The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large amounts of untrusted data from external websites (Phase 1.6, 7.1, 8.1, etc.). Maliciously crafted content on a target domain could attempt to influence the agent's logic or prompt behavior.
Ingestion points: Data enters the agent's context through tools like curl, katana, and waybackurls which retrieve page content, metadata, and historical URLs.
Boundary markers: No explicit boundary markers or instructions to ignore embedded instructions are used when processing external data.
Capability inventory: The skill utilizes powerful tools including curl, nmap, ffuf, and bash for network interaction and file management.
Sanitization: The skill does not implement specific sanitization for retrieved content before it is processed by the agent.
[REMOTE_CODE_EXECUTION]: Automated security alerts identified multiple 'remote code execution' patterns. Technical review confirms these are false positives. The flagged commands (e.g., `curl -sS -D
... -o /dev/null) are used solely to inspect HTTP headers and status codes from the target; the response body is discarded to /dev/null`, preventing any execution of downloaded content.
[DATA_EXFILTRATION]: The skill communicates with external reconnaissance services such as crt.sh. These are well-known and standard resources for public asset discovery and are used here in a legitimate context for information gathering.
[COMMAND_EXECUTION]: The skill extensively uses standard security CLI tools (nmap, ffuf, dnsx, etc.). These tools are required for the stated purpose of the skill and are invoked with parameters restricted to the target scope.

wstg-web-pentest