Agent Development for Claude Code Plugins

The skill explicitly allows agents to be configured with a 'Bash' tool, enabling arbitrary command execution on the host system. This presents a direct path to system compromise if an agent is misused or compromised.

tools: ["Read", "Write", "Grep", "Bash"]

Agents can be granted 'Read', 'Write', 'Grep', and 'Glob' tools, providing extensive access to the file system. This capability allows for data exfiltration, modification, or deletion of sensitive files.

tools: ["Read", "Write", "Grep", "Bash"], Read-only analysis: ["Read", "Grep", "Glob"]

By default, if the 'tools' field is omitted, an agent has access to all available tools. This significantly increases the attack surface for any newly created agent, violating the principle of least privilege.

Default: If omitted, agent has access to all tools

The AI-assisted agent generation method allows a user's description to directly influence the generated agent's system prompt and triggering conditions. A malicious user could inject instructions to create an agent with harmful behaviors or unintended triggers.

Create an agent configuration based on this request: "[YOUR DESCRIPTION]"