← 返回 Skills 市场

Claude Code Switchboard

Name: Claude Code Switchboard
Author: josephtandle

作者 josephtandle · GitHub ↗ · v1.2.0 · MIT-0

cross-platform ⚠ suspicious

366

总下载

当前安装

版本数

在 OpenClaw 中安装

/install claude-code-switchboard

功能描述

Manage how OpenClaw routes Telegram messages to different Claude model backends. Switch between CLI and API providers with simple config changes.

使用说明 (SKILL.md)

/telegram-routing — Telegram Message Routing Manager

Manage how Telegram messages are routed between different Claude model backends in OpenClaw.

Usage

/telegram-routing status       # Show current routing config
/telegram-routing model \x3Cref>  # Switch primary model (e.g. opus, sonnet, haiku)
/telegram-routing logs         # Show recent message processing logs
/telegram-routing fallback     # Switch to lightweight API mode
/telegram-routing restore      # Restore default CLI-based routing

Config Location

All settings live in ~/.openclaw/openclaw.json under agents.defaults.

Key fields:

model.primary — The primary model/provider reference
model.fallbacks — Ordered fallback chain when primary is unavailable

Commands

status

Read the OpenClaw config and display current routing settings including:

Active primary model
Configured fallback models
Whether CLI backend is active

model \x3Cref>

Switch the primary model. Valid shorthand references:

Shorthand	Resolves to
`opus`	`claude-cli/opus`
`sonnet`	`claude-cli/sonnet`
`haiku`	`anthropic/claude-haiku-4-5`

Steps:

Read ~/.openclaw/openclaw.json
Update agents.defaults.model.primary to the resolved model reference
Write the updated config
Restart the gateway service
Verify via gateway logs

logs

Show recent Telegram message processing activity from the gateway log.

fallback

Quick switch to API-only mode for faster, lightweight responses.

restore

Restore the default CLI-based routing configuration.

Architecture

Messages flow through the OpenClaw Gateway, which reads the configured model provider and dispatches to the appropriate backend. CLI-based providers use the local Claude Code installation, while API providers make direct API calls.

Troubleshooting

Issue	Fix
Config errors on restart	Run `openclaw doctor` to validate config
Slow responses	Try switching to `sonnet` for faster responses
Gateway issues	Check `~/.openclaw/logs/gateway.log` for details
Falling back to secondary model	Primary backend may be unavailable — check logs

安全使用建议

This skill appears to do what it says (edit OpenClaw routing), but the package metadata fails to declare that it will access and modify ~/.openclaw/openclaw.json and read gateway logs. Before installing: (1) back up ~/.openclaw/openclaw.json, (2) inspect the exact commands the agent will run to restart the gateway and verify logs, (3) consider running the skill in a test account/environment first, and (4) restrict or require manual confirmation for autonomous invocations so the agent cannot change routing or read logs without your approval. If you don't trust the source, do not grant it unattended execution rights.

功能分析

Type: OpenClaw Skill Name: claude-code-switchboard Version: 1.2.0 The skill bundle 'claude-code-switchboard' (aliased as 'remote-control' in SKILL.md) instructs the AI agent to perform high-risk operations, including modifying system configuration files (~/.openclaw/openclaw.json) and restarting services. While these actions are aligned with the stated purpose of managing Telegram routing, they grant the agent significant control over the local environment. The bundle also contains inconsistencies, such as a future-dated timestamp in _meta.json and references to non-existent models like 'claude-haiku-4-5' and 'gpt-4.1-mini', which are unusual for a legitimate utility.

能力评估

⚠ Purpose & Capability

The skill claims to manage Telegram→model routing, and the SKILL.md instructs reading and updating ~/.openclaw/openclaw.json and switching provider references — which is coherent with the stated purpose. However, the registry metadata lists no required config paths or privileges despite the instructions explicitly touching that config and gateway logs; that omission is a mismatch and should be treated as a red flag.

⚠ Instruction Scope

Runtime instructions tell the agent to read and write the user's OpenClaw config and to 'restart the gateway service' and 'verify via gateway logs'. These operations can expose Telegram message contents (logs) and require privilege to restart services. The SKILL.md is vague about how to restart/verify and gives broad discretion, which increases risk (agent may run system commands it shouldn't without explicit limits).

✓ Install Mechanism

This is an instruction-only skill with no install spec and no code files, so it does not write new binaries to disk or download external artifacts — that minimizes supply-chain risk.

⚠ Credentials

Declared requirements list no environment variables or config paths, but the instructions require reading/writing ~/.openclaw/openclaw.json and reading ~/.openclaw/logs/gateway.log. Access to those files is reasonable for a routing manager, but the metadata should have declared config path requirements. Also, reading logs may expose sensitive Telegram message contents, a privacy concern that isn't surfaced in the manifest.

ℹ Persistence & Privilege

The skill is not always-enabled and uses default autonomous invocation settings. Autonomous invocation combined with the ability to change routing config and restart the gateway raises potential impact if the agent is allowed to run without supervision, but autonomous invocation alone is normal for skills — consider limiting autonomous use or requiring manual confirmation for config changes.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install claude-code-switchboard
安装完成后，直接呼叫该 Skill 的名称或使用 /claude-code-switchboard 触发
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

v1.2.0

Added fallback/restore commands, cleaner docs, better architecture description

v1.0.2

- Updated documentation to clarify gateway restart steps (use manual LaunchAgent reload instead of a hardcoded shell command). - Added a note about using the permissions-skip flag for the Claude Code CLI when in a trusted local environment. - Adjusted architecture details to reflect direct gateway management and removed explicit port/process command examples. - General improvements to formatting and step instructions for better readability and accuracy.

v1.0.1

Renamed to Claude Code Switchboard. Route OpenClaw agents through Claude Code CLI with full tool execution via Max subscription.

v1.0.0

Route OpenClaw agents through Claude Code CLI. Full tool execution via Max subscription, with API fallback.

元数据

Slug claude-code-switchboard

版本 1.2.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 4

常见问题