claude-overnight 1.53.0 → 1.54.0

package/README.md

@@ -576,10 +576,11 @@ A fixed-plan `tasks.json` (without `flexiblePlan: true`) bypasses orchestration
 
 The `src/prompt-evolution/` engine and `claude-overnight-evolve` CLI power a self-evolution pipeline that optimises prompts (the planner prompt here, MCP-browser's supervisor prompts, or any prompt in a user's repo) via Pareto-frontier mutation with LLM-as-judge and heuristic scoring.
 
-**Multi-hour runs aren't meant for your laptop.** Two ways to run it:
+**Multi-hour runs aren't meant for your laptop.** Three ways to run it:
 
 1. **`npx claude-overnight-evolve …`** — quickest. Fine for smoke tests or short runs; needs `ANTHROPIC_API_KEY` in env and keeps running only as long as your shell is open. Output: `~/.claude-overnight/prompt-evolution/<runId>/`.
 2. **Self-hosted Docker** — [`self-host/`](self-host/README.md) ships a tiny runner image + optional HTTP server (enqueue + read-back) you can run on any VPS. Laptop can be off.
+3. **Fornace hosted** — already have a fornace project? `POST /api/projects/:id/prompt-evolution/enqueue` runs the same engine in your project's container. Body: `{ prompt, target, evalModel, generations, population, env?, anthropicApiKey?, anthropicBaseUrl?, anthropicModel? }`. Poll `GET /:runId` for status + inline `report.md`. See the [fornace integration doc § 2.6](https://github.com/Fornace/MCP-Browser/blob/main/docs/integration.md#26-prompt-evolution--apiprojectsidprompt-evolution).
 
 Experiment credentials — any Anthropic-compatible provider (Anthropic direct, OpenRouter, Kimi, DashScope, a local proxy) — are injected via env vars: `ANTHROPIC_BASE_URL`, `ANTHROPIC_API_KEY`, `EVAL_MODEL`, `MUTATE_MODEL`. Self-host reads them from `self-host/.env` (or per-run `env:` in the enqueue body).
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-overnight",
-  "version": "1.53.0",
+  "version": "1.54.0",
   "description": "Overnight parallel coding agents in git worktrees, with a self-curating skill memory that improves while the run is going. Mix Claude Opus as planner, Kimi 2.6 or Cursor composer-2 as cheap fast worker, Gemini or Qwen for bulk implementation. Multi-wave autonomous loop that plans, executes, reviews, and steers itself until the objective is met. Crash-safe resume, rate-limit aware, usage cap preserves headroom for your interactive Claude Code.",
   "type": "module",
   "bin": {

package/plugins/claude-overnight/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-overnight",
-  "version": "1.53.0",
+  "version": "1.54.0",
   "description": "Claude Code skill for understanding, installing, and inspecting claude-overnight runs: overnight parallel coding agents in git worktrees with a self-curating skill memory, multi-wave steering, three-layer review, and crash-safe resume. Mix Opus planner with Kimi 2.6, Cursor composer-2, Gemini, Qwen, or any Anthropic-compatible worker.",
   "author": {
     "name": "Francesco Fornace"