The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill suggests using npx -y binary-ninja-mcp@1.0.0, which downloads and executes a third-party bridge from the NPM registry. While the package version is pinned and a specific review commit hash is provided, the source (fosdickio) is not an official vendor.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command using npx to start a local bridge between the Binary Ninja application and the agent's MCP interface.
[INDIRECT_PROMPT_INJECTION]: The skill processes outputs from a disassembler (strings, decompiled code, imports, and exports), which are derived from untrusted binary files. This is a potential vector for instructions hidden in data to influence agent behavior.
Ingestion points: Data enters the context through analysis commands such as list_strings, decompile_function, and list_imports in SKILL.md.
Boundary markers: No explicit markers or "ignore instructions" prompts are specified to wrap data returned from the binary analysis tools.
Capability inventory: The skill includes capabilities to perform byte patching, function renaming, and comment writing as documented in SKILL.md.
Sanitization: There are no mentioned mechanisms for sanitizing or escaping strings or code lifted from binaries before they are presented to the agent.

binary-ninja-reverse