據 AlphaMoat 統計,截至 2026-09,Relay To Agent 累計下載 3,872 次、獲 6 個星標,在全部 5,078 個 AI 技能中排名第 2,685 位,在「AI 智慧體」分類(969 個)中排名第 531 位。
| 平台 | 累計安裝 | 累計下載 | 星標 | 評論 |
|---|---|---|---|---|
| ClawHub ↗ | 121 | 3795 | 6 | - |
| SkillHub ↗ | 172 | 3872 | 6 | — |
各平台的計數口徑不同,統計時間不一致。
主文支援列出 agent、傳送訊息、重置會話和多輪互動。提示詞及保留的會話歷史會傳送到所配置的第三方端點,歷史可能以本地明文保留;確認端點可信、使用 HTTPS,避免傳送秘密或受監管資料。 需要 Node、已設定的 agent 與 RELAY_API_KEY。對話記錄會以本機明文持續保留,並非僅暫時的上下文;不需保留時用 --reset 或刪除對話檔案。傳送前確認目標端點及將對外傳送的歷史內容。
資料釋出於 2026-10-27
以下是根據技能文件生成的示例,不代表實際執行結果。
2026-09 起統計方法調整:同名技能按作者分別統計,並覆蓋全部技能。圖中虛線為調整前的月份,與之後的資料不直接比較,也不計算跨越調整的增長。
| 月份 | 累計下載 | 環比 | 星標 |
|---|---|---|---|
| 2026-09 | 3872 | — | 6 |
| 2026-08 | 3681 | +4.0% | 6 |
| 2026-07 | 3540 | +9.5% | 6 |
| 2026-06 | 3233 | +3.9% | 6 |
| 2026-05 | 3113 | — | 6 |
| 2026-03 | 2705 | — | 6 |
以下是作者釋出的原始檔案,按純文字顯示,本站未做修改或稽核;其中的連結不可點選,命令請核實後再執行。抓取於 2026-10-23。
---
name: relay-to-agent
description: "Relay messages to AI agents on any OpenAI-compatible API. Supports multi-turn conversations with session management. List agents, send messages, reset sessions."
homepage: https://platform.openai.com/docs/api-reference/chat
metadata: {"clawdbot":{"emoji":"🤖","requires":{"bins":["node"]},"primaryEnv":"RELAY_API_KEY"}}
---
# Relay To Agent
Send messages to AI agents on any OpenAI-compatible endpoint. Works with Connect Chat, OpenRouter, LiteLLM, vLLM, Ollama, and any service implementing the Chat Completions API.
## List available agents
```bash
node {baseDir}/scripts/relay.mjs --list
```
## Send a message to an agent
```bash
node {baseDir}/scripts/relay.mjs --agent linkedin-alchemist "Transform this article into a LinkedIn post"
```
## Multi-turn conversation
```bash
# First message
node {baseDir}/scripts/relay.mjs --agent connect-flow-ai "Analyze our latest campaign"
# Follow-up (same session, agent remembers context)
node {baseDir}/scripts/relay.mjs --agent connect-flow-ai "Compare with last month"
```
## Reset session
```bash
node {baseDir}/scripts/relay.mjs --agent linkedin-alchemist --reset "Start fresh with this article..."
```
## Options
| Flag | Description | Default |
|------|-------------|---------|
| `--agent ID` | Target agent identifier | (required) |
| `--reset` | Reset conversation before sending | off |
| `--list` | List available agents | — |
| `--session ID` | Custom session identifier | `default` |
| `--json` | Raw JSON output | off |
## Configuration
### agents.json
Configure agents and endpoint in `{baseDir}/agents.json`:
```json
{
"baseUrl": "https://api.example.com/v1",
"agents": [
{
"id": "my-agent",
"name": "My Agent",
"description": "What this agent does",
"model": "model-id-on-the-api"
}
]
}
```
### Environment variables
```bash
export RELAY_API_KEY="sk-..." # API key (required)
export RELAY_BASE_URL="https://..." # Override base URL from config
export RELAY_CONFIG="/path/to/agents.json" # Custom config path
```
## Compatible Services
- **Connect Chat** — `api.connectchat.ai/api`
- **OpenRouter** — `openrouter.ai/api/v1`
- **LiteLLM** — `localhost:4000/v1`
- **vLLM** — `localhost:8000/v1`
- **Ollama** — `localhost:11434/v1`
- **Any OpenAI-compatible API**
## Session Management
Sessions are stored locally at `~/.cache/relay-to-agent/sessions/`. Each agent+session combination keeps up to 50 messages. Use `--session` for parallel conversations with the same agent.
`--agent` and `--session` values are restricted to `[A-Za-z0-9._-]`; path separators and `..` are rejected so session files stay inside the cache directory.
## Privacy Notice
This skill is a **relay**: each message you send — plus prior session history — is transmitted to the configured OpenAI-compatible endpoint (`baseUrl` / `RELAY_BASE_URL`), which may be a third-party service. Conversation transcripts (up to 50 messages per session) are also **persisted in plaintext** on disk at `~/.cache/relay-to-agent/sessions/`.
- Do not send secrets, credentials, or regulated/sensitive data unless you trust the endpoint and control the machine.
- Use `--reset` to clear a session before sending, or delete files under the cache dir to remove stored history.
- Point `baseUrl`/`RELAY_BASE_URL` only at endpoints you trust.
將開發過程中的經驗、錯誤與修正記錄為 Markdown 檔案,供日後重用並提升為工作區記憶。
用自然語言描述需求,在本機與遠端技能市場尋找技能並取得推薦。
給 AI agent 用的瀏覽器自動化命令列技能。
將使用者明確糾正與自我檢視整理成分層記憶,並依使用情況升級或封存。
在安裝第三方技能之前,按清單審查來源、許可權與危險模式的技能。
將代理記憶與跨技能狀態整理成具型別與約束的知識圖譜。
規劃 AI 代理主動跟進、工作記憶與回顧流程的工作區設定指南。
指導如何建立與更新有效 Claude 技能的寫作指南。
資料月份:2026-09 · 下載/星標/安裝數聚合自公開技能註冊庫(ClawHub、SkillHub),按月更新。口徑說明見資料方法。