The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python utility scripts (scripts/init_agent.py and scripts/validate_agent.py) via the shell to automate the creation and validation of agent files. These scripts are bundled with the skill and are used for deterministic tasks like file scaffolding and structure checking.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-defined requirements and reads existing agent/skill configurations to generate new agent system prompts, creating a surface for potential instruction injection from malicious user input or existing files.
Ingestion points: User input captured via $ARGUMENTS and the AskUserQuestion tool; existing agent definitions and skill files read during the delegation check phase (SKILL.md Step 6).
Boundary markers: The skill mandates a structured template for generated agents (Persona, Process, Output Format) which provides logical separation, although it does not explicitly require defensive delimiters for processed user data.
Capability inventory: The skill utilizes Read, Write, Grep, Glob, and Bash (restricted to running the skill's own scripts).
Sanitization: The init_agent.py script implements input normalization and regex-based validation for agent names to prevent path traversal. The validate_agent.py script uses yaml.safe_load() to ensure secure parsing of frontmatter content.

create-agent