The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by instructing the agent to ingest and analyze data from arbitrary external targets ($TARGET). A malicious target could serve content designed to influence the agent's behavior. * Ingestion points: The skill uses curl to fetch HTML, JSON, and JavaScript chunks from external URLs across all phases. * Boundary markers: There are no boundary markers or instructions to the agent to treat target data as untrusted content. * Capability inventory: The skill has the ability to execute shell commands including curl, grep, and python3. * Sanitization: The processing logic uses Python's json and re modules to extract data, which provides structural validation but does not screen for natural language injection attacks.
[COMMAND_EXECUTION]: The skill employs various command-line tools for reconnaissance and testing. An automated detection for remote code execution was identified where target output is piped to python3; however, analysis shows this is used to parse JSON and HTML props from the target using static local Python code, which is a standard part of the tool's intended auditing functionality.

hunt-nextjs