@caeliq/claude-code-router 2.1.7 → 2.1.9

package/README.md

@@ -20,6 +20,7 @@ Claude Code Router is an adaptive LLM gateway for Claude Code. It routes each re
 
 This fork is based on [claude-code-router](https://github.com/musistudio/claude-code-router) and includes several enhancements:
 
+- **Codex breaking change — required chaining:** Codex is no longer a converting owner. Providers **must** set `"transformer": { "use": ["openai-responses", "codex"] }`. A `codex`-only chain is rejected. `openai-responses` owns the Responses wire (including encrypted reasoning items); `codex` is ChatGPT auth/headers middleware (`store: false`, `stream: true`). Responses clients passthrough `input[]`.
 - **Improved LLM Support**: Fixed streaming for Gemini/Gemma and enhanced OpenAI API handling.
 - **Reasoning & Streaming Refactor**: Modularized streaming and reasoning logic into reusable utilities for better maintainability.
 - **Mistral Integration**: Added specific handling for Mistral's reasoning parameters and decoupled transformation logic.
@@ -27,7 +28,7 @@ This fork is based on [claude-code-router](https://github.com/musistudio/claude-
 - **Code Quality**: Localized codebase (English comments), improved error handling, and addressed Copilot review feedback.
 - **Dependency Security**: Keeps the workspace audit-clean with scoped security floors. Docusaurus's archived `image-size` dependency is replaced only on the MDX loader edge by a private `image-size/fromFile` compatibility adapter backed by maintained `probe-image-size`; it is build tooling and does not enter CCR runtime packages.
 - **Gemini Stability & Tool Use Fixes**: Corrected `thoughtSignature` placement in Gemini request bodies (Gemini 3 expects it as a sibling field on the `functionCall` part itself, and validates only the first such part per step); filtered synthetic `ccr_` placeholder signatures from outgoing Gemini requests to prevent Gemini 500 errors; fixed `tool_result` content-array serialization in the Anthropic transformer so models receive plain text instead of JSON-wrapped arrays (resolves "Error editing file" in Claude Code); fixed Fastify `onSend` hook to prevent `invalid type 'object'` unhandled rejections on error responses.
-- **Codex (ChatGPT) Integration**: Added Codex transformer for the ChatGPT backend API (Responses API), supporting both OAuth-based authentication (`ccr codex-auth`) and PAT auth via `api_key: "at-..."`, plus SSE streaming, reasoning/thinking content, tool calls with web search, and image handling.
+- **Codex (ChatGPT) Integration**: ChatGPT backend over the Responses API. **Requires** `"use": ["openai-responses", "codex"]`. Supports OAuth (`ccr codex-auth`) and PAT (`api_key: "at-..."`), SSE streaming, encrypted reasoning replay, tool calls with web search, and image handling.
 - **Gateway and Codex model discovery**: The server's OpenAI-compatible `GET /v1/models` enriches configured models with native-provider metadata from models.dev (friendly names, descriptions, context/output limits, reasoning effort levels, and Claude Desktop family tiers). `MODEL_ID_OUTPUT` controls whether discovery emits literal canonical IDs (the default) or masks otherwise-filtered IDs as reversible `claude-<lowercase UTF-8 hex>` aliases. Inbound always accepts both forms. Separately, the CLI's `ccr codex-config` command uses the same exact-model/native-provider matching rule for Codex's local picker catalog: a unique model-id row is authoritative, duplicate ids are resolved to the native provider, and a true miss falls back to a 200K context window with `low`/`medium`/`high` effort. It also wires Codex's `config.toml` to route over the Responses API.
 - **Cursor SDK Integration**: Added `cursor-sdk` transformer that runs Cursor models in-process via `@cursor/sdk`. Default **bridge** mode keeps Claude Code as the tool host (Cursor built-ins denied); supports `plan` / `agent` modes, `crsr_` / `CURSOR_API_KEY` auth, `ccr model get cursor` model discovery, and Docker runtime install of the SDK native packages.
 - **Claude Subscription Integration**: Added `claude-auth` support for routing through a Claude Pro or Max subscription via OAuth (`ccr claude-auth`), using the `claude-auth` + `Anthropic` transformer chain.
@@ -74,7 +75,8 @@ The `config.json` file has several key sections:
 
 - **`PROXY_URL`** (optional): You can set a proxy for API requests, for example: `"PROXY_URL": "http://127.0.0.1:7890"`. Loopback targets (`localhost`, `127.0.0.1`, `::1`) always bypass the proxy. Hosts listed in the `NO_PROXY` / `no_proxy` environment variables also bypass it (comma-separated hosts, `.domain` / `*.domain`, CIDR, optional `:port`).
 - **`LOG`** (optional): You can enable logging by setting it to `true`. When set to `false`, no log files will be created. Default is `true`.
-- **`LOG_LEVEL`** (optional): Set the logging level. Available options are: `"fatal"`, `"error"`, `"warn"`, `"info"`, `"debug"`, `"trace"`. Default is `"debug"`.
+- **`LOG_LEVEL`** (optional): Set the logging level. Available options are: `"fatal"`, `"error"`, `"warn"`, `"info"`, `"debug"`, `"trace"`. Default is `"info"`.
+- **`LOG_SSE_EVENTS`** (optional): When `true`, logs every SSE event on both **provider→CCR** and **CCR→client** while `LOG_LEVEL` is `debug`. Default is `false` — debug still records terminal cache/usage summaries without per-delta file I/O. Pair with `LOG_REQUEST_BODY` to also capture request bodies on **client→CCR** and **CCR→provider** (all protocols). Each record includes a `direction` field.
 - **Logging Systems**: The Claude Code Router uses two separate logging systems:
   - **Server-level logs**: HTTP requests, API calls, and server events are logged using pino in the `~/.claude-code-router/logs/` directory with filenames like `ccr-*.log`
   - **Application-level logs**: Routing decisions and business logic events are logged in `~/.claude-code-router/claude-code-router.log`