The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to download and install a software suite from a third-party GitHub repository (https://github.com/hexian2001/H-pentest.git). This source is not categorized as a verified or trusted vendor.
[COMMAND_EXECUTION]: Installation and operation involve multiple shell commands, including docker-compose, pip install, npm install, and manual execution of Python initialization scripts. These commands provide a significant surface for local system impact during setup.
[REMOTE_CODE_EXECUTION]: The platform includes built-in tools (execute_python, kali_execute) designed to allow agents to generate and execute arbitrary code and CLI commands. While the documentation mentions a Docker sandbox for Python, this capability represents a core high-risk function of the skill.
[DATA_EXFILTRATION]: The skill facilitates network operations to external target URLs using libraries like requests and integrated security tools like Nuclei. This involves sending data to and receiving data from potentially unknown external domains.
[PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection through the processing of untrusted data from external targets.
Ingestion points: The Worker Agent receives and processes data (observations) from target websites via tools like execute_python and nuclei_scan as described in SKILL.md.
Boundary markers: There are no documented delimiters or clear instructions within the prompts to distinguish or ignore malicious instructions embedded in the data retrieved from targets.
Capability inventory: The platform possesses extensive capabilities including Python code execution, Kali Linux tool execution, and network request capabilities as detailed in SKILL.md.
Sanitization: The instructions do not specify any validation or sanitization of content retrieved from external sources before it is interpreted by the AI agents.

h-pentest-ai-platform