v1.0.8

v1.0.8

Highlights

• Deduplication of tool calls: Completely resolved duplicate tool call issues within a single assistant message where coding tools (such as Claude Code) simultaneously send tool execution information in both the content array and the tool_calls field.
• Standardization to tool-call: Uniformly reformatted all tool calls inside assistant (assistant) messages and their corresponding execution results (tool messages) to match the CommandCode API specification:

{
  "type": "tool-call",
  "toolCallId": "...",
  "toolName": "...",
  "input": {...}
}

v1.0.7

v1.0.7

Highlights

• Completely fixed the ModelMessage[] schema structure error when coding tools continue conversations after executing tool calls.
• Perfect compatibility with Claude Code, Codex, Kilo Code, and Open Code when running complex tasks that require executing multiple tool calls sequentially.
• Rebuilt and compiled binary files for Windows, Linux AMD64, and Linux ARM64.

Fixes

• Converted the format of tool results sent to the API from the old format:

{
  "role": "tool",
  "content": [
    {
      "type": "tool_result",
      "tool_use_id": "...",
      "content": "..."
    }
  ]
}

v1.0.6

CommandCode Proxy Server v1.0.6

Highlights

• Fixed CommandCode ModelMessage[] schema error when continuing conversations after tool execution.
• Improved compatibility with coding tools such as Claude Code, Codex, Kilo Code, and other OpenAI-compatible clients.
• Rebuilt Windows, Linux AMD64, and Linux ARM64 binaries.

Fixes

• Fixed converted OpenAI tool result messages to use CommandCode role: "tool" instead of role: "user".

Previous converted shape:

{
  "role": "user",
  "content": [
    {
      "type": "tool_result",
      "tool_use_id": "...",
      "content": "..."
    }
  ]
}

v1.0.5

CommandCode Proxy Server v1.0.5

Highlights

• Fixed CommandCode tool result message schema for coding tools.
• Improved compatibility for Claude Code, Codex, Kilo Code, and other OpenAI-compatible coding clients when continuing after tool execution.
• Rebuilt Windows, Linux AMD64, and Linux ARM64 binaries.

Fixes

• Changed converted tool result content from the old shape:

{
  "type": "tool-result",
  "toolCallId": "...",
  "toolName": "...",
  "text": "..."
}

v1.0.4

CommandCode Proxy Server v1.0.4

Highlights

• Improved compatibility with OpenAI-style coding clients such as Claude Code, Codex, Kilo Code, and other tools that rely on chat completions, tool calls, and streaming responses.
• Added broader OpenAI request/response handling for tool use, multimodal-style content arrays, and common client fields.
• Added a minimal /v1/responses compatibility endpoint.
• Added /chat/completions as an alias for /v1/chat/completions.
• Added debug logging toggle in main.go for inspecting client requests and CommandCode upstream stream events during troubleshooting.

Changes

• Added support for additional OpenAI-compatible request fields:
  • stream_options
  • tool_choice
  • parallel_tool_calls
  • response_format
  • stop
  • top_p
  • presence_penalty
  • frequency_penalty
  • user
• Added support for OpenAI Responses-style input through POST /v1/responses.
• Improved conversion of OpenAI tool definitions into CommandCode tool schema.
• Improved conversion of assistant tool calls and tool result messages into CommandCode-compatible message content.
• Added handling for CommandCode tool stream events:
  • tool-use
  • tool-delta
  • tool-input-start
  • tool-input-delta
  • tool-call
• Normalized CommandCode finish reasons such as tool-calls into OpenAI-compatible tool_calls.
• Improved support for non-streaming tool-call responses with OpenAI-compatible tool_calls.
• Improved handling of OpenAI multimodal/content array message formats.
• Converted image content parts into text placeholders for better upstream compatibility.
• Added OpenAI-style JSON error responses.
• Improved upstream error status forwarding for CommandCode 4xx responses.
• Added debug logging helpers with output truncation to avoid excessive log size.
• Updated project version to v1.0.4.

Endpoints

Available endpoints:

POST /v1/chat/completions
POST /chat/completions
POST /v1/responses
GET  /v1/models
GET  /health

v1.0.3

What's Changed

• Fix tool call and tool result message conversion for CommandCode API
• Fix tool call streaming response (tool-use, tool-delta events)
• Fix non-streaming response finish_reason for tool calls
• Pass client tools to CommandCode API
• Always send stream: true to CommandCode API

v1.0.2

What's Changed

• Fix: always send stream: true to CommandCode API
• Update version to v1.0.2

v1.0.1

What's Changed

• Add multimodal content support (text + images)
• Add tool calls support
• Fix stream flag to respect client's setting
• Update version to v1.0.1

v1.0.0

v1.0.0

Initial release of CommandCode Proxy Server.

Highlights

• Added an OpenAI-compatible proxy server for CommandCode API.
• Supports /v1/chat/completions, /v1/models, and /health.
• Supports both streaming and non-streaming chat completion responses.
• Added configurable CLI options for host, port, API key, and version output.
• Added model alias mapping for supported CommandCode models.
• Added GitHub tag-based update checking.
• Added automatic CommandCode CLI version lookup from npm registry for upstream headers.
• Added project documentation and ignore rules.

CLI

command-code-proxy -host 127.0.0.1 -port 55990