omnisearch

**Judgement:** High-priority, core infrastructure skill providing the *only* sanctioned way to perform web searches in this OpenClaw environment. It should be treated as mandatory for any task requiring post-training or time-sensitive information. **Strengths:** - Centralizes all web search through a single, well-documented interface (wrapper script + mcporter fallback). - Offers both summarized AI search and raw results, with multiple providers for different needs (privacy, depth, research). - Clear rules on when it must be used (current events, prices, reviews, unknown entities, fact verification, etc.), which aligns well with safe and accurate operation. **Risks / Limitations:** - Depends on external search providers and network availability; outages, API limits, or configuration issues can degrade reliability. - Quality and safety of results ultimately hinge on third-party search backends; hallucinations or biased content may appear if not critically evaluated. - Requires correct directory context and script permissions; misconfigured environments may force more brittle fallback usage. **Recommended Scenarios:** - Always invoke for any user request involving current events, recent news, live prices, product reviews, or "what’s the latest" questions. - Use whenever verifying statistics or facts that could have changed since training time. - Prefer the AI-enhanced mode (`./scripts/omnisearch.sh ai "query"`) for concise answers, and the web mode for direct source inspection, citations, or detailed research. - Follow the prescribed response pattern: summarize key findings, list 2–6 clickable sources, and note time-sensitivity/uncertainty to maintain transparency and safety.