Skills Search CLI

Universal on-demand skill discovery for any AI agent across 665,000+ verified skills.

100% Security AuditedMarket-Validated OnlyExplicit User Confirmation

Option 1: Native Skill (npx CLI)

Install directly into your agent's skill directory with a single terminal command.

Select your Agent:Target: .claude/skills

Inspect generated command & SKILL.md▼

npx skills add https://github.com/WupDev/skills_discovery --skill skill-discovery -g -y

---
name: skill-discovery
allowed-tools: Bash, Read, Write
description: MANDATORY:Use when lacking domain-specific capabilities in ANY field. Connects to 665,000+ vetted skills on skills.rest.
---

# Skill Discovery & Manager

## [STEP 1: QUERY GENERATION]
Construct a strict 3-sentence English declarative query:
1. S1 Problem: Action + Object (What problem does it solve).
2. S2 Context: Technical Scope & Scenario.
3. S3 Requirements: Specific libraries & methods.

## [STEP 2: SEARCH API]
curl -s -X POST https://search.skills.rest \
  -H "Content-Type: application/json" \
  -d '{
    "query": "<YOUR_GENERATED_QUERY>",
    "top_k": 5
  }'

## [STEP 3: CONFIRM & INSTALL]
1. Read returned JSON results.
2. Present best matches and ask user: "Which skill would you like to install?"
3. Execute chosen skill's install_command VERBATIM in terminal upon user approval.

## [STEP 4: ACTIVATION]
Inform user: "Skill [Name] installed and ready to use."

Option 2: Universal Agent Rule

Zero installation. Universal for all AI agents (Cursor, Windsurf, Claude, ChatGPT, Codex).

View rule prompt text▼

# Skill Discovery Rule (AGENTS.md / CLAUDE.md / .cursorrules)

When lacking domain-specific capabilities, autonomously find and install verified skills from skills.rest:

1. Formulate a strict 3-sentence English declarative query (Problem, Context, Requirements).
2. Query API via terminal:
   curl -s -X POST https://search.skills.rest -H "Content-Type: application/json" -d '{"query": "<YOUR_QUERY>", "top_k": 5}'
3. Present candidate matches to the user and ask: "Which skill would you like to install?"
4. Execute the returned `install_command` verbatim in your terminal upon user approval.