How do I install the Brain Search skill?

Run npx skillmds@latest add search-atlas-group/brain-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 Brain 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 Brain 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 Brain Search free to use?

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

Who published Brain Search?

search-atlas-group (@search-atlas-group) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Brain Search by search-atlas-group — Agent Skill

Name: Brain Search
Author: search-atlas-group

Brain Search

Keyword and semantic search over your local markdown notes via the offline qmd index. Use BEFORE concluding your notes have nothing on a topic, BEFORE re-asking the user something they may already have told you, and on session start for any resumed topic, project, or prior decision. Sub-second, offline, no API cost.

File contents

brain-search — find what your notes already know

Silence in your head is not silence in the notes. Search before you answer. Do not grep by guessed filename, and do not ask the user what the notes already record.

Vendor-neutral: any runtime that runs shell commands can use this. Claude Code loads it from ~/.claude/skills/brain-search/; Codex and Gemini users copy this file into their own skills directory. The commands are identical.

Use it

qmd search "<topic>" -c brain -n 5   # BM25 keyword — ~150ms, exact terms, default
qmd vsearch "<topic>" -n 5           # vector — paraphrase/synonym, seconds
qmd get <path>                       # read a full hit
qmd collection list                  # check index freshness

Use your own collection name if it is not brain.

Which mode

Start with search: it wins most queries in about a tenth of a second. Escalate to vsearch only when it comes back thin. Avoid qmd query; it can time out mid-turn.

Keyword search rewards the distinctive phrase and is diluted by padding words. Search the two or three terms that appear inside the note, not your whole sentence.

Reading results

Hits print as qmd://<collection>/<path> with a score and a snippet. Open the note before acting on it — the snippet is the file's first lines, not the answer.

When it finds nothing

Say what you searched for. "My notes have nothing on X" is legitimate only after both search and vsearch come back empty. Refresh a stale index with qmd update, or with the refresh-brain.py script the installer left in your Claude Code config folder.

---name: brain-searchdescription: Keyword and semantic search over your local markdown notes via the offline qmd index. Use BEFORE concluding your notes have nothing on a topic, BEFORE re-asking the user something they may already have told you, and on session start for any resumed topic, project, or prior decision. Sub-second, offline, no API cost.---# brain-search — find what your notes already knowSilence in *your* head is not silence in the notes. Search before you answer. Do not grepby guessed filename, and do not ask the user what the notes already record.Vendor-neutral: any runtime that runs shell commands can use this. Claude Code loads itfrom `~/.claude/skills/brain-search/`; Codex and Gemini users copy this file into their ownskills directory. The commands are identical.## Use it```shqmd search "<topic>" -c brain -n 5 # BM25 keyword — ~150ms, exact terms, defaultqmd vsearch "<topic>" -n 5 # vector — paraphrase/synonym, secondsqmd get <path> # read a full hitqmd collection list # check index freshness```Use your own collection name if it is not `brain`.## Which modeStart with `search`: it wins most queries in about a tenth of a second. Escalate to`vsearch` only when it comes back thin. Avoid `qmd query`; it can time out mid-turn.Keyword search rewards the distinctive phrase and is diluted by padding words. Search thetwo or three terms that appear *inside* the note, not your whole sentence.## Reading resultsHits print as `qmd://<collection>/<path>` with a score and a snippet. Open the note beforeacting on it — the snippet is the file's first lines, not the answer.## When it finds nothingSay what you searched for. "My notes have nothing on X" is legitimate only after both`search` and `vsearch` come back empty. Refresh a stale index with `qmd update`, or withthe `refresh-brain.py` script the installer left in your Claude Code config folder.

Frequently asked questions