Nacos 3.2 Skill Registry Launches: Securing Enterprise AI Capabilities

In the past year the team debated low‑code platforms versus high‑code frameworks; low‑code tools like Dify are easy to use but inflexible, while high‑code approaches based on ReAct (LLM + Prompt + Tool) offer flexibility but suffer from hallucinations and costly prompt debugging. Anthropic’s Skill emerged as a middle ground, providing flexibility with encapsulated capabilities and boundary constraints, leading to rapid adoption in enterprises and a booming public Skill market (ClawHub, SkillHub) with tens of thousands of Skills.

Security analyses by Snyk on a sample of 3,984 Skills from ClawHub revealed that 36.82% contain vulnerabilities, 13.4% are critical, and issues such as secret leaks and prompt injection are common, posing major challenges for enterprise deployment.

Four inter‑related challenges were identified for bringing Skills into enterprises:

Security challenges: persistent risks of malicious code, known vulnerabilities, and data leakage without effective admission controls.

Permission challenges: unclear visibility, usage, modification, and publishing rights leading to over‑privileged access.

Stability challenges: version chaos, uncontrolled upgrades, and difficult rollbacks affecting business continuity.

Governance challenges: missing audit logs, incomplete traceability, and lack of compliance evidence.

These problems cannot be solved by a single capability; a platform‑level governance system is required.

Nacos 3.2 Skill Registry is presented as an enterprise‑grade AI resource governance platform that inserts a "trust after verification" layer between Agents and Skills. It unifies Skill auditing, management, distribution, and traceability, forming a closed‑loop control system.

Quick Start provides one‑line installers for macOS/Linux ( curl -fsSL https://nacos.io/nacos-installer.sh | bash) and Windows PowerShell (

iwr -UseBasicParsing https://nacos.io/nacos-installer.ps1 | iex

), automatically launching the Nacos console.

1. Skill Security Audit Pipeline ships with a built‑in plugin that scans for over ten common risks and offers a standard interface for custom extensions. The pipeline enforces a "reject if not passed" policy, turning security from a documentation requirement into a system‑enforced constraint, with multi‑dimensional scanning, risk grading, and auditable decisions.

2. Multi‑Version Management & Gray Release supports a full lifecycle (DRAFT, REVIEWING, GRAY, FORMAL, OFFLINE). Labels such as dev/latest/stable bind versions to control rollout scope, enabling gradual traffic increase and instant rollback by switching label mappings.

3. Permission Model & Multi‑Layer Isolation adopts a three‑layer boundary design:

RBAC roles (publisher, reviewer, read‑only) define clear responsibilities.

Namespace isolation separates teams, environments, and tenants.

Skill‑level visibility controls public, private, or scoped access, specifying who can view, use, modify, or publish each Skill.

4. Full‑Link Audit & Traceability records upload, audit, publish, and invocation logs, capturing who performed each action, when, and on which version, enabling complete traceability for compliance.

5. Multi‑Path Skill Ingestion & Distribution offers various acquisition methods:

Agent autonomous discovery via curl -s https://download.nacos.io/SKILL.md to list and install Skills.

CLI one‑click install ( nacos-cli skill-list, nacos-cli skill-get mysql-query -o ~/.skills).

Shell script batch pulling using environment variables and npx skills add mysql-query redis-query.

The registry integrates with HiClaw (Agent runtime) and HiMarket (private Skill/Worker marketplace) to form a complete private AI management ecosystem, where Nacos provides unified AI resource management (Prompt, Skill, MCP, AgentCard) and extensible plugins.

Looking ahead, Nacos aims to become the core control plane for AI resources, deepening governance capabilities such as active‑usage detection, data‑intelligent layer with OT‑based tracing, full‑link audit, semantic search via vector databases, multi‑protocol adapters (A2A, ACP, Matrix), and a Coding Agent plugin that evolves from Markdown commands to native MCP endpoints.

In conclusion, Nacos 3.2 marks the first step toward embracing the AI era, offering a secure, controllable, and extensible foundation for enterprise AI workloads.