The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to guide the agent through the execution of coval CLI commands (e.g., coval agents create, coval agents list, coval agents update) to manage agent infrastructure. These commands are executed directly in the shell environment.
[INDIRECT_PROMPT_INJECTION]: The instructions require the agent to collect untrusted user input and interpolate it into shell command strings, which presents a command injection surface.
Ingestion points: Phase 2 and Phase 3 in SKILL.md collect agent names, endpoint URLs, and system prompts directly from the user; $ARGUMENTS is also used for pre-filling.
Boundary markers: The skill lacks explicit boundary markers or instructions to treat the ingested strings as data only.
Capability inventory: The agent has the capability to execute shell commands via the CLI integration.
Sanitization: The skill does not provide logic to sanitize, escape, or validate these user-provided strings before they are placed inside double-quoted shell arguments, potentially allowing an attacker to escape the quotes and execute arbitrary commands.
[EXTERNAL_DOWNLOADS]: The skill references official vendor websites at https://coval.dev and https://app.coval.dev for account registration and API key management. These are legitimate resources belonging to the skill author.

setup-agent