Merge nucleic/sleek-thistle-egret-fyej into dev

corpus/replay.py

@@ -7,7 +7,12 @@ separately and does NOT count against parity (error-message wording legitimately
 differs between shells); a stderr-only difference is reported for review.
 
 Usage: replay.py [--nash PATH] [--bash PATH] [--corpus PATH] [--json PATH]
+                 [--observe-spool DIR]
 Parity target (M0/M3 gate): >= 99%.
+
+--observe-spool DIR turns nash observation ON during the replay (spool
+transport; the env is set identically for bash, where it is inert), proving
+the gate hooks don't perturb command semantics (docs/NASH.md §10.1).
 """
 
 import argparse
@@ -115,8 +120,17 @@ def main():
         default=os.path.join(os.path.dirname(os.path.abspath(__file__)), "corpus.jsonl"),
     )
     ap.add_argument("--json", help="also write the full report to this path")
+    ap.add_argument("--observe-spool", help="enable nash observation, spooling to this dir")
     args = ap.parse_args()
 
+    observe_env = None
+    if args.observe_spool:
+        os.makedirs(args.observe_spool, exist_ok=True)
+        observe_env = {
+            "NUCLEIC_SHELL_SPOOL": args.observe_spool,
+            "NUCLEIC_SESSION_ID": "corpus-replay",
+        }
+
     entries = []
     with open(args.corpus) as f:
         for line in f:
@@ -126,8 +140,8 @@ def main():
 
     results = []
     for entry in entries:
-        b = run_one(args.bash, entry["cmd"])
-        n = run_one(args.nash, entry["cmd"])
+        b = run_one(args.bash, entry["cmd"], extra_env=observe_env)
+        n = run_one(args.nash, entry["cmd"], extra_env=observe_env)
         verdict = classify(b, n)
         results.append({"id": entry["id"], "cmd": entry["cmd"], "verdict": verdict,
                         "bash": b, "nash": n})