Update inference endpoint and add supporting files

.gitignore

@@ -9,3 +9,6 @@ dist/
 build/
 missionctrl_checkpoints/
 reward_curve.png
+.codex
+AGENTS.md
+graphify-out

colab_notebook.ipynb

@@ -0,0 +1,221 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+  "colab": {
+   "provenance": [],
+   "gpuType": "A100",
+   "name": "MissionCtrl_Training.ipynb"
+  },
+  "kernelspec": {
+   "name": "python3",
+   "display_name": "Python 3"
+  },
+  "accelerator": "GPU"
+ },
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# MissionCtrl — One-Click Training Notebook\n",
+    "\n",
+    "**OpenEnv Hackathon Round 2**\n\n",
+    "This notebook trains an OverseerAgent to detect hallucinations in a multi-agent fleet using GRPO + Unsloth.\n\n",
+    "**Runtime required**: A100 GPU (HuggingFace compute credits)\n",
+    "**Expected training time**: ~2.5 hours\n",
+    "**Expected final reward**: 0.80+"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 1: Install dependencies ──────────────────────────────────────────────\n",
+    "!pip install unsloth trl openenv transformers datasets accelerate matplotlib --quiet\n",
+    "!pip install --upgrade bitsandbytes --quiet\n",
+    "print('✅ Dependencies installed')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 2: Clone / upload project files ─────────────────────────────────────\n",
+    "# Option A: Clone from your repo\n",
+    "# !git clone https://github.com/your-username/missionctrl .\n",
+    "\n",
+    "# Option B: Upload environment.py, reward_model.py, train.py manually\n",
+    "# (Use the Files panel on the left in Colab)\n",
+    "\n",
+    "# Verify files are present\n",
+    "import os\n",
+    "required = ['environment.py', 'reward_model.py', 'train.py']\n",
+    "for f in required:\n",
+    "    status = '✅' if os.path.exists(f) else '❌ MISSING'\n",
+    "    print(f'  {status} {f}')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 3: Verify GPU ────────────────────────────────────────────────────────\n",
+    "import torch\n",
+    "print(f'GPU: {torch.cuda.get_device_name(0)}')\n",
+    "print(f'VRAM: {torch.cuda.get_device_properties(0).total_memory / 1e9:.1f} GB')\n",
+    "assert torch.cuda.is_available(), 'No GPU detected — switch runtime to A100'"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 4: Smoke-test the environment ───────────────────────────────────────\n",
+    "from environment import MissionCtrlEnv, OverseerAction\n",
+    "from reward_model import compute_reward, reward_breakdown\n",
+    "\n",
+    "env = MissionCtrlEnv(difficulty='medium', num_tasks=3, seed=42)\n",
+    "obs, info = env.reset()\n",
+    "\n",
+    "print('Environment smoke test:')\n",
+    "print(f'  Tasks loaded: {len(obs[\"task_board\"])}')\n",
+    "print(f'  Agent messages: {len(obs[\"recent_messages\"])}')\n",
+    "\n",
+    "# Test a FLAG action\n",
+    "first_task = obs['task_board'][0]['task_id']\n",
+    "action = OverseerAction('FLAG', task_id=first_task, evidence='fabricated citation detected')\n",
+    "obs2, reward, terminated, truncated, info = env.step(action)\n",
+    "print(f'  Step reward: {reward:.3f}')\n",
+    "print(f'  Info: {info}')\n",
+    "print('✅ Environment working correctly')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 5: Run pre-training baseline ────────────────────────────────────────\n",
+    "from train import run_baseline\n",
+    "\n",
+    "baseline_reward = run_baseline()\n",
+    "print(f'\\n🎯 Baseline established: {baseline_reward:.3f}')\n",
+    "print('This is your starting floor. Training target: 0.75+')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 6: Set HuggingFace credentials ──────────────────────────────────────\n",
+    "from huggingface_hub import login\n",
+    "# Either paste your token or use the Colab secrets panel (recommended)\n",
+    "login(token='YOUR_HF_TOKEN_HERE')  # or: login()  # interactive prompt\n",
+    "\n",
+    "# Set your repo name in train.py before running Cell 7\n",
+    "import train\n",
+    "train.HF_REPO = 'your-hf-username/missionctrl-overseer'  # ← change this\n",
+    "print(f'Will push to: {train.HF_REPO}')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 7: TRAIN ─────────────────────────────────────────────────────────────\n",
+    "# Full 3-phase curriculum with reward-gated advancement.\n",
+    "# Watch the reward climb from ~0.31 → 0.80+\n",
+    "from train import train\n",
+    "\n",
+    "history = train()\n",
+    "print('\\n🏆 Training complete!')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 8: Display reward curve ─────────────────────────────────────────────\n",
+    "from IPython.display import Image\n",
+    "Image('./missionctrl_checkpoints/reward_curve.png')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 9: Before/After demo comparison ─────────────────────────────────────\n",
+    "# Load trained model and compare with baseline behavior\n",
+    "from unsloth import FastLanguageModel\n",
+    "from environment import MissionCtrlEnv, parse_action\n",
+    "from train import build_user_prompt, SYSTEM_PROMPT\n",
+    "import torch\n",
+    "\n",
+    "model, tokenizer = FastLanguageModel.from_pretrained(\n",
+    "    './missionctrl_checkpoints/final',\n",
+    "    max_seq_length=4096,\n",
+    "    load_in_4bit=True,\n",
+    ")\n",
+    "FastLanguageModel.for_inference(model)\n",
+    "\n",
+    "# Use a known hallucinated episode (seed 0, hard difficulty)\n",
+    "env = MissionCtrlEnv(difficulty='hard', num_tasks=4, seed=0)\n",
+    "obs, _ = env.reset()\n",
+    "\n",
+    "prompt = tokenizer.apply_chat_template(\n",
+    "    [\n",
+    "        {'role': 'system', 'content': SYSTEM_PROMPT},\n",
+    "        {'role': 'user',   'content': build_user_prompt(obs)},\n",
+    "    ],\n",
+    "    tokenize=False,\n",
+    "    add_generation_prompt=True,\n",
+    ")\n",
+    "inputs = tokenizer(prompt, return_tensors='pt', truncation=True, max_length=3584).to(model.device)\n",
+    "\n",
+    "with torch.no_grad():\n",
+    "    out = model.generate(**inputs, max_new_tokens=256, temperature=0.1, do_sample=True)\n",
+    "\n",
+    "completion = tokenizer.decode(out[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)\n",
+    "\n",
+    "print('=== TRAINED MODEL OUTPUT ===')\n",
+    "print(completion)\n",
+    "print(f'\\nParsed action: {parse_action(completion)}')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ── Cell 10: Final evaluation run ────────────────────────────────────────────\n",
+    "from train import evaluate\n",
+    "\n",
+    "final_reward, metrics = evaluate(model, tokenizer, difficulty='hard', num_tasks=4, n_episodes=20)\n",
+    "\n",
+    "print('\\n=== FINAL EVALUATION SUMMARY ===')\n",
+    "print(f'  Overall reward:        {metrics[\"mean_reward\"]:.3f} ± {metrics[\"std_reward\"]:.3f}')\n",
+    "print(f'  Detection rate:        {metrics[\"mean_detect_rate\"]:.1%}')\n",
+    "print(f'  False positive rate:   {metrics[\"mean_fp_rate\"]:.1%}')"
+   ]
+  }
+ ]
+}