The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to unsafe interpolation of untrusted data in assets/print-terminal.sh during Markdown generation.
Ingestion points: Data is ingested from the system clipboard via pbpaste or from local files via cat in assets/print-terminal.sh.
Boundary markers: The content is wrapped in triple backticks (```text), but these markers are not escaped, allowing an attacker to escape the code block if the input contains triple backticks.
Capability inventory: The skill uses Bash to write to temporary files and executes pandoc, xelatex, and lpr.
Sanitization: While ANSI escape codes are stripped using sed, there is no escaping of Markdown or LaTeX special characters, which could lead to LaTeX injection when Pandoc processes the file.
[DATA_EXFILTRATION]: The skill reads from the system clipboard and arbitrary user-specified files. This represents a risk of exposing sensitive information (e.g., passwords or tokens recently copied to the clipboard) if the content is printed to a network printer.
[COMMAND_EXECUTION]: The skill executes various system commands and utilities, including pandoc for document processing, xelatex for PDF rendering, and lpr for communicating with the network printer.

terminal-print