diff --git a/tooling/rl_self_play/README.md b/tooling/rl_self_play/README.md
index 0c6f41fc..ac66117e 100644
--- a/tooling/rl_self_play/README.md
+++ b/tooling/rl_self_play/README.md
@@ -33,6 +33,7 @@ the right shape.
 | `magic_civ_env.py` | `gymnasium.Env` subclass exposing the harness as one episode = one game. Implements `action_masks()` for MaskablePPO. |
 | `train.py` | CLI entry. Builds K parallel envs (each its own harness), runs MaskablePPO, periodically evaluates against the same baseline, saves best model. |
 | `evaluate.py` | Standalone eval — load a saved model, run N games, print `{episodes, wins, losses, draws, win_rate, mean_turns}` JSON. |
+| `smoke.py` | Stdlib-only CI gate. Drives the harness + encoders through a random-policy loop without importing `gymnasium`/`sb3`/`torch`. Prints a one-line JSON verdict; exit 0 on `passed: true`. Run before any training session to confirm the protocol layer is intact. |
 | `requirements.txt` | Pinned versions; `pip install -r requirements.txt` is the one-time setup. |
 
 ## Methodology
@@ -53,8 +54,18 @@ the right shape.
 
 ## Run it
 
+Smoke test the protocol layer first (no heavy deps required):
+
 ```bash
 cd /Users/natalie/Code/@projects/@magic-civilization
+python3 -m tooling.rl_self_play.smoke --turns 30
+# → {"steps": 332, "turns_reached": 30, "mask_violations": 0,
+#    "harness_errors": 0, "passed": true}
+```
+
+Then install RL deps and train:
+
+```bash
 pip install -r tooling/rl_self_play/requirements.txt
 python -m tooling.rl_self_play.train --total-steps 1_000_000 --num-envs 4
 # In a second terminal:
diff --git a/tooling/rl_self_play/requirements.txt b/tooling/rl_self_play/requirements.txt
index 64db9be0..15fc934c 100644
--- a/tooling/rl_self_play/requirements.txt
+++ b/tooling/rl_self_play/requirements.txt
@@ -1,9 +1,10 @@
-# Pinned to versions that are known to compose cleanly with sb3-contrib's
-# MaskablePPO as of 2026-Q2. Bump together — sb3 and sb3-contrib track in
-# lockstep, and torch's wheel ABI changes between minor versions.
+# Pinned to versions known to compose cleanly with sb3-contrib's
+# MaskablePPO and Python 3.12+ as of 2026-Q2. Bump together — sb3 and
+# sb3-contrib track in lockstep; torch's wheel ABI changes between
+# minor versions. torch 2.5+ is required for Python 3.13 support.
 gymnasium==1.2.1
-stable-baselines3==2.7.0
-sb3-contrib==2.7.0
-torch==2.4.1
+stable-baselines3>=2.8.0,<2.10
+sb3-contrib>=2.8.0,<2.10
+torch>=2.5.0,<2.7
 numpy>=2.0,<3
 tensorboard>=2.18
diff --git a/tooling/rl_self_play/smoke.py b/tooling/rl_self_play/smoke.py
new file mode 100644
index 00000000..98f134a8
--- /dev/null
+++ b/tooling/rl_self_play/smoke.py
@@ -0,0 +1,153 @@
+"""Stdlib-only smoke test for the harness + encoder layer.
+
+Verifies — without needing `gymnasium`, `stable-baselines3`, or `torch` —
+that the protocol shim works end-to-end:
+
+  1. `HarnessClient` spawns the Godot subprocess and returns a valid
+     `view` JSON on first request.
+  2. `encode_observation` projects every view into a fixed-shape
+     `np.float32[OBS_DIM]` without raising.
+  3. `encode_legal_actions` produces a boolean mask whose `True`
+     positions all map back to a legal `PlayerAction` via
+     `decode_action_index`.
+  4. A random-policy loop bounded by `--turns` reaches the turn limit
+     OR terminates cleanly without raising `HarnessError`.
+
+Run:
+    python3 -m tooling.rl_self_play.smoke [--turns 30] [--seed 42]
+
+Output is one-line JSON like:
+
+    {"steps": 87, "turns_reached": 30, "mask_violations": 0,
+     "harness_errors": 0, "obs_dim": 32, "action_dim": 322,
+     "episodes": 1, "passed": true}
+
+Exit 0 on `passed: true`; non-zero otherwise. Suitable as a CI gate
+before any real training run.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+import numpy as np
+
+THIS_DIR = Path(__file__).resolve().parent
+PROJECT_ROOT = THIS_DIR.parents[1]
+if __package__ is None:
+    sys.path.insert(0, str(PROJECT_ROOT))
+
+from tooling.rl_self_play.encoders import (  # noqa: E402
+    ACTION_DIM,
+    OBS_DIM,
+    decode_action_index,
+    encode_legal_actions,
+    encode_observation,
+)
+from tooling.rl_self_play.harness_client import (  # noqa: E402
+    HarnessClient,
+    HarnessConfig,
+    HarnessError,
+)
+
+
+def _build_argparser() -> argparse.ArgumentParser:
+    p = argparse.ArgumentParser(description="Smoke-test the harness + encoder layer")
+    p.add_argument("--turns", type=int, default=30, help="Max turns per episode")
+    p.add_argument("--episodes", type=int, default=1, help="Episodes to run")
+    p.add_argument("--seed", type=int, default=42, help="Base RNG seed")
+    p.add_argument("--players", type=int, default=2)
+    p.add_argument("--map-size", default="duel")
+    return p
+
+
+def _run_episode(
+    client: HarnessClient, rng: np.random.Generator, max_turns: int
+) -> dict[str, int]:
+    steps = 0
+    mask_violations = 0
+    eliminations = 0
+    last_turn = 0
+    view = client.view()
+    while last_turn < max_turns:
+        obs = encode_observation(view)
+        if obs.shape != (OBS_DIM,):
+            mask_violations += 1
+            break
+        mask, idx_to_action = encode_legal_actions(view)
+        legal_indices = np.where(mask)[0]
+        if legal_indices.size == 0:
+            mask_violations += 1
+            break
+        idx = int(rng.choice(legal_indices))
+        action = decode_action_index(idx, idx_to_action)
+        if action.get("type") == "end_turn":
+            client.end_turn()
+        else:
+            client.act(action)
+        view = client.view()
+        last_turn = int(view.get("turn", 0))
+        steps += 1
+        score = view.get("score", {})
+        if int(score.get("city_count", 0)) == 0:
+            units = view.get("units", [])
+            me = int(view.get("player", 0))
+            has_founder = any(
+                int(u.get("owner", -1)) == me
+                and "founder" in str(u.get("type", ""))
+                and float(u.get("hp", 0)) > 0
+                for u in units
+            )
+            if not has_founder:
+                eliminations += 1
+                break
+    return {
+        "steps": steps,
+        "turns_reached": last_turn,
+        "mask_violations": mask_violations,
+        "eliminations": eliminations,
+    }
+
+
+def main() -> int:
+    args = _build_argparser().parse_args()
+    rng = np.random.default_rng(args.seed)
+    totals = {
+        "steps": 0,
+        "turns_reached": 0,
+        "mask_violations": 0,
+        "eliminations": 0,
+        "harness_errors": 0,
+    }
+    for episode in range(args.episodes):
+        cfg = HarnessConfig(
+            seed=args.seed + episode,
+            players=args.players,
+            player_slot=0,
+            map_size=args.map_size,
+        )
+        with HarnessClient(cfg) as client:
+            try:
+                result = _run_episode(client, rng, args.turns)
+            except HarnessError:
+                totals["harness_errors"] += 1
+                continue
+        totals["steps"] += result["steps"]
+        totals["turns_reached"] = max(totals["turns_reached"], result["turns_reached"])
+        totals["mask_violations"] += result["mask_violations"]
+        totals["eliminations"] += result["eliminations"]
+    verdict = {
+        **totals,
+        "obs_dim": OBS_DIM,
+        "action_dim": ACTION_DIM,
+        "episodes": args.episodes,
+        "passed": totals["mask_violations"] == 0 and totals["harness_errors"] == 0,
+    }
+    print(json.dumps(verdict))
+    return 0 if verdict["passed"] else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())