The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Downloads the CAPEv2 source code from a third-party repository (https://github.com/kevoreilly/CAPEv2) and installs dependencies via pip.
[COMMAND_EXECUTION]: Utilizes system utilities including VBoxManage, docker, vmrun, and firejail to manage virtualization, containers, and application sandboxing.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Malware samples are ingested and executed within sandboxed environments for behavior monitoring (SKILL.md, Step 4).
Boundary markers: The skill emphasizes operational isolation (e.g., Host-only networking) but lacks specific prompt boundary markers to prevent the agent from being influenced by instructions embedded within the samples.
Capability inventory: Employs capabilities to execute code within isolated guest environments using tools like VBoxManage guestcontrol and docker exec.
Sanitization: No sanitization or filtering protocols are described for processing untrusted data extracted from the analyzed samples before returning it to the agent's context.

re-sandbox