The Agent Skills Directory

[COMMAND_EXECUTION]: The run_cmd function in scripts/agent.py uses subprocess.run(cmd, shell=True). Throughout the script, this function is called with shell commands constructed using f-string interpolation of variables like image_path, path, inode, and keyword. A malicious user or an agent processing a maliciously crafted disk image (where file names contain shell control characters) could trigger arbitrary command execution. For example, in search_keywords, the keyword argument is piped directly into grep within a shell string.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests untrusted data from disk images (via fls and mmls command output) and parses it without sanitization or boundary markers. A disk image containing malicious file names or metadata could influence the agent's behavior or exploit the command execution vulnerability mentioned above.
[EXTERNAL_DOWNLOADS]: The SKILL.md file provides instructions for downloading forensic software and reference data from official repositories, specifically the Sleuth Kit releases on GitHub and the NIST NSRL hash database on Amazon S3. These are legitimate resources for the skill's stated purpose.

analyzing-disk-image-with-autopsy