How do I install the Web Search skill?

Run npx skillmds@latest add hec-ovi/web-search in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Web Search safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Web Search?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Web Search free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Web Search?

hec-ovi (@hec-ovi) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Web Search by hec-ovi — Agent Skill

Name: Web Search
Author: hec-ovi

Web Search

Search the web across four backends (DuckDuckGo, Tavily, Firecrawl, Serper). Use only for things outside the app's local data — news, current prices, opening hours, recent events, places not in the destination catalog. The local catalog skills come first.

hec-ovi bcea11d 2.5 KB Updated Sep 17, 2026

File contents

When to use

Call a search tool when:

The user asks something time-sensitive (today's news, current prices, recent reviews, opening hours TODAY).
The catalog skills can't answer (city not in destination-catalog, product not in product-catalog).
The user explicitly says "search", "look up", "what is the latest".

Do NOT call when:

A local skill covers the question (destination-catalog, product-catalog, weather-forecast). Try those first.
The question is fully answerable from registered state or stable background knowledge.

How to use: pick ONE backend per question

Default order of preference:

search_serper — general factual lookup (Google index, fast).
search_tavily — research-heavy questions where you want longer per-result context (multi-paragraph snippets).
search_firecrawl — when your next step will be to scrape a specific page (Firecrawl can search-then-scrape).
search_duckduckgo — fallback that needs no API key. Note: shared IPs hit CAPTCHAs occasionally; expect intermittent failures.

If a backend returns { ok: false, code: "missing_key" } or { ok: false, code: "rate_limited" }, fall back to the next preferred backend. Do NOT report the failure as if it were the answer; transparently retry with the next option.

Query craft

A good query is short, specific, and uses real-world terms.

DO: "lisbon airport metro hours 2026", "best travel adapter europe usb-c 2026"
DON'T: "can you search for the current operating hours of the metro at lisbon airport this year"

Strip filler. Pin the year on time-sensitive queries.

Output narration

Quote the source domain inline ("per nytimes.com,..."). Don't paste raw URLs unless the user asks.
Combine 2-3 results for any non-trivial claim.
If sources disagree, flag the disagreement explicitly.
If the freshest source is older than the question requires, say so.

Anti-patterns

Do NOT call a backend when a local catalog skill would answer.
Do NOT call multiple backends in parallel "just in case".
Do NOT return the raw { ok: false } envelope to the user — fall back, then narrate the actual answer.
Do NOT recommend a product / hotel / restaurant from a single search snippet without naming the source.

---name: web-searchdescription: Search the web across four backends (DuckDuckGo, Tavily, Firecrawl, Serper). Use only for things outside the app's local data — news, current prices, opening hours, recent events, places not in the destination catalog. The local catalog skills come first.---# When to useCall a search tool when:- The user asks something time-sensitive (today's news, current prices, recent reviews, opening hours TODAY).- The catalog skills can't answer (city not in `destination-catalog`, product not in `product-catalog`).- The user explicitly says "search", "look up", "what is the latest".Do NOT call when:- A local skill covers the question (`destination-catalog`, `product-catalog`, `weather-forecast`). Try those first.- The question is fully answerable from registered state or stable background knowledge.# How to use: pick ONE backend per questionDefault order of preference:1. **`search_serper`** — general factual lookup (Google index, fast).2. **`search_tavily`** — research-heavy questions where you want longer per-result context (multi-paragraph snippets).3. **`search_firecrawl`** — when your next step will be to scrape a specific page (Firecrawl can search-then-scrape).4. **`search_duckduckgo`** — fallback that needs no API key. Note: shared IPs hit CAPTCHAs occasionally; expect intermittent failures.If a backend returns `{ ok: false, code: "missing_key" }` or`{ ok: false, code: "rate_limited" }`, fall back to the next preferredbackend. Do NOT report the failure as if it were the answer; transparentlyretry with the next option.# Query craftA good query is short, specific, and uses real-world terms.- DO: `"lisbon airport metro hours 2026"`, `"best travel adapter europe usb-c 2026"`- DON'T: `"can you search for the current operating hours of the metro at lisbon airport this year"`Strip filler. Pin the year on time-sensitive queries.# Output narration- Quote the source domain inline ("per nytimes.com,..."). Don't paste raw URLs unless the user asks.- Combine 2-3 results for any non-trivial claim.- If sources disagree, flag the disagreement explicitly.- If the freshest source is older than the question requires, say so.# Anti-patterns- Do NOT call a backend when a local catalog skill would answer.- Do NOT call multiple backends in parallel "just in case".- Do NOT return the raw `{ ok: false }` envelope to the user — fall back, then narrate the actual answer.- Do NOT recommend a product / hotel / restaurant from a single search snippet without naming the source.

Frequently asked questions