The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes shell commands for external tools (opencode, codex) by substituting user-provided variables like <model> and <dir> into command templates defined in REFERENCE.md. This pattern is vulnerable to argument injection if a user provides a malicious string containing shell metacharacters or additional flags.
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to use flags such as --dangerously-skip-permissions and approval_policy=never. These flags are designed to suppress interactive user approval prompts for file modifications and tool execution, effectively granting the downstream agent unrestricted access to the local environment and bypassing built-in safety mechanisms.
[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection. It ingests untrusted user input and pipes it into a secondary AI agent. Because the skill suppresses the secondary agent's permission prompts and safety filters, a malicious prompt could successfully instruct the downstream agent to perform harmful actions on the user's system without oversight.

assign