[codex] fix nested spec parsing in CLI

cli/trellis

@@ -267,28 +267,89 @@ def yaml_read_nested(text, parent, field):
     return None
 
 
+def parse_numeric_scalar(value):
+    """Parse a YAML scalar that should contain a numeric value."""
+    if value in (None, "", "null", "{}"):
+        return None
+    try:
+        return float(str(value).strip().strip('"').strip("'"))
+    except (TypeError, ValueError):
+        return None
+
+
+def extract_self_eval_score(text):
+    """Read the recorded self-eval score from supported spec shapes."""
+    for parent in ("self_eval", "perf_eval"):
+        for field in ("total", "score"):
+            score = parse_numeric_scalar(yaml_read_nested(text, parent, field))
+            if score is not None:
+                return score
+        score = parse_numeric_scalar(yaml_read_field(text, parent))
+        if score is not None:
+            return score
+
+    return parse_numeric_scalar(yaml_read_field(text, "score"))
+
+
+def parse_phase_statuses(text):
+    """Parse phase statuses from the phases block without counting unrelated statuses."""
+    lines = text.splitlines()
+    statuses = []
+    in_phases = False
+    i = 0
+
+    while i < len(lines):
+        line = lines[i]
+        stripped = line.strip()
+        indent = len(line) - len(line.lstrip())
+
+        if not in_phases:
+            if re.match(r'^phases:\s*$', line):
+                in_phases = True
+            i += 1
+            continue
+
+        if stripped and indent == 0:
+            break
+
+        match = re.match(r'^\s+-\s+id:\s*"?(phase\d+)"?', line)
+        if not match:
+            i += 1
+            continue
+
+        item_indent = len(match.group(0)) - len(match.group(0).lstrip())
+        status = "pending"
+        i += 1
+
+        while i < len(lines):
+            field_line = lines[i]
+            field_stripped = field_line.strip()
+            field_indent = len(field_line) - len(field_line.lstrip())
+
+            if field_stripped and field_indent <= item_indent:
+                break
+
+            status_match = re.match(
+                r'^\s+status:\s*"?(pending|in_progress|completed|failed|skipped)"?',
+                field_line,
+            )
+            if status_match:
+                status = status_match.group(1)
+
+            i += 1
+
+        statuses.append(status)
+
+    return statuses
+
+
 def count_phases(text):
     """Count phases and their statuses."""
-    total = 0
-    completed = 0
-    failed = 0
-    in_progress = 0
-    for m in re.finditer(r'^\s+-\s+id:\s*"?phase\d+"?', text, re.MULTILINE):
-        total += 1
-    for m in re.finditer(r'^\s+status:\s*"?completed"?', text, re.MULTILINE):
-        completed += 1
-    for m in re.finditer(r'^\s+status:\s*"?failed"?', text, re.MULTILINE):
-        failed += 1
-    for m in re.finditer(r'^\s+status:\s*"?in_progress"?', text, re.MULTILINE):
-        in_progress += 1
-    # Subtract the top-level status from phase counts
-    top_status = yaml_read_field(text, "status")
-    if top_status == "completed":
-        completed = max(0, completed - 1)
-    elif top_status == "failed":
-        failed = max(0, failed - 1)
-    elif top_status == "in_progress":
-        in_progress = max(0, in_progress - 1)
+    phase_statuses = parse_phase_statuses(text)
+    total = len(phase_statuses)
+    completed = sum(1 for status in phase_statuses if status == "completed")
+    failed = sum(1 for status in phase_statuses if status == "failed")
+    in_progress = sum(1 for status in phase_statuses if status == "in_progress")
     return total, completed, failed, in_progress
 
 
@@ -1073,18 +1134,16 @@ def cmd_start(args):
 
 def check_self_eval(text, task_id):
     """Warn if self-eval looks like a rubber stamp."""
-    # Look for self_eval or perf_eval score
-    score_match = re.search(r'(?:self_eval|perf_eval|score):\s*(\d+)', text)
-    if not score_match:
+    score = extract_self_eval_score(text)
+    if score is None:
         print(f"  {c(C_YELLOW, 'warn')}: no self-eval score found in spec")
         return
-
-    score = int(score_match.group(1))
     has_deviations = bool(re.search(r'deviations:', text, re.MULTILINE))
     has_improvements = bool(re.search(r'improvements:', text, re.MULTILINE))
+    score_display = int(score) if float(score).is_integer() else score
 
     if score >= 9 and not has_deviations and not has_improvements:
-        print(f"  {c(C_YELLOW, 'warn')}: self-eval {score}/10 with no deviations or improvements noted")
+        print(f"  {c(C_YELLOW, 'warn')}: self-eval {score_display}/10 with no deviations or improvements noted")
         print(f"         scores above 8 should document at least one deviation or improvement")
     elif score == 10:
         print(f"  {c(C_YELLOW, 'note')}: perfect 10/10 - are you sure? 10 means flawless with improvements beyond spec")
@@ -1112,7 +1171,7 @@ def cmd_complete(args):
     text = spec.read_text()
 
     # Check for exec results - warn if no criteria were executed
-    has_results = bool(re.search(r'result:\s*"?(pass|fail)"?', text, re.MULTILINE))
+    has_results = any(ac.get("result") in ("pass", "fail") for ac in parse_acceptance_criteria(text))
     if not has_results:
         print(f"  {c(C_YELLOW, 'warn')}: no exec results recorded. Run '{c(C_BOLD, f'trellis exec {args.task_id}')}' first")
 
@@ -2333,13 +2392,14 @@ def cmd_report(args):
         risks[risk] = risks.get(risk, 0) + 1
 
         # Self-eval scores
-        score_match = re.search(r'total:\s*(\d+)', text)
-        if score_match:
-            self_eval_scores.append(int(score_match.group(1)))
+        score = extract_self_eval_score(text)
+        if score is not None:
+            self_eval_scores.append(score)
 
         # Exec results
-        passes = len(re.findall(r'result:\s*"?pass"?', text))
-        fails = len(re.findall(r'result:\s*"?fail"?', text))
+        criteria = parse_acceptance_criteria(text)
+        passes = sum(1 for ac in criteria if ac.get("result") == "pass")
+        fails = sum(1 for ac in criteria if ac.get("result") == "fail")
         if passes or fails:
             exec_pass += passes
             exec_fail += fails