Give Codex, Claude Code, and compatible AI agents fast, token-conscious Windows file discovery through the Everything Command-line Interface (ES).
Click the animation to watch the full 45-second demonstration.
The skill does not force every search through Everything. It routes each request to the cheapest suitable tool:
| Search situation | Preferred tool |
|---|---|
| Exact known path | Test-Path or Get-Item |
| Filename or content inside a known project | rg --files or rg |
| Unknown location, cross-drive search, or indexed metadata | es.exe |
It also caps results at 20 by default, returns plain paths, and asks the agent to refine broad queries before requesting more output.
- Find a document when its folder is unknown.
- Search across indexed drives by filename, extension, size, or date.
- Avoid recursive whole-computer scans and oversized tool responses.
- Keep file discovery read-only unless the user separately requests an action.
- Windows
- Everything running with the required locations indexed
- ES available as
es.exeonPATH
The skill has been tested locally with Codex, Windows PowerShell 5.1, and PowerShell 7. Other tools that support the open Agent Skills format may also load it.
Install globally with the open skills CLI and choose the supported agent or agents when prompted:
npx skills add DerrickZeng33/find-files-with-everything-cli --skill find-files-with-everything -gBrowse the skill on skills.sh.
Everything and ES are separate prerequisites and are not included in this repository.
On 2026-08-12, a synthetic folder containing 20,000 empty files and 200 unique matching spreadsheets produced these results over 21 warm runs with PowerShell 7.6.4, ES 1.1.0.37, and Everything 1.5.0.1418:
| Query | Median | Paths returned | Output characters | Approx. tokens* |
|---|---|---|---|---|
| Everything index, location unknown, capped at 20 | 33.6 ms | 20 | 1,779 | 445 |
| PowerShell recursive scan, known root, uncapped | 23.5 ms | 200 | 17,799 | 4,450 |
| PowerShell recursive scan, known root, capped at 20 | 2.8 ms | 20 | 1,779 | 445 |
The 90% output reduction versus the uncapped command came from the skill's result cap, not from ES alone. The known-root capped scan was faster, which is why the skill keeps local project searches on local tools. Everything's advantage is unknown-location discovery through its existing index without making the agent recursively scan drives.
*Approximate tokens use characters divided by four; actual tokenization varies. To protect privacy, no recursive scan was run outside the synthetic folder. The folder was deleted immediately after the benchmark. These are controlled observations, not universal performance guarantees.
The repository contains no telemetry or result-storage component and bundles no credentials or executables. The skill performs read-only searches, directs agents away from ES options that change Everything state, and tells them not to export results unless asked.
This is an unofficial community project. It is not affiliated with or endorsed by voidtools. Everything and ES belong to their respective owners.
This project is licensed under the MIT License.

