smart-search-cli
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill is centered on the agent executing the local 'smart-search' CLI tool and its various subcommands to perform web-based tasks.
- [INDIRECT_PROMPT_INJECTION]: The skill provides the agent with the ability to fetch and process arbitrary content from the web, which introduces an attack surface where external content could attempt to influence agent behavior. • Ingestion points: 'smart-search fetch', 'smart-search search', 'smart-search research', and 'smart-search exa-search' (File: references/cli-core.md). • Boundary markers: The skill instructs the agent to 'summarize only what fetched text supports' and uses a 'fetch_before_claim' evidence policy (File: references/deep-research-mode.md). • Capability inventory: Includes shell command execution, local file writing, and configuration modification (File: references/cli-core.md). • Sanitization: Relies on the agent's internal safety filters and synthesis instructions to verify claims against fetched text.
- [EXTERNAL_DOWNLOADS]: The tool facilitates interaction with numerous external search and data providers (including xAI, Zhipu, Exa, Context7, Tavily, Firecrawl, TinyFish, DeepSeek, AnySearch, and Sciverse) and allows fetching data from any user-provided URL as part of its core research functionality.
Audit Metadata