The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to interpret and execute instructions from a dev.yml file found in the project root. This file is an external, untrusted source of data.
Ingestion points: Configuration data is read from dev.yml in the current project directory.
Boundary markers: The skill lacks instructions for the agent to validate or sanitize the commands found in the configuration file before execution.
Capability inventory: The skill describes a tool capable of executing arbitrary shell commands through various keys (e.g., up, build, test, commands). It also includes capabilities for file system modification (e.g., cp .env.example .env) and network access via package managers.
Sanitization: No sanitization or safety checks are performed on the shell strings defined in the configuration.
[COMMAND_EXECUTION]: The described dev tool is designed to execute shell commands for project management. This includes 'custom tasks' with user-defined shell logic for environment checks (met?) and provisioning (meet).

dev-cli