Skip to content

CHANGELOG.md

Latest commit

 

History

History
135 lines (105 loc) · 6.18 KB

CHANGELOG.md

File metadata and controls

135 lines (105 loc) · 6.18 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

v1.2.0 (2026-07-31)

Features

  • Cost sync: sync Command Code model pricing to CLI agent configs
    • commandcode-bridge cost-sync CLI command with interactive agent selection
    • Detects installed CLI agents (OpenCode, Aider, Goose)
    • Filters models by bridge provider only (matches "Command Code" in provider name, localhost only)
    • Reads user's configured models from each agent's config
    • Adds/updates per-model cost fields (input, output, cache_read per 1M tokens)
    • Hardcoded pricing table with 44 models, exactly matching the bridge /v1/models API
    • Live API validation: pricing is checked against the bridge /v1/models before syncing
    • Lists ALL bridge providers found in config (OpenAI and Anthropic compatible), not just one
    • JSONC parser with trailing comma support for OpenCode configs
  • TUI page 7 "Cost Sync" with keymap [7]
    • Shows ALL Command Code models with pricing (grouped: Open Source, Premium)
    • Models in bridge config: bright white + "(will be implemented)"
    • Models not in config: greyed out
    • [c] keymap triggers sync with progress feedback (hint shown on the page)
    • [up/down] agent selection when multiple agents detected

v1.1.1 (2026-07-29)

Fixes

  • Ctrl+C: status bar notification only (no quit panel). Use [q] to quit.
  • Auto-increment port: if configured port is in use, bridge tries port+1, +2, ... up to 100 attempts, logs the switch
  • Help screen now includes -v / --version flag

v1.1.0 (2026-07-28)

Features

  • Anthropic-compatible Messages API (/v1/messages and /messages)
    • Full SSE streaming: content_block_start/delta/stop + message_start/delta/stop
    • Tool use blocks with input_json_delta streaming
    • Thinking blocks (reasoning mapped to Anthropic format)
    • Non-streaming endpoint with proper Anthropic response shape
    • Anthropic Messages protocol -> CC wire format conversion (system, tools, messages)
    • stop_reason mapping: tool-calls -> tool_use, length -> max_tokens
  • Self-update CLI command
    • commandcode-bridge update -- download latest release .tgz from GitHub and install via npm install -g
    • --version / -v flag prints version string
    • Update cache in ~/.config/commandcode-bridge/update-cache.json (1-hour TTL, avoids rate limits)

TUI

  • Dual endpoint header: OpenAI (http://127.0.0.1:{port}/v1) + Anthropic (http://127.0.0.1:{port})
  • Keymaps: [o] copies OpenAI URL, [a] copies Anthropic base URL
  • Ctrl+C now opens quit confirmation panel (not force exit)
    • Quit panel: Y / Enter / second Ctrl+C all confirm quit
    • Status bar shows red quit hint when panel is open
  • Footer, help page, and Proxy Config (page 6) synced with both endpoints

Refactor

  • Server code split into three clean files:
    • server_controller.dart -- HTTP server, request routing, shared endpoints (/models, /health, /token, /info)
    • openai_handler.dart -- OpenAI-compatible handler (streaming + non-streaming)
    • anthropic_handler.dart -- Anthropic-compatible handler (streaming + non-streaming)
  • Common upstream body builder shared between handlers
  • Removed diagnostic/debug logging; production-level logging only
  • Renamed ProxyServer -> ServerController, extracted handlers to own classes
  • Update OpenCode protocol mapping for production upstream format
  • Fixed config.date field missing from OpenAI handler (was broken after refactor)
  • Extracted version into lib/src/models/version.dart constant

Docs

  • README.md trimmed to preview; detailed docs moved to docs/
  • New docs: INSTALL.md, API-REFERENCE.md, TUI.md, ARCHITECTURE.md
  • API-REFERENCE.md covers OpenAI, Anthropic, and Zed client configurations
  • AGENTS.md synced with new file structure, endpoints, and naming conventions

v1.0.0 (2026-07-27)

Initial stable release of CommandCode Bridge, a single-account proxy bridge for Command Code CLI with TUI dashboard.

Features

  • TUI dashboard with 6 info pages (account, plan, usage, rate limits, models, proxy config)
  • Progress bars for credit usage, billing period, success rate, token ratios, rate limits
  • Plan-aware model sorting (Go = opensource first, Pro+ = premium first)
  • OpenAI-compatible HTTP proxy (/v1/chat/completions, /v1/models, /v1/health, /v1/token, /v1/info)
  • Real-time log sidebar with fullscreen mode
  • Port configuration panel with availability scan (persisted across restarts)
  • Clipboard integration for endpoint URL and model codenames
  • Color-coded notification system (success, info, warning, error)
  • Cross-platform support: Linux (primary), macOS (experimental), Windows (experimental)

Proxy

  • Full OpenAI-compatible chat completions (streaming + non-streaming)
  • Mapping of system, tool, and assistant messages to Command Code wire format
  • Tool-call round-trip support (assistant tool calls -> tool results)
  • Completion token translation (max_tokens / max_completion_tokens)
  • Correct UTF-8 multi-byte character handling in streaming responses
  • finish_reason override to "tool_calls" when tool calls present

TUI

  • Auto-refresh active page with background refresh
  • Header displaying last-used model, updateable in real-time
  • Relative time display on billing period and reset countdowns
  • Log panel toggle (Ctrl+L) with half-screen width
  • 44 models with plan-coded access display (green accessible, grey blocked)
  • Scrollable model list with Enter-to-copy

Distribution

  • npm tarball packaging with native Dart binary
  • Cross-platform binary compilation (Linux, macOS, Windows)
  • Zero overhead Node.js launcher (commandcode-bridge command)
  • No Dart SDK required for end-users
  • GitHub Actions CI/CD: test, release, and post-release validation workflows

Fixes

  • Signal handling: SIGINT/SIGTERM properly captured in headless server mode
  • UTF-8 chunk boundary decoding: no more FormatException on multi-byte sequences
  • Non-streaming chunk-boundary leftover handling
  • Tool-call preservation across subsequent tool-result turns
  • Log ordering: newest entries displayed at top

Infrastructure

  • GitHub Actions matrix build across 3 platforms
  • Automated tarball packaging via npm pack
  • Post-release user simulation (install, smoke test, uninstall)
  • Bug report issue template