The Agent Skills Directory

[COMMAND_EXECUTION]: The skill translates SQL-like queries into radare2 shell commands. The documentation explicitly mentions that assembly mnemonics in the assemble, search_asm, and patches tables are charset-validated (permitting only letters, digits, spaces, and specific symbols like ,.+-*/[]():%#_$) to ensure they are injection-safe before being spliced into internal command strings.
[INDIRECT_PROMPT_INJECTION]: As a binary analysis tool, the skill is designed to ingest untrusted data from external files (via strings, read_cstr, and data_items tables). This creates a surface where an agent could be exposed to adversarial instructions embedded within a malicious binary's strings. However, this is an inherent risk of the skill's primary purpose of reverse engineering and not a flaw in the skill's implementation.
[DATA_EXPOSURE]: The skill facilitates reading arbitrary bytes and memory segments from a target binary. While this involves sensitive data access, it is restricted to the scope of the file being analyzed by radare2 and aligns with the tool's intended use case.

data