The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses javascript_tool to execute scripts within the user's browser context. It utilizes synthetic ClipboardEvent dispatch and DOM manipulation (e.g., document.activeElement.dispatchEvent) to interact with the Medium editor. This runtime code execution is the primary mechanism for the skill to function on a platform that restricts standard automation.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided markdown content and converts it into HTML for publishing, which introduces a surface for indirect prompt injection. Malicious instructions embedded in the input content could potentially influence the agent's behavior during the multi-step publishing and verification process.
Ingestion points: Markdown article content provided as input to the agent (SKILL.md).
Boundary markers: None specified for the content being processed.
Capability inventory: Use of javascript_tool for DOM interaction and extension-based tools for browser control.
Sanitization: The instructions include a manual conversion process from markdown to a restricted set of HTML tags (e.g., <h1>, <p>, <ul>, <li>), which serves as a basic sanitization layer.

medium-post