The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of ai_instructions for a separate voice agent. These instructions are persisted via the ttai:update_scenario tool, creating a surface where user-supplied content or harvested data could influence downstream agent behavior.
Ingestion points: The ai_instructions field within the scenario data structure, as described in SKILL.md and references/steps-format.md.
Boundary markers: The templates recommend using structured Markdown headers (e.g., ## BROWSER TOOL INSTRUCTIONS) to separate injected behavior rules from other content.
Capability inventory: The skill utilizes tools for scenario management (ttai:update_scenario) and authentication (ttai:authenticate_browser), which communicate with the vendor's API to modify persistent agent configurations.
Sanitization: The instructions do not define specific escaping or validation procedures for narration cues or action descriptions provided by the user before they are included in the generated instructions.
[DYNAMIC_EXECUTION]: The documentation in references/selector-guide.md provides static JavaScript templates designed to be executed in a browser context for auditing DOM elements and extracting selector metadata.
[CREDENTIALS_UNSAFE]: The skill describes a system where interaction arguments (e.g., login values) are stored in plaintext within scenario configurations. It mitigates this risk by explicitly instructing the user to only utilize sandbox or demo-specific credentials and provides an OAuth-based authentication flow for persistent sessions via ttai:authenticate_browser.

browser-demo-builder