v2.1.0 — Universal Router

The right skill loads automatically.

Send any prompt. auto-skill-finder detects intent, picks the best installed skill, compresses it, and injects caveman mode. No slash commands. No setup per prompt.

Install in 30s Documentation
bash — UserPromptSubmit hook
$fix the auth bug in my middleware
running auto-skill-finder...
> intent: code
> mode: caveman-code
> skill: caveman-code (score:11)
done. injected context.
$
0 % context skipped
0 % fewer output tokens
0 agents supported
The Problem

Manual workflows slow you down.

Every AI agent has a skill system. But none auto-route. You have to type the slash command every single time.

Manual selection

Remembering which skill exists and typing /skill-name before every prompt.

Wasted context

Loading every skill on every prompt burns your token budget instantly.

How it works

A 4-step pipeline before generation.

01

Detect intent

Classifies your prompt as code or chat using a fast rule-based classifier.

scripts/detect_intent.py
02

Activate mode

Code tasks get validation-first discipline. Chat gets terse prose.

caveman-code / caveman-chat
03

Score & Route

Scores installed skills based on keywords and description match.

scripts/skill_finder.py
04

Inject Context

Compresses the matched skill instructions and passes them to the agent.

hooks/UserPromptSubmit
Examples

Talk naturally.

Manual
/github-pr review this diff
→
Auto-Routed
review my pull request
github-pr
Manual
/jest write unit tests
→
Auto-Routed
write tests for this function
jest
Installation

One command.

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/prantikmedhi/auto-skill-finder/main/install.sh | bash

Windows

irm https://raw.githubusercontent.com/prantikmedhi/auto-skill-finder/main/install.ps1 | iex

Supported Agents

Claude Code

Native Hooks

Cursor

.cursor/rules/

Gemini CLI

Manifest

Codex

system-prompt