The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on various shell tools for static analysis, including wasm-objdump, wasm2wat, and wasm-decompile. It provides instructions for installing these tools via system package managers like apt, brew, and scoop, which may require administrative privileges. Furthermore, the skill's instructions suggest interpolating the user-provided $ARGUMENTS into shell commands, which can lead to command injection if not properly sanitized.- [EXTERNAL_DOWNLOADS]: The skill includes instructions to download WASM modules from remote URLs using curl. It also references the installation of the WABT toolkit from its official GitHub repository and recommends installing several Python packages for WASM execution.- [REMOTE_CODE_EXECUTION]: The skill provides implementation templates for executing untrusted WASM modules using the wasmer and wasmtime libraries in Python, as well as the native WebAssembly API in Node.js. Running untrusted binaries is an inherent risk in dynamic analysis.- [PROMPT_INJECTION]: As the skill is designed to ingest and process untrusted external files (WASM modules), it is vulnerable to indirect prompt injection. Maliciously crafted data within the analyzed files (such as strings or metadata) could attempt to influence the agent's behavior during the analysis process. Ingestion points: untrusted .wasm files downloaded via curl or read from disk. Boundary markers: absent. Capability inventory: shell execution, file system access, and network operations. Sanitization: none documented.

wasm-reverse