The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions include the use of sudo to perform administrative network operations. This includes ARP spoofing via arpspoof, modifying system routing tables with route, and performing raw packet captures using tcpdump.
[COMMAND_EXECUTION]: The skill requires the agent to invoke a broad suite of external command-line security tools to perform binary triage and forensics, such as tshark, capa, volatility3, pe-sieve, and strace.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to install third-party packages from PyPI (e.g., pip install dissect.cobaltstrike) and provides links to multiple external GitHub repositories for specialized analysis scripts and unpackers.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted malware artifacts, which constitutes a significant attack surface for indirect prompt injection.
Ingestion points: The agent reads and processes strings, binary metadata, PCAP traffic, and memory contents extracted from potentially malicious external samples.
Boundary markers: While the skill advises the use of isolated environments, it does not prescribe specific prompt delimiters or "ignore instructions" wrappers when processing extracted data.
Capability inventory: The agent is provided with high-privilege capabilities including network manipulation, file system access, and the ability to install and execute external tools.
Sanitization: There are no explicit instructions to sanitize or validate strings and data recovered from analysis targets before they are handled in a prompt context.

malware-ctf