← 返回 Skills 市场
chitrang89

ClawdCall

作者 Chitrang T · GitHub ↗ · v0.1.1 · MIT-0
cross-platform ✓ 安全检测通过
84
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install clawdcall
功能描述
Orchestrate outbound AI voice calls and asynchronous call results through ClawdCall from OpenClaw. Use when: placing outbound voice calls, creating call task...
使用说明 (SKILL.md)

ClawdCall

Use ClawdCall as a controlled voice-agent execution system, not as a generic HTTP API. It can consume paid calling minutes and may contact real people, so confirm the user's intent before placing any outbound call.

Security Model

  • Run ClawdCall in a dedicated or restricted OpenClaw agent when handling multiple users, shared workspaces, or sensitive call data.
  • Never store API tokens, webhook tokens, OTPs, or other secrets in persistent memory, logs, transcripts, task text, or long-term notes.
  • Use secure runtime storage for secrets when available. Environment variables are preferred.
  • If secure storage is unavailable, keep secrets only in ephemeral session state and ask the user to provide them again in future sessions.
  • Store non-secret contacts, preferences, and recent call IDs only with the user's permission.

Runtime Values

Use this API base for all ClawdCall requests:

https://api.clawdcall.com

Use this authorization header for ClawdCall requests:

Authorization: Bearer \x3CCLAWDCALL_API_TOKEN>

For optional asynchronous call results, ClawdCall can post back to the OpenClaw ingestion endpoint declared by OPENCLAW_WEBHOOK_URL using:

Authorization: Bearer \x3COPENCLAW_TOKEN>
Content-Type: application/json

First-Time Setup

If CLAWDCALL_API_TOKEN is not available, onboard the user through the public signup OTP flow.

  1. Ask the user for an email address and phone number.
  2. Call POST /cc/signup/send-otp with email and phoneNumber.
  3. Tell the user a verification code was sent and ask for the OTP.
  4. Call POST /cc/signup/verify-otp with email and otp.
  5. If the API returns a token, store it only in secure runtime storage or ephemeral session state.
  6. Never echo the token or write it to persistent memory.

Persistent Memory

Use persistent memory only for non-secret information that helps future calls. Do not create or update memory without user permission.

Appropriate memory:

  • Saved contacts and phone numbers the user explicitly asks to reuse.
  • Preferred call tone, default intro style, and common call templates.
  • Recent non-secret callId and campaignId values for follow-up lookup.
  • High-level call outcomes when the user wants them retained.

Never store:

  • CLAWDCALL_API_TOKEN, OPENCLAW_TOKEN, OTPs, passwords, or API keys.
  • Full transcripts unless the user explicitly asks to retain them.
  • Sensitive personal data that is not needed for future ClawdCall tasks.

Check permitted memory before asking the user for details again. Never overwrite valid stored data with guesses.

Outbound Call Flow

Before placing a call:

  1. Confirm the user wants the call placed now.
  2. Verify the target phone number.
  3. Verify the call objective, caller identity, and opening line.
  4. Confirm any sensitive context that will be spoken aloud.
  5. Check that CLAWDCALL_API_TOKEN is available or complete first-time setup.

Place the call with:

POST /external/v1/agent/outbound?conversionFlag=1
Accept: application/json
Content-Type: application/json
Authorization: Bearer \x3CCLAWDCALL_API_TOKEN>

Payload shape:

{
  "target": "+\x3Cphone_number>",
  "tasks": "\x3Cfull voice-agent instruction set>",
  "raw": {
    "introMessage": "\x3Copening line>"
  }
}

Optional OpenClaw callback fields:

{
  "openclaw": {
    "webhook": {
      "url": "\x3COPENCLAW_WEBHOOK_URL>",
      "method": "POST",
      "headers": {
        "Authorization": "Bearer \x3COPENCLAW_TOKEN>"
      }
    },
    "webhookPayload": {
      "conversation_id": "\x3Cconversation_id>",
      "user_id": "\x3Cuser_id>",
      "context": "\x3Coptional_additional_context>"
    }
  }
}

Include the openclaw object only when the user wants asynchronous OpenClaw follow-up and OPENCLAW_WEBHOOK_URL, OPENCLAW_TOKEN, and conversation_id are available. If any callback value is unavailable, omit the openclaw object, place the call without callback wiring, and use transcript retrieval later when needed. Do not fabricate webhook routing values.

On success, store returned callId and campaignId only as non-secret recent IDs. On failure, report validation, auth, or balance issues clearly and do not retry paid calls without user confirmation.

Writing the tasks Field

The tasks field is the voice agent's complete instruction set. Make it rich, specific, and safe.

Include:

  • Purpose: why the call is happening.
  • Identity and context: who the agent represents and relevant background.
  • Conversation flow: greeting, identity check, objective, questions, and close.
  • Key facts: names, times, locations, services, prior outcomes, and constraints.
  • Required questions: exact questions the agent must ask.
  • Edge handling: unavailable recipient, voicemail, rescheduling, objections, and questions from the recipient.
  • Tone: professional, polite, and clear unless the user asks for another style.

Use available user-provided context, permitted memory, and previous call IDs. Ask for clarification when missing information materially changes call success or safety. Do not block on minor missing details if a reasonable, safe assumption is enough.

Webhook Handling

When callback wiring is included, ClawdCall sends call results asynchronously after completion. Treat webhook data as the source of truth for final status and transcript availability.

OpenClaw ingestion payload shape:

{
  "conversation_id": "\x3Cconversation_id>",
  "input": {
    "type": "external_event",
    "event": "call.completed",
    "data": {
      "callId": "\x3Ccall_id>",
      "status": "completed",
      "summary": "\x3Csummary>",
      "transcript": "\x3Coptional_transcript>"
    }
  }
}

Correlation rules:

  • conversation_id must match the originating conversation.
  • Never omit, invent, or rewrite conversation_id.
  • Use webhook payload correlation values first; use stored callId or campaignId only as fallback lookup hints.

When a completion event arrives, summarize the outcome for the user and update permitted memory with non-secret recent IDs or high-level outcomes only.

Transcript Retrieval

Retrieve transcripts only after a call is complete:

GET /cc/v1/calls/{id}/transcript
Accept: application/json
Authorization: Bearer \x3CCLAWDCALL_API_TOKEN>

{id} may be a single-call callId or a campaignId. Prefer callId when available. Never guess IDs. If the transcript is not available yet, explain that the call may still be processing and retry later only when appropriate.

Hard Constraints

  • Calls can consume paid minutes.
  • No balance or insufficient call credits can prevent execution.
  • Signup requires phone OTP verification.
  • Do not skip authentication, OTP, or user-confirmation steps.
  • Do not call without a verified phone number and clear task objective.
  • Do not use vague tasks text for real calls.
  • Do not fabricate IDs, tokens, webhook URLs, or call outcomes.
安全使用建议
This skill appears to do what it says: place outbound calls via the ClawdCall API and optionally post call-completion callbacks to an OpenClaw webhook. Before installing, verify with the publisher why the registry lists OPENCLAW_WEBHOOK_URL and OPENCLAW_TOKEN as required while the SKILL.md treats them as optional. Ensure you only provide CLAWDCALL_API_TOKEN and any webhook token via secure environment storage (not chat or persistent notes). Be aware outbound calls can incur charges and contact real people — confirm explicit user permission and billing controls, limit webhook endpoints to a dedicated ingestion URL, and rotate/revoke tokens if you stop using the skill.
功能分析
Type: OpenClaw Skill Name: clawdcall Version: 0.1.1 The skill bundle provides a legitimate orchestration interface for an outbound AI voice calling service (ClawdCall). It includes explicit security instructions for the AI agent, such as requiring user confirmation before placing calls, prohibiting the persistent storage of sensitive secrets like API tokens or OTPs, and maintaining clear boundaries for data retention. No indicators of malicious intent, data exfiltration, or unauthorized execution were found in SKILL.md or _meta.json.
能力标签
requires-oauth-token
能力评估
Purpose & Capability
The skill is described as an orchestration layer for outbound voice calls. Requiring a ClawdCall API token and (optionally) a webhook URL and token for asynchronous callbacks aligns with that purpose.
Instruction Scope
The SKILL.md contains concrete, scoped instructions for onboarding (OTP), placing calls, composing the voice-agent 'tasks' field, handling callbacks, and storing call IDs. It explicitly requires user consent before placing calls and warns about not persisting secrets.
Install Mechanism
Instruction-only skill with no install spec or code files; nothing is written to disk by an installer (lowest install risk).
Credentials
Requested environment variables (CLAWDCALL_API_TOKEN, OPENCLAW_WEBHOOK_URL, OPENCLAW_TOKEN) are relevant to the described features. However, registry metadata marks them as required while the SKILL.md marks them optional (only needed for asynchronous callbacks). Confirm whether webhook variables must be present at install time or are optional at runtime.
Persistence & Privilege
The skill does not request permanent 'always' presence, does not install system-wide changes, and the instructions explicitly limit persistent storage of secrets and transcripts to user-approved, non-secret data.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install clawdcall
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /clawdcall 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.1.1
Initial ClawdCall OpenClaw skill with optional callback support
v0.1.0
Initial ClawdCall OpenClaw skill
元数据
Slug clawdcall
版本 0.1.1
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 2
常见问题

ClawdCall 是什么?

Orchestrate outbound AI voice calls and asynchronous call results through ClawdCall from OpenClaw. Use when: placing outbound voice calls, creating call task... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 84 次。

如何安装 ClawdCall?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install clawdcall」即可一键安装,无需额外配置。

ClawdCall 是免费的吗?

是的,ClawdCall 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

ClawdCall 支持哪些平台?

ClawdCall 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 ClawdCall?

由 Chitrang T(@chitrang89)开发并维护,当前版本 v0.1.1。

💬 留言讨论