faf-wizard

Warn

Audited by Gen Agent Trust Hub on Aug 10, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions recommend the installation of global NPM packages (faf-cli) and the use of npx -y claude-faf-mcp@latest to set up an MCP server. These resources originate from third-party sources that are not verified or recognized as trusted.\n- [REMOTE_CODE_EXECUTION]: The use of npx -y to execute the claude-faf-mcp package constitutes a remote code execution pattern, as it downloads and runs code from the NPM registry in a single step without version pinning or source verification.\n- [COMMAND_EXECUTION]: The provided workflow guides users to execute multiple shell commands (faf auto, faf migrate, faf sync) using a third-party CLI tool, which may perform arbitrary operations on the local file system or network.\n- [PROMPT_INJECTION]: The skill describes a tool that ingests untrusted data from project files to influence AI behavior, presenting an indirect prompt injection risk.\n
  • Ingestion points: The tool scans project manifest files (e.g., package.json, Cargo.toml, pyproject.toml), directory structures, and README.md files as part of its Phase 1 and Phase 2 detection process.\n
  • Boundary markers: There are no instructions or delimiters specified to ensure that the agent ignores commands or instructions embedded within the scanned project files.\n
  • Capability inventory: The skill involves the execution of a CLI tool and an MCP server that can read project files and generate context files (project.faf) used by multiple AI platforms.\n
  • Sanitization: The skill does not mention any validation, filtering, or escaping mechanisms to sanitize the content extracted from untrusted external sources before it is processed by the AI.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 10, 2026, 09:43 AM
Security Audit — agent-trust-hub — faf-wizard