The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill is designed to ingest and process documentation from any website that implements the llms.txt standard. This is a primary vector for indirect prompt injection.
Ingestion points: Step 4 uses curl to download content from an arbitrary example.com (placeholder for user-provided URLs) into docs/llms.txt.
Boundary markers: Absent. The skill treats the fetched llms.txt content as trusted 'LLM-optimized markdown' without delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill executes curl, skill-seekers scrape, grep, head, and cp. The cp command writes the untrusted content to output/myskill/references/complete.md, potentially influencing future agent actions or other skills.
Sanitization: Absent. Validation in Step 2 only checks for HTTP errors and basic file structure, not for malicious instructions embedded within the markdown content.
Unverifiable Dependencies & Remote Code Execution (HIGH): The skill relies on an external command skill-seekers (referenced to a non-trusted GitHub repository jmagly/Skill_Seekers).
Evidence: Step 4 executes skill-seekers scrape --llms-txt docs/llms.txt. Executing unverified tools on untrusted data fetched from the network is a high-risk pattern.
External Downloads (MEDIUM): The skill performs multiple network requests to arbitrary domains using curl to probe for and download various llms.txt variants. While curl -I (headers only) is low risk, the subsequent download of full content for processing elevates the risk when combined with the lack of sanitization.

llms-txt-support