Find your landing page message
Connect a product’s value to a specific audience and write a clear first-screen message.
Advanced web search via Tavily through AIsa API. Supports search depth, topic filtering (general/news/finance), time ranges, domain inclusion/exclusion, and LLM-generated answers. Use when: the user needs web search, research, source discovery, or content extraction.
--- name: tavily-search-18a05c549d98 description: 'Advanced web search via Tavily through AIsa API. Supports search depth, topic filtering (general/news/finance), time ranges, domain inclusion/exclusion, and LLM-generated answers. Use when: the user needs web search, research, source discovery, or content extraction.' compatibility: Designed for Agent Skills compatible clients such as OpenClaw, Claude Code, Hermes, and GitHub-backed skill catalogs. Requires system binaries python3, environment variables AISA_API_KEY and internet access to api.aisa.one. metadata: author: AIsa version: 1.0.0 homepage: https://aisa.one repository: https://github.com/baofeng-tech/agent-skills tags: x,search,research,finance,llm,aisa platforms: agentskills.io,agentskill.sh,github primary_env: AISA_API_KEY allowed-tools: Read Bash Grep --- # AIsa Tavily Search Advanced web search powered by Tavily through the AIsa API. Offers fine-grained control over search depth, topic categories, time ranges, domain filtering, and optional LLM-generated answer summaries. ## Setup This skill requires the `AISA_API_KEY` environment variable. When installed as a Claude plugin, the key is configured via the environment variables. ## Usage Run the search client with the `tavily` subcommand: ```bash python3 scripts/search_client.py tavily --query "<search query>" [options] ``` ### Arguments | Argument | Required | Default | Description | |----------|----------|---------|-------------| | `--query` / `-q` | Yes | — | Search query | | `--count` / `-c` | No | 5 | Maximum results (1–20) | | `--depth` | No | basic | Search depth: `basic`, `advanced`, `fast`, `ultra-fast` | | `--topic` | No | — | Topic filter: `general`, `news`, `finance` | | `--time-range` | No | — | Time range filter | | `--include-answer` | No | false | Include an LLM-generated answer summary | ### Examples ```bash # Basic search python3 scripts/search_client.py tavily --query "OpenAI latest announcements" --count 10 # Advanced news search with answer python3 scripts/search_client.py tavily --query "AI regulation 2026" --depth advanced --topic news --include-answer # Finance-focused search python3 scripts/search_client.py tavily --query "NVIDIA earnings Q1 2026" --topic finance --include-answer ``` ## Output The script prints structured results including: - **Title** — Page title - **URL** — Direct link - **Date** — Publication date (when available) - **Content** — Relevant excerpt - **Answer** — LLM-generated summary (when `--include-answer` is used) ## When to Use Use this skill when the user needs advanced search with specific filtering requirements: news-only results, finance-focused results, time-bounded searches, or when they want an AI-generated answer alongside raw results. This is the most feature-rich search tool in the plugin.--- name: tavily-search-18a05c549d98 description: 'Advanced web search via Tavily through AIsa API. Supports search depth, topic filtering (general/news/finance), time ranges, domain inclusion/exclusion, and LLM-generated answers. Use when: the user needs web search, research, source discovery, or content extraction.' compatibility: Designed for Agent Skills compatible clients such as OpenClaw, Claude Code, Hermes, and GitHub-backed skill catalogs. Requires system binaries python3, environment variables AISA_API_KEY and internet access to api.aisa.one. metadata: author: AIsa version: 1.0.0 homepage: https://aisa.one repository: https://github.com/baofeng-tech/agent-skills tags: x,search,research,finance,llm,aisa platforms: agentskills.io,agentskill.sh,github primary_env: AISA_API_KEY allowed-tools: Read Bash Grep --- # AIsa Tavily Search Advanced web search powered by Tavily through the AIsa API. Offers fine-grained control over search depth, topic categories, time ranges, domain filtering, and optional LLM-generated answer summaries. ## Setup This skill requires the `AISA_API_KEY` environment variable. When installed as a Claude plugin, the key is configured via the environment variables. ## Usage Run the search client with the `tavily` subcommand: ```bash python3 scripts/search_client.py tavily --query "<search query>" [options] ``` ### Arguments | Argument | Required | Default | Description | |----------|----------|---------|-------------| | `--query` / `-q` | Yes | — | Search query | | `--count` / `-c` | No | 5 | Maximum results (1–20) | | `--depth` | No | basic | Search depth: `basic`, `advanced`, `fast`, `ultra-fast` | | `--topic` | No | — | Topic filter: `general`, `news`, `finance` | | `--time-range` | No | — | Time range filter | | `--include-answer` | No | false | Include an LLM-generated answer summary | ### Examples ```bash # Basic search python3 scripts/search_client.py tavily --query "OpenAI latest announcements" --count 10 # Advanced news search with answer python3 scripts/search_client.py tavily --query "AI regulation 2026" --depth advanced --topic news --include-answer # Finance-focused search python3 scripts/search_client.py tavily --query "NVIDIA earnings Q1 2026" --topic finance --include-answer ``` ## Output The script prints structured results including: - **Title** — Page title - **URL** — Direct link - **Date** — Publication date (when available) - **Content** — Relevant excerpt - **Answer** — LLM-generated summary (when `--include-answer` is used) ## When to Use Use this skill when the user needs advanced search with specific filtering requirements: news-only results, finance-focused results, time-bounded searches, or when they want an AI-generated answer alongside raw results. This is the most feature-rich search tool in the plugin.
Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Includes source scripts; review them before use. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.