The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-supplied fields (ideas, documents, and consultant questions) stored in Busabase Bases and interpolates them into the agent's context. While the skill includes robust frontend sanitization, the agent context itself remains a surface for potential instructions embedded in these fields.
Ingestion points: Data is read from the ideas, documents, and questions bases via busabase-provider.js.
Boundary markers: No explicit delimiters are used in the prompt construction logic, although the agent is instructed to treat the content as user-provided vocabulary.
Capability inventory: The skill has permissions to create and update records in the database and hand off data to the $kelly-app-skill-creator tool.
Sanitization: Uses DOMPurify for sanitizing HTML and SVG in the frontend display, but there is no explicit filtering of natural language instructions for the LLM context.
[SAFE]: The skill implements excellent defensive measures for its frontend component:
HTML Sanitization: Uses DOMPurify in markdown-renderer.js with a restrictive configuration to sanitize user-supplied Markdown and SVG content.
Mermaid Security: Configures the Mermaid diagramming engine to securityLevel: "strict" and disables htmlLabels to prevent potential script execution through diagrams.
[SAFE]: All external dependencies are restricted to well-known libraries (hono, marked, dompurify, mermaid) or official platform SDKs (busabase-sdk). No suspicious or unverifiable packages were detected.

kelly-ideas