The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes user-supplied task descriptions to generate execution plans for sub-agents and sequential workflows. Maliciously crafted tasks could attempt to inject instructions into the generated prompts or shell commands. • Ingestion points: User task descriptions are ingested in the SKILL.md orchestration logic during Phase 1. • Boundary markers: The skill uses structured prompts for sub-agents and provides a mandatory confirmation phase in Phase 2 using the AskUserQuestion tool. • Capability inventory: The skill has access to Bash, Write, Edit, Agent, Task, and TeamCreate tools, which could be misused if the orchestration logic is subverted. • Sanitization: There is no automated sanitization of user input; the skill relies on the user to manually review and approve the execution plan.
[DYNAMIC_EXECUTION]: The skill dynamically spawns agents and creates tasks based on the runtime analysis of user requirements. It also invokes the hoyeon:ralph skill, which represents a runtime dependency on external logic not contained within this skill.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute steps within the Sequential Pipeline pattern, which may involve running code or scripts derived from user-defined subtasks.

agent-orchestrate