feat: add reasoning/thinking output support for OpenAI and Anthropic routes

fix: Anthropic route thinking blocks validation + context token display

fix: support thinking/redacted_thinking content blocks in Anthropic schema

feat: add image input support for all API routes

fix: normalize tool schemas missing `properties` field

fix: window counter review issues — opacity modifier, catch-up spam, rounding

feat: add per-window usage counters with time-driven reset

fix: stop resetting local usage counters on rate limit window rollover

feat: add per-account empty response tracking with email in logs

fix: detect empty Codex responses and auto-retry with account rotation (#18)

chore: deduplicate model selector, auto-extract chromium version, harden update pipeline

fix: harden prompt injection suppression and non-streaming error handling

fix: filter non-Codex models from backend to prevent resolveModelId fallback breakage

fix: handle error SSE events from Codex API instead of silently dropping them

fix: support thinking.type "adaptive" for latest Claude Code

fix: Docker container auto-detect host proxy via host.docker.internal

fix: use known-profile list instead of runtime detection for TLS impersonation

fix: auto-fallback for unsupported TLS impersonate profiles

feat: dynamic model list fetching from Codex backend (#9)

feat: OpenCode platform support, vitest tests, changelog & Docker hardening

feat: native tool calling support across all protocols (#8)

fix(api): return 400 invalid_json for malformed chat payload

refactor: remove dead code, unused config, and fix type violations

feat: accept tool/function messages across all protocols

feat: suppress Codex Desktop directives in model responses

feat: support developer role and array content in OpenAI protocol

fix: startup log shows config proxy_api_key instead of random hash

fix: proxy_api_key from config not used in frontend or auth validation

feat: add pnpm/bun package manager support for local development

fix: prevent .env inline comments from being parsed as JWT tokens

feat: add libcurl-impersonate FFI transport for Chrome TLS fingerprint

refactor: architecture audit fixes round 2 (P0-P2)

feat: migrate dashboard to Preact + Vite

refactor: architecture audit fixes (P0-P2 stability & reliability)

refactor: fingerprint alignment, auto-maintenance, Docker deployment

fix: remove fake Claude/Gemini model aliases, use dynamic catalog

refactor: extract shared proxy handler, eliminate route duplication

fix: remove dead login.html reference causing UI fallback on fresh installs

feat: auto-detect local proxy (mihomo/clash/v2ray)

feat: auto proxy detection for curl + tighten dashboard layout

ui: redesign dashboard with Tailwind CSS + wire backend APIs

fix: remove all `any` types from source code

feat: multi-protocol support — Anthropic Messages API + Google Gemini API

feat: desktop context injection, multi-turn sessions, TLS impersonation & auto-update

feat: anti-detection hardening — curl TLS, auto-version, jitter & Chromium headers

feat: reactive API config, live code examples, quota display & curl fix

feat: native OAuth PKCE login with fixed localhost:1455 redirect

feat: dashboard multi-account management UI with OAuth login (#3)

feat: architecture audit — disguise hardening, update automation, robustness

refactor: remove dead code, extract hardcoded values into config