cloudflare-full-stack-scaffold

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION] (MEDIUM): The skill documentation instructs the execution of multiple shell scripts located in the scripts/ directory, including setup-project.sh, init-services.sh, enable-auth.sh, and enable-ai-chat.sh.
  • Evidence: These scripts are described as having capabilities to modify the filesystem (cp -r), run build commands (npm install), and update configuration files (wrangler.jsonc, .dev.vars).
  • Risk: The scripts are not provided for direct analysis, and their description indicates they interact with sensitive setup processes.
  • [EXTERNAL_DOWNLOADS] (LOW): The workflow relies on npm install to download a large number of third-party packages.
  • Evidence: The skill references package.json containing dependencies for React, Hono, AI SDK, and Clerk.
  • Risk: Standard dependency risk associated with supply-chain attacks, exacerbated by the 'all Cloudflare services' scope.
  • [CREDENTIALS_UNSAFE] (LOW): The skill specifically targets the handling of sensitive credentials.
  • Evidence: enable-auth.sh and enable-ai-chat.sh are described as 'prompting for API keys' and updating .dev.vars files.
  • Risk: While this is a standard developer workflow, the handling of keys by unverified scripts is a sensitive data surface.
  • [PROMPT_INJECTION] (LOW): The skill implements a 'Session handoff protocol' via SCRATCHPAD.md to bridge context between AI sessions.
  • Category 8 (Indirect Prompt Injection) Evidence Chain:
  • Ingestion points: SCRATCHPAD.md file reads and writes.
  • Boundary markers: None mentioned in the README for delimiting handoff data.
  • Capability inventory: Shell script execution, filesystem writes, and network operations (via wrangler).
  • Sanitization: No mention of sanitizing the content passed between sessions via the scratchpad.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 06:07 PM