The Agent Skills Directory

Indirect Prompt Injection (LOW): Multiple templates (e.g., code-reviewer, data-analyst, security-auditor, support-assistant) define agent configurations that ingest untrusted data from the local environment or web searches while possessing sensitive tools like Bash or file system writing. This configuration creates a vulnerability surface where malicious content in processed files or search results could potentially influence the agent's actions.\n
Ingestion points: Use of Read, Grep, Glob, and WebSearch tools across various templates (e.g., templates/agents/code-reviewer.md, templates/specialized/legal-assistant.md).\n
Boundary markers: No specific delimiters or "ignore" instructions are defined in the templates to separate external data from system instructions.\n
Capability inventory: Templates for code-reviewer, data-analyst, financial-analyst, and security-auditor explicitly include the Bash tool. Templates for technical-writer and hr-assistant include Write and Edit tools.\n
Sanitization: No evidence of input validation, escaping, or sanitization logic within the prompt instructions.

prompt-library