The Agent Skills Directory

[SAFE]: The skill acts as an instructional assistant for Cloudflare One products, relying on official documentation from Cloudflare's developer portal, which is a trusted source. It contains no malicious code, unauthorized data access patterns, or obfuscation.
[SAFE]: The instructions promote security-first configurations, emphasizing least-privilege principles, staged rollouts for risky changes, and thorough validation of identity-based policies. It includes specific guardrails to prevent broad production changes without explicit approval.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates a standard attack surface for indirect prompt injection as part of its core functionality.
Ingestion points: It retrieves data from documentation at developers.cloudflare.com and processes user-supplied context (account IDs, application details) to generate configuration proposals.
Boundary markers: The skill relies on its structured workflow rather than explicit delimiters or specific warnings to ignore embedded instructions in the ingested data.
Capability inventory: The agent is instructed to inspect existing Cloudflare resources and propose or execute configuration changes via Cloudflare APIs and MCP tools.
Sanitization: No specific sanitization or escaping mechanisms for external content are defined in the instructions.
Contextual Assessment: This vulnerability surface is considered low risk as it is necessary for the assistant's primary purpose and is balanced by the instruction's emphasis on manual validation and staged rollout procedures.

cloudflare-one