docs: add CHANGELOG.md, fix README category count

Author: dorukardahan

CHANGELOG.md

@@ -0,0 +1,59 @@
+# Changelog
+
+## [3.0.0] — 2026-04-05
+
+Complete architecture rewrite. Skill-based routing replaced with OpenClaw plugin.
+
+### Breaking Changes
+- Anthropic/Claude removed — subscriptions no longer cover OpenClaw ([source](https://x.com/bcherny/status/2040206440556826908))
+- AGENTS.md routing snippets replaced by `before_model_resolve` plugin hook
+- All example configs restructured — old `claude-*` and `specialist-agents` directories removed
+- Minimum OpenClaw version raised to 2026.4.2+ (was 2026.2.6+)
+
+### Added
+- **Plugin-based routing** via OpenClaw `before_model_resolve` hook (<1ms, same session, zero token overhead)
+- **Two-stage routing**: capability filter (context window, vision, TTFT) → benchmark ranking
+- **6 subscription providers**: Google, OpenAI, Kimi, Z AI (GLM), MiniMax, Alibaba (Qwen)
+- **201 models** with 15 benchmark categories from Artificial Analysis API v2
+- **Risk-tiered failure policy**: low/medium/high risk classifications
+- **Routing observability**: decision log at `~/.openclaw/logs/zeroapi-routing.log`
+- **Benchmark staleness warnings**: 30-day warning, 60-day hard gate
+- **Per-job cron model assignment** (was per-workspace)
+- **Vision detection heuristic** for image-related prompts
+- **Word-boundary matching** to prevent false positive keyword matches
+- **Config validation** with runtime shape checking
+- **48 tests** across 6 test suites (classifier, filter, selector, config, logger, integration)
+- **5 example configs**: google-only, google-openai, google-openai-glm, google-openai-glm-kimi, full-stack
+
+### Changed
+- Benchmark methodology updated to AA Intelligence Index v4.0.4 (10 benchmarks, 4 categories)
+- Coding index reweighted: 0.85 × terminalbench + 0.15 × scicode (was 0.67 + 0.33)
+- Orchestration uses composite: 0.6 × tau2 + 0.4 × ifbench (was tau2 only)
+- Fast category enforces TTFT < 5s hard filter
+- Fallback chains limited to max 3 per category (cross-provider required)
+
+### Removed
+- Anthropic provider and all Claude model references
+- AGENTS.md routing snippet approach (replaced by plugin)
+- `sessions_spawn` delegation model (replaced by same-session model switching)
+- Old example directories: claude-only, claude-codex, claude-gemini, specialist-agents
+
+## [2.3.0] — 2026-02-15
+
+Cross-provider fallback chains and LLM-friendly documentation.
+
+## [2.2.0] — 2026-02-15
+
+Progressive disclosure restructure — SKILL.md as entry point, references/ for details.
+
+## [2.1.1] — 2026-02-12
+
+Bug fixes.
+
+## [2.1.0] — 2026-02-12
+
+Flash-Lite model ID fix.
+
+## [2.0.0] — 2026-02-10
+
+Initial public release with benchmark-driven routing skill.

README.md

@@ -40,7 +40,7 @@ The model is switched for that turn only. The session, conversation history, and
 
 ## Task Categories
 
-The plugin matches keywords in each message to one of five routing categories. No match stays on the default model.
+The plugin matches keywords in each message to one of six routing categories. No match stays on the default model.
 
 | Category | Primary Benchmark | Routing Signals | Example Prompts |
 |----------|------------------|-----------------|-----------------|