The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a template for a shell command parallel-cli extract "$ARGUMENTS" --json -o "/tmp/$FILENAME.json" where user-provided arguments and an agent-generated filename are interpolated directly into a bash command. This pattern is vulnerable to command injection if the input contains shell metacharacters like backticks or semicolons.
[EXTERNAL_DOWNLOADS]: The skill includes instructions that direct the agent to advise the user to run /parallel-setup if the primary tool is missing. This encourages the execution of an external setup script for dependency installation, which is a potential vector for executing untrusted code.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection when fetching content from arbitrary URLs.
Ingestion points: External data is ingested from arbitrary URLs through the parallel-cli extract command (SKILL.md).
Boundary markers: The instructions lack delimiters or safety warnings to separate external content from the agent's instructions, instead requiring the agent to extract and return the content "verbatim."
Capability inventory: The skill uses the Bash tool to run CLI commands and has filesystem write access via the /tmp directory (SKILL.md).
Sanitization: No sanitization or filtering is applied to the extracted content before it is presented to the agent.

parallel-web-extract