The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains a 'Quick Start' instruction to download and execute a remote script using curl -fsSL https://cli.inference.sh | sh. Piped execution of remote content bypasses security reviews and executes unverified code with the privileges of the current user.
[DYNAMIC_EXECUTION]: The skill utilizes the infsh/python-executor tool to execute Python code blocks at runtime. This pattern allows for arbitrary code execution, which is used here to generate charts via Matplotlib.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands via the infsh CLI tool and also recommends installing additional dependencies through npx skills add commands.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect injection by accepting structured data (HTML and Python code) as tool inputs. If these inputs are populated using untrusted external data without strict sanitization, an attacker could manipulate the generated visuals or the code being executed.
Ingestion points: Tool inputs for html-to-image and python-executor in SKILL.md.
Boundary markers: Absent. The skill interpolates data directly into code/HTML templates.
Capability inventory: Subprocess execution via infsh and Python execution.
Sanitization: Not demonstrated in the provided instructions.

pitch-deck-visuals