The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system commands and custom CLI tools to perform its auditing tasks.
Uses agent-browser to control a web browser for product walkthroughs, capturing snapshots, and inspecting network logs.
Executes shell commands like mkdir, chmod, printf, and jq for directory management and configuration.
Uses openssl to generate random passwords for test accounts.
[DYNAMIC_EXECUTION]: The skill uses heredocs to generate and execute bash scripts at runtime, specifically in references/disposable-inbox.md for interacting with the mail.tm API.
These scripts interpolate variables such as $PERSONA, $ADDR, and $PRODUCT, which could lead to command injection if persona names or project titles contain shell metacharacters.
[EXTERNAL_DOWNLOADS]: The skill interacts with external services to facilitate testing.
Fetches data from https://api.mail.tm to manage disposable email accounts and retrieve verification codes.
Accesses the product's staging and production environments (zenstory.ai, vercel-labs protected URLs) for auditing.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted content from the web, creating a vulnerability surface for indirect prompt injection.
Ingestion points: Reads text from the audited website (via agent-browser snapshot) and registration emails (via the mail.tm API).
Boundary markers: Relies on persona-based instructions and

new-user-product-audit