perform file system operations via load_file_bytes and bytes table updates.
Sanitization: No sanitization or filtering of content retrieved from the database is mentioned.
[COMMAND_EXECUTION]: The skill relies on the Bash tool to execute the idasql CLI for database operations, server management, and binary analysis tasks.
[DATA_EXFILTRATION]: The skill includes functionality to start HTTP and MCP servers that can be configured to bind to public network interfaces (--bind 0.0.0.0). This capability, combined with access to sensitive binary analysis data, could lead to data exposure if authentication tokens are not used or are compromised.

connect