The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the infsh command-line tool via Bash to interact with the inference.sh platform for authentication and application execution.
[REMOTE_CODE_EXECUTION]: The skill is designed to transmit user-provided Python scripts to a remote sandboxed environment for execution on the vendor's infrastructure.
[DATA_EXFILTRATION]: The environment includes networking libraries such as requests, playwright, and selenium that enable network communication and data retrieval from external web sources and APIs.
[EXTERNAL_DOWNLOADS]: The documentation points to an installation script for the required CLI tool hosted on the official inference-sh GitHub repository.
[PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection by processing untrusted content from the web.
Ingestion points: Untrusted data enters the agent context through web scraping tools (beautifulsoup4, selenium, playwright) and HTTP clients (requests, httpx) listed in SKILL.md.
Boundary markers: No instructions are provided to the agent to treat scraped content as untrusted or to use specific delimiters to separate data from instructions.
Capability inventory: The skill allows for subsequent shell command execution via the infsh tool and further network requests based on processed data.
Sanitization: No sanitization or validation of the retrieved external content is performed before it is processed or returned to the agent.

python-executor