The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to translate user requirements into configuration files such as .instructions.md, .agent.md, and hooks.json. This process of ingesting untrusted natural language data to generate files that influence agent behavior creates a potential attack surface for indirect prompt injection.
Ingestion points: User input collected during the "interview" process using the ask-questions tool or through general chat interaction.
Boundary markers: The skill does not currently implement specific delimiters or instructions to treat user-provided content as data rather than instructions when generating the configuration files.
Capability inventory: The skill utilizes file system tools to read from the codebase and write new customization files to sensitive paths (e.g., .github/, .claude/).
Sanitization: There are no explicit validation or sanitization steps mentioned to filter out instruction-override patterns from user-provided requirements before they are written to disk.

agent-customization