The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references downloads for Ghidra from its official GitHub repository (NationalSecurityAgency/ghidra) and JDK versions from reputable providers like Temurin and Oracle. These are well-known and legitimate official sources for security and development tools.
[COMMAND_EXECUTION]: Includes standard shell commands for package management (apt, brew, choco, pip) and for executing Ghidra's headless analysis utility (analyzeHeadless). These commands are necessary for the skill's intended functionality and are used in a legitimate context.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external binary files (untrusted data) through decompilation and analysis, which creates an inherent attack surface.
Ingestion points: Binary files (e.g., PE, ELF, Mach-O) imported into the Ghidra environment for analysis as described in SKILL.md.
Boundary markers: The instructions explicitly warn in SKILL.md that decompilation results are inferences and should be verified against assembly listings to avoid misinterpretation.
Capability inventory: The skill utilizes shell command execution, Jython scripting within the Ghidra environment, and file system operations for exporting results.
Sanitization: Not present, as is standard for binary analysis tools; the skill instead provides warnings to mitigate the risk of the agent misinterpreting malicious binary content.

re-ghidra