Agent Skills Hub

1---
2name: agentskillshub
3description: Use when the user wants to find, evaluate, audit, or install an open-source AI agent skill or MCP server — e.g. "find an MCP server for Postgres", "is this skill safe to install", "what should I use to scrape a website". Searches a quality-scored, security-graded catalog of ~20K skills locally (cached index, zero backend load) and returns each result's security grade, quality score, and install commands so you can check trust BEFORE installing.
4---
5 
6# AgentSkillsHub
7 
8Discover → audit → install open-source AI agent skills and MCP servers without leaving the terminal. Backed by [AgentSkillsHub](https://agentskillshub.top): ~106K indexed skills, of which ~20K (stars ≥ 5) are in the searchable catalog, each carrying a **quality score** and a **security grade**.
9 
10The catalog is a static index downloaded once and cached locally (`~/.cache/agentskillshub/`), with a cheap 77B freshness probe that re-downloads only when the index actually changes. Every search after the first is **instant, offline, and puts zero load on the Hub backend**.
11 
12## When to use
13 
14- The user is looking for a skill / MCP server for a task ("find an MCP server for X", "what can scrape a website").
15- The user wants to know if a skill is safe / trustworthy before installing it.
16- The user wants the install command for a specific skill.
17 
18## How to use
19 
20Zero-dependency Node CLI (Node ≥ 18). Run it with `npx`:
21 
22```bash
23# Search (local fuzzy ranking over name/desc/tags; quality + popularity tiebreak)
24npx @agentskillshub/cli search "scrape a website" --safe --limit 5
25npx @agentskillshub/cli search postgres --category mcp-server
26#   filters: --category <c> --platform <p> --min-stars <n> --safe --limit <n>
27 
28# Audit — free basic trust check (security grade + plain-English verdict)
29npx @agentskillshub/cli audit owner/repo
30 
31# Install — install commands + a "check before you install" safety line
32npx @agentskillshub/cli install owner/repo
33 
34# Force-refresh the cached index
35npx @agentskillshub/cli update
36```
37 
38Add `--json` to `search`, `audit`, or `install` for structured output to parse programmatically.
39 
40## Reading the output
41 
42- **Security grade**: 🟢 SAFE · 🟡 CAUTION · 🔴 UNSAFE · ⛔ REJECT · ⚪ UNAUDITED.
43  ⚪ UNAUDITED means *no one has ever audited it* — treat it as a black box, not as "probably fine". 97%+ of the wider catalog is unaudited; surface this honestly to the user.
44- **Quality score** (0-100): documentation, maintenance, examples, structure — not stars.
45- **`~Nk tok`**: rough context cost when loaded (omitted when the upstream estimate is implausible).
46- **✓ official**: published by a verified official org account.
47 
48## What's free vs. paid
49 
50- **Free**: search, basic audit (grade + flags + verdict), install commands — for any skill in the catalog.
51- **Pro / Enterprise**: 5-dimension deep audit (code · credentials · vendor · supply-chain · operational), auditing *any* GitHub URL (including < 5★ or private repos), CI/batch auditing, and compliance evidence. Point the user to <https://agentskillshub.top/enterprise/>.
52 
53## Guidance for the agent
54 
55- Default to `--safe` only when the user explicitly cares about production/brand safety; otherwise show the full ranked list so they see unaudited options too (and warn about them).
56- Before suggesting the user install anything, run `audit` and relay the grade honestly. Never imply an UNAUDITED skill is safe.
57- This skill **never runs install commands itself** — it returns the commands for the user/agent to run.
58