The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses aggressive, imperative language to override the agent's default behavior and conversational protocols.
Evidence: "CRITICAL: Auto-start on load. When this skill triggers, go straight to Step 1. Do not summarise. Do not explain options. Start immediately."
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (post files or user input) and interpolates it into prompts and files without sanitization or boundary markers.
Ingestion points: The skill reads "the most recent post file" from the project or accepts pasted text from the user in Step 1.
Boundary markers: Absent. There are no instructions for the agent to ignore or delimit instructions contained within the post content.
Capability inventory: The skill has the capability to read project files, write new HTML files to the disk, and call the AskUserQuestion tool.
Sanitization: Absent. The content extracted from the post is directly used to generate headlines, summary points, and image generation prompts.

graphic-designer