The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download binary release artifacts for multiple architectures from the GitHub repository DavidValin/universal-prompt-language. These resources are owned by the skill author.
[COMMAND_EXECUTION]: The skill provides explicit instructions to install the downloaded binary using sudo install -m 0755 upl /usr/local/bin/upl. This requires administrative privileges to modify system directories. It also uses the CLI for prompt rendering via upl build-from-json.
[REMOTE_CODE_EXECUTION]: The pattern of downloading a binary from a remote repository and subsequently executing it with system privileges constitutes a remote code execution risk.
[PROMPT_INJECTION]: The skill recursively searches for and reads .upl and .txt files from local directories, parsing metadata fields (name, description) to influence agent behavior. This ingestion of untrusted local data represents a surface for indirect prompt injection.
Ingestion points: Recursive file search in ~/.upl/prompts or user-specified directories (e.g., find ~/.upl/prompts -type f).
Boundary markers: The skill contains a manual safety rule (Rule 20) advising the agent not to execute shell commands contained in prompt text, but no technical boundary markers or sanitization logic are defined in the instructions.
Capability inventory: The skill allows file system discovery, temporary file writing, and arbitrary CLI execution via the upl binary.
Sanitization: None detected; the agent relies on the upl CLI's internal handling of parameters provided via JSON files.

run-upl-prompts