The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include shell commands utilizing elevated privileges (sudo) for critical environment setup, such as creating network interfaces (ip tuntap), configuring routes (ip route), and modifying local host resolution files (tee -a /etc/hosts). These operations are necessary for establishing the required VPN tunnel and network connectivity to the training lab environment.
[EXTERNAL_DOWNLOADS]: The skill references an external documentation repository on GitBook (0xb0b.gitbook.io) as a primary source for detailed lab walkthroughs and command payloads. It encourages fetching dynamic content from this source using a query parameter.
[PROMPT_INJECTION]: A vulnerability surface for indirect prompt injection is present due to the integration of external data from a third-party source. The agent is instructed to fetch and process content from an external GitBook page, which could potentially contain malicious instructions if the source content were to be tampered with.
Ingestion points: The Querying the writeups live section in SKILL.md identifies that external data enters the agent context through queries to the GitBook API/endpoint.
Boundary markers: No explicit markers or specific instructions to disregard embedded commands in the external data are present.
Capability inventory: The skill leverages an extensive array of high-capability offensive tools, such as impacket, netexec, responder, and evil-winrm, which could be abused if an injection attack succeeded.
Sanitization: There is no evidence of validation or sanitization of the fetched external content before the agent interprets or processes it.

hacksmarter-labs