The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is explicitly designed to process untrusted data in the form of malware samples (e.g., sample.exe). Analyzing the output of tools like strings or metadata from these files presents a surface for indirect prompt injection if the samples contain malicious instructions targeted at the LLM.
Ingestion points: Untrusted file samples (sample.exe) processed via strings, file, and rabin2 (SKILL.md).
Boundary markers: None present in the instructions to separate untrusted data from agent instructions.
Capability inventory: Execution of command-line tools such as file, sha256sum, strings, diec, exeinfope, rabin2, and dumpbin (SKILL.md).
Sanitization: No mention of escaping or sanitizing tool outputs before processing by the agent.
[COMMAND_EXECUTION]: The skill contains multiple command-line examples for file analysis and identification. While these are standard forensic tools, they involve executing subprocesses on the host system.
Evidence: Use of file, sha256sum, strings, diec, exeinfope, rabin2, and dumpbin in SKILL.md.
[PERSISTENCE]: The skill documents numerous persistence mechanisms used by malware, such as Registry Run keys (HKCU/HKLM\Software\Microsoft\Windows\CurrentVersion\Run), Scheduled tasks (schtasks), and Service creation (sc.exe). While listed for identification purposes, these are high-risk operations.
Evidence: Detailed list under the 'Persistence Mechanisms' section in SKILL.md.
[PRIVILEGE_ESCALATION]: The analysis workflow includes techniques that typically require administrative privileges, such as modifying services, scheduled tasks, and accessing system-level registry keys.
Evidence: References to sc.exe, schtasks, and HKLM registry paths in SKILL.md.

malware-analyst