The Agent Skills Directory

[SAFE]: The skill consists entirely of markdown documentation, configuration examples (JSONC/TOML), and architectural references. There are no executable scripts included within the skill itself.
[EXTERNAL_DOWNLOADS]: The reference guides correctly point to official packages and registries. For example, it suggests installing @cloudflare/workers-types, @cloudflare/puppeteer, and official SDKs for OpenAI and Anthropic. These are recognized as safe and well-known dependencies for the tasks described.
[DATA_EXPOSURE_EXFILTRATION]: Code snippets use environment variable placeholders (e.g., process.env.CLOUDFLARE_API_TOKEN, process.env.OPENAI_API_KEY) and standard secret management patterns. No hardcoded credentials or sensitive local file path access was found.
[INDIRECT_PROMPT_INJECTION]: The skill documents various ingestion surfaces, such as Email Workers (processing inbound email) and AI Search (indexing website/R2 content). These represent the functional surface of the platform services rather than malicious intent. The guides include best practices for validation and error handling.
[REMOTE_CODE_EXECUTION]: Documentation for the Cloudflare Sandbox service (e.g., references/sandbox/patterns.md) includes examples of executing commands like curl | sh within an isolated container. This is consistent with the intended purpose of the Sandbox product for secure code execution and does not represent a risk to the host environment.
[PROMPT_INJECTION]: The instructional language uses 'IMPORTANT' and 'CRITICAL' markers solely for technical documentation callouts (e.g., requirements for API bindings or security best practices). No override patterns or safety bypass attempts were detected in the skill instructions.

cloudflare-deploy