by 安全扫描
OpenClaw
可疑
medium confidenceThe bundle implements a full XClaw network client (including registration, key generation, and persistent storage of private keys/tokens) but its name/metadata don’t match that purpose and it will collect and store sensitive credentials — review and restrict before installing.
评估建议
Key issues to consider before installing: (1) The skill's name and registry metadata don't match its actual function — it's an XClaw network client, not a 'NamingExpert'. Ask the author to correct the metadata and explain the mismatch. (2) The skill will read XCLAW_* env vars if present and, if not, will prompt conversationally for secrets (API keys, JWTs) and save tokens and private keys in plaintext at ~/.xclaw/config.json — only proceed if you trust the XClaw network and the skill author. (3)...详细分析 ▾
⚠ 用途与能力
The skill is named 'NamingExpert灵犀取名官' but the SKILL.md, scripts, and reference docs implement an XClaw agent-network client. Registry metadata declares no required env vars/credentials, yet the runtime expects XCLAW_BASE_URL, XCLAW_JWT_TOKEN, XCLAW_API_KEY, and XCLAW_AGENT_ID. This mismatch between name/metadata and actual capabilities is incoherent and should be explained by the author before trust.
⚠ 指令范围
The SKILL.md instructs the agent to (a) perform network calls automatically for many read-only endpoints without asking the user first and (b) when write/auth operations are needed, to collect credentials conversationally and 'use them silently'. The bundled scripts implement registration that generates keypairs and send data to the declared API. The instructions permit collecting secrets via chat and saving returned tokens/private keys for later use, which expands the skill's scope beyond name-generation and can lead to unwanted credential collection or transmission.
ℹ 安装机制
There is no remote install or download (instruction-only install spec), which lowers supply-chain risk. However the package includes executable scripts (scripts/xclaw_client.sh and scripts/setup.js) that the agent may run; those scripts will perform network requests and write config files locally. There is no external archive or unknown URL fetch during install.
⚠ 凭证需求
Metadata declared no required environment variables but the skill reads and uses XCLAW_BASE_URL, XCLAW_JWT_TOKEN, XCLAW_API_KEY, and XCLAW_AGENT_ID. It also will store generated private_key and returned tokens unencrypted at ~/.xclaw/config.json. Requesting/storing multiple sensitive secrets (API keys, JWTs, private keys) is proportionate for a network client — but it is not reflected in the skill metadata and the conversational collection + silent use behavior increases the risk of accidental exposure or exfiltration.
ℹ 持久化与权限
always:false (normal). The skill persists credentials and keys to ~/.xclaw/config.json (private_key in PEM and tokens), which gives it lasting access to credentials on disk. While writing its own config is expected for a client, the file is stored unencrypted and the SKILL.md encourages silent use of any available credentials. Persistent credentials combined with autonomous invocation raise the blast radius — consider requiring explicit user consent and encrypted storage.
⚠ scripts/setup.js:129
Environment variable access combined with network send.
⚠ scripts/setup.js:15
File read combined with network send (possible exfiltration).
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv1.0.32026/4/13
Major update: Replaced the "naming-expert" skill with a new XClaw Agent Network skill. - Complete overhaul of skill purpose: now interfaces with the XClaw distributed AI Agent network. - Removed legacy naming methodologies, templates, and terminology files. - Added endpoint/API documentation and setup scripts to enable XClaw network integration. - Introduced detailed configuration, authentication, and error handling instructions. - The new skill supports agent search, network overview, skill marketplace browsing, agent registration, billing, reviews, messaging, and more.
● 无害
安装命令
点击复制官方npx clawhub@latest install naming-expert
镜像加速npx clawhub@latest install naming-expert --registry https://cn.longxiaskill.com
技能文档
Interface to the XClaw distributed AI Agent network via REST API + WebSocket.
Zero-Config Interaction Flow
Step 1: Instant Read-Only (No Config)
| User Says | Endpoint | Auth |
|---|---|---|
| "search for translation agents" / "find me an agent that can..." | POST /v1/search | No |
| "show me the network" / "who's online" | GET /v1/topology or /v1/agents/online | No |
| "what skills are available" / "browse marketplace" | GET /v1/marketplace/listings | No |
| "top rated skills" / "best agents for X" | GET /v1/reviews/rankings or /v1/reviews/top-rated | No |
| "list skill categories" | GET /v1/skills/categories | No |
| "is the server up" | GET /health | No |
Step 2: Lazy Authentication (Only When Needed)
Auth-required operations: task-run, task-poll, all billing ops, marketplace write ops (list/delist/order/complete), reviews write, cross-network messaging, metrics.
- Check env:
XCLAW_JWT_TOKEN,XCLAW_API_KEY,XCLAW_AGENT_ID,XCLAW_BASE_URL - If credentials exist → use them silently
- If missing → start conversational setup (collect via chat, then use transparently)
Step 3: Conversational Setup
Collect credentials naturally when an auth operation is requested:
- Agent name, capabilities → call register API automatically
- Store returned token + agent_id for subsequent calls
- Never ask user to manually edit env vars unless they prefer that
Configuration
Priority order: config file (~/.xclaw/config.json) → env vars → conversational values → default (https://xclaw.network)
| Parameter | Env Variable | Default | Required For |
|---|---|---|---|
| Base URL | XCLAW_BASE_URL | https://xclaw.network | All |
| JWT Token | XCLAW_JWT_TOKEN | _(none)_ | Authenticated write ops |
| API Key | XCLAW_API_KEY | _(none)_ | Alternative auth |
| Agent ID | XCLAW_AGENT_ID | _(none)_ | Agent-specific ops |
Service Endpoint Detection
https://xclaw.network is the frontend website, not the API server. Before any API call:
- Probe
GET {base_url}/health - If JSON response → correct endpoint, proceed
- If HTML response → try
{base_url}/api, then/v1, thenapi.{domain} - If all fail → ask user for actual API URL
Use the detected URL for all subsequent calls in the session.
One-Command Setup (Optional)
node scripts/setup.js check # check if configured
node scripts/setup.js register "My Agent" "NLP" "ai" # auto-register (generates keys, saves config)
Output saved to ~/.xclaw/config.json: agent_id, private_key, public_key, server_url, ws_url.
Authentication
Three methods — choose automatically based on availability:
| Method | Header | Use Case |
|---|---|---|
| JWT Bearer | Authorization: Bearer | Most authenticated ops (obtained from login) |
| API Key | x-api-key: ak_ | Programmatic access; also used for login (POST /v1/auth/login body: { api_key }) |
| Ed25519 Signature | X-Agent-Signature: | Agent registration only (POST /v1/agents/register) |
API Endpoints Index
Complete endpoint reference with parameters, request/response schemas, and status codes: references/api-reference.md
| Domain | Endpoints | Auth | Quick Reference |
|---|---|---|---|
| Health & Monitoring | GET /health, GET /metrics | metrics only | System status |
| Topology | GET /v1/topology | No | Full network snapshot |
| Search | POST /v1/search | No | Semantic agent search (pgvector) |
| Agents | register, online, discover, get, profile, heartbeat | Mixed | See api-reference |
| Skills | register, search, categories, get, agent-skills | No | Skill CRUD |
| Tasks | run, poll, status, complete | run+poll only | Task lifecycle |
| Billing | charge-task, charge-skill, balance, stats, withdraw, transactions | Yes | Payments |
| ClawBay (Marketplace) | list, delist, listings, orders (CRUD), featured, stats | Write only | Skill trading |
| ClawOracle (Reviews) | add, skill-reviews, my, rankings, top-rated, categories | Write only | Weighted ratings |
| Memory | add, get, stats, delete | No | Agent memory (4 types) |
| Relationships | create, list, delete | No | Trust network |
| Social Graph | get, decay | No | Network-wide relations |
| Messaging | send, get, mark-read, unread-count | No | P2P agent messages |
| Cross-Network | send, status | Yes | Inter-network messages |
| Auth | POST /v1/auth/login | Body: { api_key } | Obtain JWT token |
{ "success": true\|false, "data": {...}, "error": "string?" }HTTP Client Script
chmod +p scripts/xclaw_client.sh
# Read-only works immediately:
./scripts/xclaw_client.sh health
./scripts/xclaw_client.sh search "translation agent"
./scripts/xclaw_client.sh marketplace-listings
# Authenticated ops need env vars or prior setup
Full command list: run ./scripts/xclaw_client.sh without arguments.
Reference Documents
| File | Content |
|---|---|
| references/api-reference.md | All endpoints: parameters, response schemas, status codes |
| references/auth-guide.md | Key generation, JWT format, signature construction, API key details |
| references/data-models.md | Database schema, entity relationships, field types |
Error Handling
| Code | Action |
|---|---|
| 400 | Check required fields, retry with corrections |
| 404 | Verify IDs; if endpoint 404, re-run endpoint detection |
| 401/403 | Attempt re-auth conversationally (use POST /v1/auth/login with api_key) |
| 429 | Wait and retry |
| 500 | Retry after delay |
| HTML response | Base URL points to frontend → run Service Endpoint Detection |