The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to analyze untrusted binary data, which provides a significant attack surface for indirect prompt injection.
Ingestion points: Untrusted binaries are ingested via the currentProgram API and processed by analysis scripts in SKILL.md and references/headless-analysis.md.
Boundary markers: The provided script templates lack delimiters or explicit instructions to prevent the agent from interpreting extracted strings (e.g., in ExtractStrings.py) as commands.
Capability inventory: The skill utilizes capabilities for writing analysis results to the local filesystem and performing binary patching in memory.
Sanitization: Extracted string and function data is processed and output without sanitization or escaping, increasing the risk of the agent obeying embedded instructions inside the binary.
[COMMAND_EXECUTION]: The documentation provides multiple shell command examples for both interactive and automated use of Ghidra.
Evidence: Usage patterns for ghidraRun, analyzeHeadless, and pyghidraRun across all files.
[EXTERNAL_DOWNLOADS]: The skill references several external resources for its installation and extended functionality.
Evidence: Points to the official National Security Agency Ghidra repository for the core suite and third-party GitHub repositories for analysis enhancements like Sekiryu and ghidriff.

ghidra