The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of content from arbitrary external websites through its scraping and crawling features. This creates an attack surface for Indirect Prompt Injection, where malicious instructions hidden in the scraped content could attempt to override the agent's behavior. 1. Ingestion points: Web content fetched via the Create Scrape, Start Crawl, and Create Answer endpoints in SKILL.md. 2. Boundary markers: The documentation does not indicate the use of delimiters or specific instructions to help the agent distinguish between data and commands within the retrieved content. 3. Capability inventory: The skill provides multiple API endpoints via the orth CLI for further web interaction and data retrieval. 4. Sanitization: While the tool offers HTML transformers (e.g., Postlight's Mercury Parser) to clean content, there is no mention of filtering for prompt-injection patterns.

olostep