The Agent Skills Directory

PROMPT_INJECTION (HIGH): The skill's primary function is to ingest untrusted external data from sources like Wikipedia, Hacker News, and Stack Exchange. This creates a significant surface for indirect prompt injection.
Ingestion points: Articles from x wkp, comments from x hn, and answers from x se enter the agent's context.
Boundary markers: No delimiters or "ignore embedded instructions" warnings are used when interpolating external data.
Capability inventory: The agent is authorized to execute shell commands via the x CLI, including configuration changes.
Sanitization: No sanitization or filtering of the fetched content is described, allowing malicious instructions in a Wikipedia edit or HN comment to potentially control the agent's next actions.
COMMAND_EXECUTION (HIGH): The skill operates by giving the agent the ability to execute arbitrary subcommands through the x CLI. If an injection attack occurs, the agent has a direct functional path to execute unintended shell commands.
EXTERNAL_DOWNLOADS (MEDIUM): The skill requires the x-cmd CLI to be pre-installed. x-cmd.com is not a trusted source according to the provided security guidelines, making the underlying toolset unverifiable and potentially dangerous.
DATA_EXFILTRATION (MEDIUM): The cfg and init subcommands for tools like ddgo and hn allow the agent to modify API endpoints and proxy settings. A sophisticated prompt injection could trick the agent into reconfiguring these tools to send data or authentication headers to an attacker-controlled server.

x-cmd-knowledge