The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to execute various AI agent CLI tools (agy, claude, agent, codex) for task delegation. It follows safe execution practices such as verifying the existence of binaries using 'command -v' and enforcing wall-clock limits via the 'timeout' utility.
[INDIRECT_PROMPT_INJECTION]: As an orchestration tool, the skill accepts user-defined tasks and passes them to external agent instances, creating a potential surface for indirect prompt injection.
Ingestion points: User goals and constraints are passed directly as prompt arguments to spawned agents (SKILL.md).
Boundary markers: The skill recommends using flags like '--disable-slash-commands' to prevent the interpretation of instructions as commands by the sub-agent (references/antigravity.md).
Capability inventory: The spawned agents possess significant capabilities including file system access, command execution, and network operations.
Sanitization: The instructions do not mandate specific sanitization of user-provided subtask descriptions before they are passed to the CLI tools.
[EXTERNAL_DOWNLOADS]: The documentation contains references to installation guides for the supported CLI tools. All links point to established, official domains of well-known technology providers including Google, Anthropic, OpenAI, and Anysphere (Cursor).

spawn-agent