The Agent Skills Directory

[COMMAND_EXECUTION]: The skill heavily utilizes shell commands, specifically curl, to interact with remote network targets. These commands interpolate variables like $TARGET and COLLAB_HOST to perform fingerprinting and vulnerability scanning.
[DATA_EXFILTRATION]: The skill includes instructions to access sensitive file paths on the remote target, such as /proc/self/environ, /proc/self/cmdline, and /proc/self/cwd, which can contain credentials or environment configuration. It also demonstrates patterns for exfiltrating data to an external collaborator host (OOB) to confirm successful exploitation.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes responses from untrusted external servers.
Ingestion points: Data retrieved from external servers via curl requests (HTTP headers and response bodies) in SKILL.md.
Boundary markers: Absent; there are no delimiters or instructions to treat external data as untrusted or to ignore embedded commands.
Capability inventory: Shell execution and network access via the curl tool.
Sanitization: Absent; the content retrieved from the network is piped into grep or reviewed by the agent without escaping or validation.

hunt-nodejs