The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the creation of agents that ingest untrusted data (such as code and files) and possess significant capabilities. However, it explicitly provides high-quality security guidance to mitigate these risks.
Ingestion points: Agents are configured to use tools like Read, Grep, Glob, and MCP tools to ingest external content (documented in references/subagents.md).
Boundary markers: The skill explicitly recommends using XML delimiters (e.g., <prior_findings>) to wrap untrusted data and help the agent distinguish between instructions and data (documented in references/writing-subagent-prompts.md).
Capability inventory: The configuration supports powerful tools including Read, Write, Edit, Bash, and various MCP-based tools (documented in references/subagents.md).
Sanitization: The skill advocates for the principle of least privilege, providing mechanisms to restrict tool access via tools allowlists and disallowedTools denylists to minimize the attack surface.

agent-creator