AnySearch: real-time search and full-text extraction for an Agent

Who it is for

Use this skill when you want a coding Agent to use one unified real-time search engine, instead of stitching several search tools together. It supports general web search, vertical-domain search, parallel batch search, and full-page body extraction. It fits people who need to check online material on the spot and are willing to finish one manual install first. An API key is optional but strongly recommended; without a key you can still use all search capabilities through anonymous access, with lower rate and quota. The page does not write a key variable name; do not invent one. If you have not yet decided which client directory the skill goes into, read the official paths first, then decide whether to install.

First try in 3 minutes

The official default is a manual install: first download the pinned v3.1.1 archive, then unzip it, rename it to anysearch, and move it into the client skill directory. The download command is below; the address must match the page:

curl -L -o anysearch-skill.zip https://github.com/anysearch-ai/anysearch-skill/archive/refs/tags/v3.1.1.zip

After unzipping, move the directory according to your client. Officially: Claude Code uses ~/.claude/skills/anysearch; OpenCode uses ~/.config/opencode/skills/anysearch; Cursor or Windsurf uses .skills/anysearch under the current project. Use only these written paths; do not change them to a directory that never appears, and do not invent a skills add command. If the platform has a skill store, officially you can search anysearch and install from the store; otherwise follow the manual steps above.

The first time, do only a low-risk general web search. In a client that already loaded anysearch, ask a concrete query, and first have the Agent restate: it will do real-time search and will not change your local files. Allow the network only after you confirm. Success signal: real-time results related to the query come back. If a quota is hit, then consider adding the optional key the official text mentions, instead of switching to an undocumented install.

Usage examples

Example 1: General web or vertical-domain search. Give the Agent a clear question or a domain-limited search and let anysearch run a real-time query. It fits a quick check of public information. The notes do not give a vertical-domain inventory, so do not invent an entry that “must support a given industry”; what you can confirm is that general search and vertical search both exist.

Example 2: Parallel batch search or full-page body extraction. When you have several queries, or you need a page’s full text rather than a summary, have the Agent use batch search or full-page extraction. Extraction visits the target page; before you run it the URL must come from you — do not let the Agent invent an address. The notes give no batch-count cap or extraction format; treat the actual return as the output. Anonymous mode has a lower quota, so confirm batch work first.

Limits

Install is pinned to the v3.1.1 archive; do not change the tag or guess an update-package address. Skill directories are only the official Claude Code, OpenCode, and Cursor/Windsurf paths. Store install only when the platform truly has a skill store and the search word is anysearch. A key is optional; anonymous use works with a lower quota, and the page does not write a variable name. The client names come from install paths; do not treat that as exclusive support. Do not fill in unwritten paths, skills add, or key placeholders. Repository page:

https://github.com/anysearch-ai/anysearch-skill