The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates a documentation workflow that ingests user input to generate technical files. This creates a surface for indirect prompt injection where user instructions could be embedded into the final output documents. This is considered an inherent functional characteristic rather than a security flaw for this type of tool.\n
Ingestion points: User input collected during chat sessions in agents/agent1-requirements.md and agents/agent2-prd.md.\n
Boundary markers: Not present; the skill interpolates user-provided text directly into markdown templates without specific delimiters.\n
Capability inventory: Writes documentation files (.md) to the PRDS/ and assets/ directories on the local filesystem.\n
Sanitization: User input is formatted into structured documentation without explicit escaping or validation mechanisms.\n- [EXTERNAL_DOWNLOADS]: The skill provides documentation templates that reference well-known and established Content Delivery Networks (CDNs) such as jsdelivr.net, unpkg.com, and cdn.tailwindcss.com for standard frontend libraries. These are documented neutrally as they originate from well-known services.\n- [DATA_EXFILTRATION]: The skill maintains a local user preference file at assets/user-profile.md to store technical background and tool preferences. This is a standard persistence mechanism for improving the user experience and does not involve sensitive system paths or external exfiltration.

vibecoding-helper