The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The installation instructions utilize sudo apt install to acquire system-level libraries and extraction utilities (e.g., mtd-utils, squashfs-tools). In an autonomous agent context, this requires elevated permissions.
[COMMAND_EXECUTION]: The skill relies heavily on shell command execution for its core functionality, including binwalk, find, grep, dd, and strings to perform binary analysis and file system extraction.
[EXTERNAL_DOWNLOADS]: The skill downloads the Binwalk v3 tool using cargo install (Crates.io) and provides links to official GitHub releases. These are well-known services used for legitimate tool acquisition.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data (firmware blobs). This creates a vulnerability surface where an attacker could embed malicious text or instructions within a binary's strings or metadata. If the agent parses the output of tools like strings or grep without proper boundary markers or sanitization, it could be influenced by the embedded data.
Ingestion points: Binary firmware files (e.g., firmware.bin) processed in SKILL.md.
Boundary markers: Absent; commands do not wrap tool output in delimiters or instructions to ignore embedded content.
Capability inventory: Includes subprocess execution of binwalk, find, grep, dd, strings, and file across SKILL.md and references/firmware-analysis.md.
Sanitization: Absent; the skill does not suggest escaping or filtering the content extracted from binaries before presentation to the agent.

binwalk