cloudflare-deploy
Warn
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: Documentation for the Cloudflare Sandbox service includes examples that execute remote shell scripts via a piped interpreter (e.g.,
curl | sh) from thecode-server.devdomain inreferences/sandbox/api.mdandreferences/sandbox/patterns.md. While these are intended for use within an isolated container environment, executing unverified remote scripts is a high-risk pattern. - [EXTERNAL_DOWNLOADS]: The skill provides instructions for using the
c3(create-cloudflare) CLI, which can download and execute application templates from arbitrary GitHub repositories as shown inreferences/c3/patterns.md. It also references numerous third-party libraries across NPM and PyPI registries for database drivers, AI SDKs, and automation tools. - [INDIRECT_PROMPT_INJECTION]: The "AI Search" (AutoRAG) service documented in
references/ai-search/README.mdingests untrusted data from R2 buckets and website crawls. This content is then used to generate AI responses via theaiSearch()method inreferences/ai-search/api.md, creating a surface for indirect prompt injection where malicious instructions in the source content could influence the agent's behavior. The documentation does not specify explicit sanitization or boundary markers for these inputs.
Audit Metadata