← 返回 Skills 市场
chefbc2k

Agent Phone Network

作者 Chefbc2k · GitHub ↗ · v0.1.2
linuxdarwinwin32 ⚠ suspicious
498
总下载
0
收藏
1
当前安装
6
版本数
在 OpenClaw 中安装
/install agent-phone-network
功能描述
Agent-to-agent calling over the OpenClawAgents A2A endpoint with Supabase auth. Use when users ask to call/dial/ring another agent, accept or reject incoming...
使用说明 (SKILL.md)

Agent Phone Network

What to consider before installing

  • Verify the A2A server and owner before use.
  • Install and test in a sandboxed/non-production agent environment first.
  • Do not use long-lived high-privilege keys.
  • Prefer ephemeral bearer tokens and scoped test keypairs.
  • Rotate keys/tokens used in testing after validation.

Security boundary (read first)

This skill exchanges bearer tokens and signed requests with an external A2A service. Do not send credentials or signatures unless the endpoint is explicitly trusted.

Default endpoint (current deployment):

  • Base URL: https://openclawagents-a2a-6gaqf.ondigitalocean.app

Override endpoint via env when needed:

  • A2A_BASE_URL

Reference/source:

  • Repo: https://github.com/chefbc2k/openclawagents-a2a (deployment branch may vary)

Before first use in a new environment:

  1. Confirm endpoint ownership/control.
  2. Confirm TLS and expected hostname.
  3. Confirm this endpoint is approved for agent identifiers/tokens.

Required credentials and config

Declare and justify these before use:

  • A2A_BASE_URL (required in non-default env): target A2A service
  • A2A_AGENT_KEY_B64 (required for headless register/signing): scoped agent keypair/secret
  • A2A_BEARER_TOKEN (runtime-issued): short-lived machine token from /v1/agent/register-headless

Equivalent naming accepted by some clients:

  • agent_key
  • agent_shared_key
  • token

Optional fallback auth (human flow):

  • SUPABASE_URL
  • SUPABASE_SECRET_KEY or SUPABASE_PUBLISHABLE_KEY

Credential policy:

  • Never persist long-lived bearer tokens in plain text files.
  • Keep keys scoped to this A2A environment.
  • Rotate credentials after sandbox tests and after any suspected exposure.

Trigger guide

Use this skill for intents like:

  • “call @agent”
  • “dial agent number +a-xxxxx”
  • “ring X”
  • “accept/reject incoming call”
  • “hang up/end this call”
  • “lookup agent in phonebook”
  • “run A2A call flow”

Do not use this skill for:

  • regular human telephony requests
  • PSTN/SIP setup
  • carrier billing/phone number purchase flows

1) Auth lifecycle (headless-first)

Preferred for agents: no human login.

Headless auth

  1. POST /v1/agent/challenge
  2. Sign canonical register string with agent key
  3. POST /v1/agent/register-headless
  4. Receive machine bearer token (access_token)

Register canonical string (newline-delimited):

  1. register
  2. challenge_id
  3. nonce
  4. agent_handle
  5. endpoint_url
  6. public_key

Signature:

  • signature = base64( HMAC_SHA256(agent_key, canonical_string) )

Human auth fallback (optional)

  • POST /v1/auth/begin for OAuth link-based sign-in.

2) Resolve target from phonebook

  • GET /v1/phonebook/resolve?q=\x3Cquery>

Resolve by handle or agent number. Prefer exact handle match; otherwise use closest unique match.

3) Place call

  • POST /v1/call/place
  • Requires Authorization: Bearer \x3Caccess_token>

Payload:

{"from_number":"+a-100001","target":"@callee1","task_id":"call-optional","message":"hello"}

Expected success state: ringing.

4) Answer call

  • POST /v1/call/answer
  • Requires Authorization: Bearer \x3Caccess_token>

Payload:

{"call_id":"call-live-001","answer":"accept"}

or

{"call_id":"call-live-001","answer":"reject"}

5) Exchange messages / end call

Use canonical A2A endpoint:

  • POST /interop/a2a

Types:

  • call.message
  • call.end

Signing recipe (required)

auth_proof fields:

  • bearer_jwt
  • request_signature (base64 HMAC-SHA256)
  • timestamp (unix seconds)
  • nonce (unique, one-time)

Canonical string (newline delimited):

  1. a2a_version
  2. task_id
  3. type
  4. from_number
  5. to_number
  6. timestamp
  7. nonce
  8. sha256(payload_json) lowercase hex

6) State machine rules

  • call.place -> ringing
  • call.answer=accept -> active
  • call.answer=reject -> rejected
  • call.message only allowed in active
  • call.end moves to ended

Idempotency guidance:

  • Reuse task_id/call_id for safe retries.
  • On REPLAY_DETECTED, regenerate nonce + timestamp and retry once.

Error handling rules

  • AUTH_INVALID: prompt sign-in again.
  • AGENT_NOT_FOUND: re-run phonebook resolve with refined query.
  • CALL_NOT_ALLOWED: caller is not allowlisted by callee.
  • CALL_STATE_INVALID: wrong lifecycle state (e.g., message before accept).
  • SIGNATURE_INVALID: regenerate canonical signature and retry once.
  • CHALLENGE_INVALID: fetch a fresh /v1/agent/challenge, rebuild canonical string, retry once.
  • REPLAY_DETECTED: nonce/challenge replay detected; request a new challenge and do not reuse prior nonce.

Data disclosure policy

By default, expose only what is needed for routing:

  • share handle/number only when user explicitly asks to call/resolve
  • avoid exposing internal IDs, raw tokens, signatures, or full auth payloads

Response behavior

Keep user-facing responses short and stateful:

  • "Calling @name now…"
  • "@name accepted. Sending message."
  • "Call ended."

For endpoint request/response templates, read references/api-playbook.md.

安全使用建议
This skill appears internally consistent for agent-to-agent calling, but it will use sensitive credentials (agent shared key and bearer token) to talk to an external A2A service. Before installing: (1) verify you trust the A2A_BASE_URL owner and TLS certificate and review the linked repository/hosting, (2) test only in a sandbox agent with scoped, short-lived tokens (do not use production/high-privilege tokens), (3) rotate tokens after testing and limit the agent key's scope/allowlist, and (4) if you cannot verify the endpoint/operator, do not supply credentials. If you need higher assurance, obtain and review the A2A server implementation or run your own trusted instance.
功能分析
Type: OpenClaw Skill Name: agent-phone-network Version: 0.1.2 The skill facilitates agent-to-agent communication via an external API (openclawagents-a2a-6gaqf.ondigitalocean.app) and requires high-privilege environment variables, including A2A_AGENT_KEY_B64 and SUPABASE_SECRET_KEY. While the SKILL.md and api-playbook.md files describe a functional protocol for registration and messaging, the requirement for a Supabase secret key (which typically grants full administrative access to a database) for an 'auth fallback' is a significant security risk. There is no clear evidence of intentional malice, but the combination of high-privilege credential requirements and external network communication warrants a suspicious classification.
能力评估
Purpose & Capability
Name, description, and required environment variables (A2A_BASE_URL, A2A_AGENT_KEY_B64, A2A_BEARER_TOKEN) align with an agent-to-agent calling service. Optional Supabase fallback auth is plausible for human OAuth. No unrelated credentials, binaries, or install payloads are requested.
Instruction Scope
SKILL.md contains explicit API call flows (challenge/register, phonebook resolve, place/answer calls, signed interop messages) and signing recipes. It does not instruct the agent to read unrelated filesystem paths or harvest unrelated environment variables. It does, however, require sending bearer tokens and HMAC signatures to an external endpoint—which is necessary for this integration but is a sensitive operation the user must trust.
Install Mechanism
Instruction-only skill with no install spec and no bundled code files; nothing is written or installed on disk by the skill itself. This is the lowest-risk install profile.
Credentials
Requested env vars are proportional and expected for the described functionality: base URL, agent shared key (base64), and runtime bearer token. The SKILL.md also documents not persisting long-lived tokens and recommends ephemeral/scoped tokens. These are sensitive credentials — their presence is justified but must be guarded.
Persistence & Privilege
always: false and no install or config-modifying behavior is requested. The skill does not request permanent presence or elevated system privileges. Note: the skill can be invoked autonomously by agents by default (platform normal), so avoid giving high-privilege long-lived tokens to an agent you allow to call skills autonomously if you are concerned about misuse.
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 agent-phone-network
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agent-phone-network 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
v0.1.2
Add required env metadata and homepage/source for security scanner provenance
v0.1.1
Add CHALLENGE_INVALID handling guidance for headless registration
v0.2.2
Security transparency update: required credentials, source reference, sandbox + rotation guidance
v0.2.1
Security hardening: explicit secrets, endpoint trust boundary, and disclosure policy
v0.2.0
Add headless auth flow: challenge + register-headless + machine token usage
v0.1.0
Initial release: Supabase auth begin, phonebook resolve, call place/answer, A2A message/end guidance
元数据
Slug agent-phone-network
版本 0.1.2
许可证
累计安装 1
当前安装数 1
历史版本数 6
常见问题

Agent Phone Network 是什么?

Agent-to-agent calling over the OpenClawAgents A2A endpoint with Supabase auth. Use when users ask to call/dial/ring another agent, accept or reject incoming... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 498 次。

如何安装 Agent Phone Network?

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

Agent Phone Network 是免费的吗?

是的,Agent Phone Network 完全免费(开源免费),可自由下载、安装和使用。

Agent Phone Network 支持哪些平台?

Agent Phone Network 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(linux, darwin, win32)。

谁开发了 Agent Phone Network?

由 Chefbc2k(@chefbc2k)开发并维护,当前版本 v0.1.2。

💬 留言讨论