The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data, such as user-supplied drafts, URLs, and tone documents, which creates an inherent attack surface. However, it includes extensive mitigations:
Ingestion points: Defined in rules/intake.md and rules/post-workflow.md, where external text and URLs are integrated into the research and drafting process.
Boundary markers: The skill enforces the use of [확인 필요] (Verification needed) slots for any information the agent cannot confirm through primary sources, preventing the agent from following potentially malicious fabricated instructions.
Capability inventory: The agent utilizes web search for research and generates structured HTML and JSON files as output.
Sanitization: Hard rules in references/format-options.md and references/rich-html-template.md explicitly command the agent to exclude all <script> tags, event handlers, and executable URL schemes from the generated body content.
[SAFE]: The skill's research phase targets well-known platforms like Naver and inblog, as well as official regulatory domains. These interactions are limited to reading public information as a visitor.
[SAFE]: The provided HTML assets (assets/rich-post.html) contain minimal, benign JavaScript used only for user-initiated 'Copy to Clipboard' functionality. This script is located outside the editable content region and does not process external data in an unsafe manner.

blog-maker