The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill is designed to fetch instructions from a remote URL and apply them as logic to your local files. The instructions explicitly state that the 'fetched content contains all the rules and output format instructions.'
Evidence: Fetching from https://raw.githubusercontent.com/cloudflare/web-interface-guidelines/main/command.md to determine agent behavior.
Risk: If the remote repository is compromised, the agent will follow malicious instructions, such as reading sensitive files and sending their content to an attacker-controlled endpoint.
EXTERNAL_DOWNLOADS (MEDIUM): The skill relies on an external source from a GitHub organization (cloudflare) that is not included in the predefined list of trusted organizations.
PROMPT_INJECTION (LOW): This skill exhibits a significant Indirect Prompt Injection surface (Category 8) because it treats untrusted remote data as high-privileged instructions.
Ingestion points: SKILL.md fetches content from a remote markdown file.
Boundary markers: Absent; there are no instructions to ignore embedded commands in the fetched guidelines.
Capability inventory: The agent has the ability to read arbitrary files based on the fetched rules.
Sanitization: Absent; the remote content is directly interpolated into the agent's task logic.

web-design-guidelines-6