The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for generating interactive data tables and math visualizations that are susceptible to indirect injection if the source data contains malicious payloads.
Ingestion points: Untrusted data enters the agent context through the data array used in the 'Sortable / Filterable Data Tables' section in SKILL.md.
Boundary markers: There are no boundary markers or instructions to the agent to escape or sanitize content before rendering.
Capability inventory: The generated widgets have access to the sendPrompt() function (Part 6), which can trigger further agent actions, and perform arbitrary DOM manipulation using innerHTML.
Sanitization: Sanitization is absent; the templates use JavaScript template literals (e.g., `<tr><td>${r[0]}</td>...`) to write directly to the document object model, which could allow a malicious data source to execute scripts or manipulate the chat interface.
[DYNAMIC_EXECUTION]: The skill instructions guide the agent to generate and execute JavaScript code within the host environment for simulations, math plotting, and data handling.
The use of requestAnimationFrame and eval-like behavior in custom event handlers (like onClick) is a core feature of the skill but increases the potential impact of any injected content.
[EXTERNAL_DOWNLOADS]: The skill references several external JavaScript libraries to enable advanced functionality.
Fetches Chart.js from Cloudflare's official CDN.
Loads Mermaid and Three.js via esm.sh and jsDelivr.
Accesses D3.js and Tone.js from well-known public package registries. These references target well-known, established services and are considered standard practice for this skill's visualization purposes.

Advanced Visualization Techniques