search-web

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The setup script (scripts/setup-mcp.sh) installs the @tiny-fish/cli package globally via npm and utilizes npx to fetch MCP servers like @upstash/context7-mcp and github-fetcher-mcp.
  • [COMMAND_EXECUTION]: The skill instructs the agent to run scripts/setup-mcp.sh, which executes various shell commands for environment detection and tool configuration.
  • [PERSISTENCE]: To maintain API access, the setup script modifies shell configuration files (~/.bashrc and ~/.zshrc) to export the TINYFISH_API_KEY, ensuring the variable is available in subsequent sessions.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a broad ingestion surface as it reads data from the web and GitHub repositories. While it lacks explicit sanitization, it enforces a strict output format in references/rules/output-format.md to distinguish external content from agent responses.
  • Ingestion points: Technical documentation (Context7), web search results (Exa, TinyFish), and GitHub content (mcp-deepwiki, github-fetcher).
  • Boundary markers: Output format rules define structured Markdown headers and source labeling.
  • Capability inventory: Shell command execution (setup-mcp.sh), file system modification (shell profiles), and network requests (MCP tools).
  • Sanitization: None detected in instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 08:28 AM
Security Audit — agent-trust-hub — search-web