The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines patterns for agents that ingest untrusted external data (such as source code), creating an attack surface for indirect prompt injection.\n- Ingestion points: Agents built from these templates use tools like Read, Grep, and Glob to ingest workspace files and utilize CLAUDE.md for context.\n- Boundary markers: The provided system prompt templates do not explicitly include instructions to use delimiters or boundary markers to separate untrusted data from instructions.\n- Capability inventory: Example templates include agents with powerful capabilities such as file modification (Write) and arbitrary command execution (Bash).\n- Sanitization: The documentation lacks specific guidance on sanitizing or validating ingested content before it is processed by the agent's core logic.

Agent Development