The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data from various sources and incorporates it into system commands, creating a potential injection surface.
Ingestion points: $ARGUMENTS and multiple interactive prompts in SKILL.md (e.g., agent names, endpoint URLs, and system prompts).
Boundary markers: No delimiters or safety instructions are provided to separate user-provided content from the command structure.
Capability inventory: The skill executes numerous subprocesses via the coval CLI for creating agents, personas, and test cases across SKILL.md.
Sanitization: There is no evidence of input validation or shell escaping for the variables interpolated into CLI commands.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the Coval CLI by providing download links and instructions.
Evidence: References installation via Homebrew, Cargo, and direct binary download from https://github.com/coval-ai/cli/releases/latest/download/coval-x86_64-pc-windows-msvc.exe. These resources are hosted on GitHub and managed by the skill's author.
[COMMAND_EXECUTION]: The skill relies on extensive use of CLI tools to perform its intended functions.
Evidence: Commands such as coval agents create, coval test-sets list, and coval runs launch are used throughout the instruction flow to manage the evaluation lifecycle.

onboard