The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a shell script scripts/validate-agent.sh for users to verify the structure and content of agent markdown files. The script performs static analysis using standard local utilities including grep, sed, and awk without elevated privileges.
[INDIRECT_PROMPT_INJECTION]: The skill includes templates in references/agent-creation-system-prompt.md and examples/agent-creation-prompt.md for generating agent configurations from user descriptions. This represents an indirect injection surface where a malicious task description could attempt to influence the agent's generated system prompt.
Ingestion points: The [YOUR DESCRIPTION] placeholder in the agent generation template.
Boundary markers: The template wraps user input in double quotes to delimit it from the system instructions.
Capability inventory: The skill defines tools such as Read, Write, Grep, Glob, and Bash that generated agents are expected to use.
Sanitization: No explicit input sanitization or validation is implemented within the provided templates.
[SAFE]: The documentation explicitly encourages the principle of least privilege by teaching developers how to restrict an agent's tool access using the tools field in YAML frontmatter. The overall content is educational and follows established patterns for AI agent development.

developing-agents