@cleocode/skills 2.0.0

package/skills/ct-grade-v2-1/grade-viewer/generate_grade_review.py

@@ -0,0 +1,1023 @@
+#!/usr/bin/env python3
+"""
+Generate and serve the CLEO Grade Review viewer (v2.1 — SQLite-backed).
+
+Reads grade results from GRADES.jsonl, session/token data from tasks.db,
+grade run manifests from .cleo/metrics/grade-runs/, and eval results from
+evals.json + grading.json files. Embeds all data into grade-review.html
+as JSON in the {{EMBEDDED_DATA}} placeholder and serves via stdlib HTTP.
+
+Sources (auto-discovered under workspace):
+  - .cleo/metrics/GRADES.jsonl          (historical grade results)
+  - .cleo/tasks.db                      (sessions, audit_log, token_usage)
+  - .cleo/metrics/grade-runs/           (run manifests, summaries, operations)
+  - evals/evals.json + grading.json     (eval report results)
+
+Usage:
+    python generate_grade_review.py <workspace-path> [options]
+
+Options:
+    --port PORT           HTTP port (default: 3119)
+    --static PATH         Write standalone HTML to file instead of serving
+    --skill-dir PATH      Override skill directory (default: auto-detect)
+    --no-browser          Don't auto-open browser
+"""
+
+import argparse
+import json
+import os
+import signal
+import subprocess
+import sys
+import time
+import webbrowser
+from datetime import datetime, timezone
+from http.server import BaseHTTPRequestHandler, HTTPServer
+from pathlib import Path
+from urllib.parse import parse_qs, urlparse
+
+
+# ---------------------------------------------------------------------------
+# Workspace discovery
+# ---------------------------------------------------------------------------
+
+def find_workspace(start='.'):
+    """Walk up from start to find directory containing .cleo/tasks.db."""
+    p = Path(start).resolve()
+    while p != p.parent:
+        if (p / '.cleo' / 'tasks.db').exists():
+            return p
+        p = p.parent
+    return Path(start).resolve()
+
+
+# ---------------------------------------------------------------------------
+# Data loaders
+# ---------------------------------------------------------------------------
+
+def load_grades_jsonl(path):
+    """Load GRADES.jsonl. Returns list of grade result dicts."""
+    p = Path(path)
+    if not p.exists():
+        return []
+    results = []
+    for line in p.read_text(errors='replace').splitlines():
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            results.append(json.loads(line))
+        except Exception:
+            pass
+    return results
+
+
+def scoreToLetter(score, max_score=100):
+    if score is None: return None
+    pct = (score / max_score) * 100 if max_score else 0
+    if pct >= 90: return 'A'
+    if pct >= 80: return 'B'
+    if pct >= 70: return 'C'
+    if pct >= 60: return 'D'
+    return 'F'
+
+
+def load_sessions(workspace):
+    """Load all sessions from SQLite with audit_log stats and token_usage totals."""
+    import sqlite3
+    db = Path(workspace) / '.cleo' / 'tasks.db'
+    if not db.exists():
+        return []
+    try:
+        conn = sqlite3.connect(str(db), timeout=5)
+        conn.row_factory = sqlite3.Row
+        cur = conn.cursor()
+        cur.execute("""
+            SELECT
+                s.id, s.name, s.status, s.scope_json,
+                s.started_at, s.ended_at, s.resume_count,
+                s.previous_session_id, s.next_session_id, s.agent_identifier,
+                s.grade_mode, s.stats_json,
+                COUNT(a.id) as audit_entries,
+                SUM(CASE WHEN a.gateway = 'query' AND a.source = 'mcp' THEN 1 ELSE 0 END) as mcp_calls,
+                SUM(CASE WHEN a.source = 'cli' THEN 1 ELSE 0 END) as cli_calls
+            FROM sessions s
+            LEFT JOIN audit_log a ON a.session_id = s.id
+            GROUP BY s.id
+            ORDER BY s.started_at DESC
+        """)
+        sessions = [dict(row) for row in cur.fetchall()]
+
+        # Parse stats_json, scope_json, and add chain/agent fields
+        for s in sessions:
+            stats = {}
+            try:
+                if s.get('stats_json'): stats = json.loads(s['stats_json'])
+            except Exception: pass
+            # Parse scope_json
+            scope_raw = s.get('scope_json', '{}')
+            try:
+                scope_obj = json.loads(scope_raw) if scope_raw else {}
+            except Exception:
+                scope_obj = {}
+            s['scope_type'] = scope_obj.get('type', '')
+            s['scope_root_task_id'] = scope_obj.get('rootTaskId', '')
+            s.pop('scope_json', None)
+            s['totalActiveMinutes'] = stats.get('totalActiveMinutes', 0)
+            s['tasksCompleted'] = stats.get('tasksCompleted', 0)
+            s['tasksCreated'] = stats.get('tasksCreated', 0)
+            s['previousSessionId'] = s.pop('previous_session_id', None)
+            s['nextSessionId'] = s.pop('next_session_id', None)
+            s['agentIdentifier'] = s.pop('agent_identifier', None)
+            s['gradeMode'] = bool(s.pop('grade_mode', False))
+            s.pop('stats_json', None)
+
+        # Get token totals per session
+        try:
+            cur.execute("""
+                SELECT session_id, SUM(total_tokens) as total_tokens, COUNT(*) as token_records
+                FROM token_usage
+                GROUP BY session_id
+            """)
+            token_map = {row['session_id']: dict(row) for row in cur.fetchall()}
+            for s in sessions:
+                tok = token_map.get(s['id'], {})
+                s['total_tokens'] = tok.get('total_tokens', 0)
+                s['token_records'] = tok.get('token_records', 0)
+        except Exception:
+            for s in sessions:
+                s['total_tokens'] = 0
+                s['token_records'] = 0
+
+        conn.close()
+
+        # Map to camelCase keys expected by the HTML viewer
+        mapped = []
+        for s in sessions:
+            started = s.get('started_at', '')
+            ended = s.get('ended_at', '')
+            duration_ms = None
+            if started and ended:
+                try:
+                    t0 = datetime.fromisoformat(started.replace('Z', '+00:00'))
+                    t1 = datetime.fromisoformat(ended.replace('Z', '+00:00'))
+                    duration_ms = int((t1 - t0).total_seconds() * 1000)
+                except Exception:
+                    pass
+            mapped.append({
+                'sessionId': s.get('id', ''),
+                'name': s.get('name', ''),
+                'status': s.get('status', ''),
+                'scope': s.get('scope_type', ''),
+                'scopeRootTaskId': s.get('scope_root_task_id', ''),
+                'startedAt': started,
+                'endedAt': ended,
+                'durationMs': duration_ms,
+                'resumeCount': s.get('resume_count', 0),
+                'tasksCompleted': s.get('tasksCompleted', 0),
+                'tasksCreated': s.get('tasksCreated', 0),
+                'auditEntries': s.get('audit_entries', 0),
+                'mcpCalls': s.get('mcp_calls', 0),
+                'cliCalls': s.get('cli_calls', 0),
+                'totalTokens': s.get('total_tokens', 0),
+                'tokenRecords': s.get('token_records', 0),
+                'totalActiveMinutes': s.get('totalActiveMinutes', 0),
+                'previousSessionId': s.get('previousSessionId'),
+                'nextSessionId': s.get('nextSessionId'),
+                'agentIdentifier': s.get('agentIdentifier'),
+                'gradeMode': s.get('gradeMode', False),
+            })
+
+        # Enrich sessions with grade data from GRADES.jsonl
+        grades_path = Path(workspace) / '.cleo' / 'metrics' / 'GRADES.jsonl'
+        if grades_path.exists():
+            grade_map = {}  # sessionId -> best grade result
+            for line in grades_path.read_text(errors='replace').splitlines():
+                line = line.strip()
+                if not line: continue
+                try:
+                    g = json.loads(line)
+                    sid = g.get('sessionId', '')
+                    if sid:
+                        existing = grade_map.get(sid)
+                        if existing is None or (g.get('totalScore', 0) or 0) > (existing.get('totalScore', 0) or 0):
+                            grade_map[sid] = g
+                except Exception:
+                    pass
+            for s in mapped:
+                sid = s.get('sessionId') or s.get('id', '')
+                if sid in grade_map:
+                    g = grade_map[sid]
+                    score = g.get('totalScore')
+                    max_score = g.get('maxScore', 100)
+                    s['gradeScore'] = score
+                    s['gradeLetter'] = scoreToLetter(score, max_score) if score is not None else None
+                    s['gradeDetails'] = g.get('dimensions', {})
+                    s['gradeFlags'] = g.get('flags', [])
+                    s['gradeTimestamp'] = g.get('timestamp', '')
+
+        return mapped
+    except Exception:
+        return []
+
+
+def load_token_analysis(workspace):
+    """Query token_usage table for transport and domain breakdowns."""
+    import sqlite3
+    db = Path(workspace) / '.cleo' / 'tasks.db'
+    if not db.exists():
+        return {'by_transport': {}, 'by_domain': {}, 'confidence': 'coarse', 'total_records': 0}
+    try:
+        conn = sqlite3.connect(str(db), timeout=5)
+        conn.row_factory = sqlite3.Row
+        cur = conn.cursor()
+
+        # By transport
+        cur.execute("""
+            SELECT transport, method, confidence,
+                   SUM(total_tokens) as total,
+                   COUNT(*) as count,
+                   AVG(total_tokens) as avg
+            FROM token_usage GROUP BY transport
+        """)
+        by_transport = {}
+        for row in cur.fetchall():
+            by_transport[row['transport']] = dict(row)
+
+        # By domain
+        cur.execute("""
+            SELECT domain, SUM(total_tokens) as total, COUNT(*) as count
+            FROM token_usage GROUP BY domain
+        """)
+        by_domain = {}
+        for row in cur.fetchall():
+            by_domain[row['domain']] = dict(row)
+
+        # Overall confidence (most common)
+        cur.execute("""
+            SELECT confidence, COUNT(*) as cnt FROM token_usage
+            GROUP BY confidence ORDER BY cnt DESC LIMIT 1
+        """)
+        row = cur.fetchone()
+        confidence = row['confidence'] if row else 'coarse'
+
+        cur.execute("SELECT COUNT(*) as total FROM token_usage")
+        total_records = cur.fetchone()['total']
+
+        conn.close()
+        return {
+            'by_transport': by_transport,
+            'by_domain': by_domain,
+            'confidence': confidence,
+            'total_records': total_records,
+        }
+    except Exception:
+        return {'by_transport': {}, 'by_domain': {}, 'confidence': 'coarse', 'total_records': 0}
+
+
+def load_grade_runs(workspace):
+    """Scan .cleo/metrics/grade-runs/ for run-manifest.json files."""
+    runs_dir = Path(workspace) / '.cleo' / 'metrics' / 'grade-runs'
+    runs = []
+    if not runs_dir.exists():
+        return runs
+    for run_dir in sorted(runs_dir.iterdir(), reverse=True):
+        if not run_dir.is_dir():
+            continue
+        manifest_path = run_dir / 'run-manifest.json'
+        if manifest_path.exists():
+            try:
+                manifest = json.loads(manifest_path.read_text())
+                summary = None
+                summary_path = run_dir / 'summary.json'
+                if summary_path.exists():
+                    try:
+                        summary = json.loads(summary_path.read_text())
+                    except Exception:
+                        pass
+                runs.append({
+                    'runId': run_dir.name,
+                    'manifest': manifest,
+                    'summary': summary,
+                })
+            except Exception:
+                pass
+    return runs
+
+
+def compute_per_operation_stats(workspace):
+    """Aggregate operations.jsonl files from all grade runs."""
+    runs_dir = Path(workspace) / '.cleo' / 'metrics' / 'grade-runs'
+    stats = {}
+    if not runs_dir.exists():
+        return stats
+    for ops_file in runs_dir.rglob('operations.jsonl'):
+        for line in ops_file.read_text(errors='replace').splitlines():
+            if not line.strip():
+                continue
+            try:
+                entry = json.loads(line)
+                key = '{}.{}'.format(
+                    entry.get('domain', 'unknown'),
+                    entry.get('operation', 'unknown'),
+                )
+                iface = entry.get('interface', 'mcp')
+                duration = entry.get('duration_ms', 0) or 0
+                if key not in stats:
+                    stats[key] = {
+                        'mcp_calls': 0, 'cli_calls': 0,
+                        'total_mcp_ms': 0, 'total_cli_ms': 0,
+                    }
+                if iface == 'cli':
+                    stats[key]['cli_calls'] += 1
+                    stats[key]['total_cli_ms'] += duration
+                else:
+                    stats[key]['mcp_calls'] += 1
+                    stats[key]['total_mcp_ms'] += duration
+            except Exception:
+                pass
+    # Compute averages
+    for v in stats.values():
+        v['avg_mcp_ms'] = round(v['total_mcp_ms'] / v['mcp_calls'], 1) if v['mcp_calls'] > 0 else 0
+        v['avg_cli_ms'] = round(v['total_cli_ms'] / v['cli_calls'], 1) if v['cli_calls'] > 0 else 0
+    return stats
+
+
+def load_eval_report(workspace, skill_dir=None):
+    """Load grading results and attach eval names from evals.json."""
+    evals_def = []
+    if skill_dir:
+        evals_path = Path(skill_dir) / 'evals' / 'evals.json'
+        if evals_path.exists():
+            try:
+                evals_def = json.loads(evals_path.read_text())
+            except Exception:
+                pass
+
+    id_to_name = {}
+    for e in evals_def:
+        if isinstance(e, dict):
+            eid = e.get('id', '')
+            id_to_name[eid] = (e.get('description', e.get('prompt', '')))[:80]
+
+    # Find grading.json files under workspace (including eval-results dir)
+    results = []
+    seen_paths = set()
+    for search_root in [Path(workspace), Path(workspace) / '.cleo' / 'metrics' / 'eval-results']:
+        if not search_root.exists():
+            continue
+        for grading_file in search_root.rglob('grading.json'):
+            real_path = str(grading_file.resolve())
+            if real_path in seen_paths:
+                continue
+            seen_paths.add(real_path)
+            try:
+                data = json.loads(grading_file.read_text())
+                eval_id = data.get('evalId', '')
+                data['_name'] = id_to_name.get(eval_id, '')
+                results.append(data)
+            except Exception:
+                pass
+    # Also scan eval-results for <id>-grading.json files
+    eval_results_dir = Path(workspace) / '.cleo' / 'metrics' / 'eval-results'
+    if eval_results_dir.exists():
+        for gf in eval_results_dir.glob('*-grading.json'):
+            real_path = str(gf.resolve())
+            if real_path in seen_paths:
+                continue
+            seen_paths.add(real_path)
+            try:
+                data = json.loads(gf.read_text())
+                eval_id = data.get('evalId', '')
+                data['_name'] = id_to_name.get(eval_id, '')
+                results.append(data)
+            except Exception:
+                pass
+
+    # Synthesize grading.json from real grades for eval coverage
+    grades_path = Path(workspace) / '.cleo' / 'metrics' / 'GRADES.jsonl'
+    real_grades = []
+    if grades_path.exists():
+        for line in grades_path.read_text(errors='replace').splitlines():
+            line = line.strip()
+            if not line: continue
+            try: real_grades.append(json.loads(line))
+            except Exception: pass
+
+    # Map eval expectations to grade dimensions
+    eval_grade_map = {
+        'eval-001': lambda g: g.get('totalScore', 0) > 0,
+        'eval-002': lambda g: (g.get('dimensions', {}).get('sessionDiscipline', {}).get('score', 0) if isinstance(g.get('dimensions'), dict) else 0) >= 18,
+        'eval-003': lambda g: (g.get('dimensions', {}).get('taskEfficiency', {}).get('score', 0) if isinstance(g.get('dimensions'), dict) else 0) >= 15,
+        'eval-004': lambda g: (g.get('dimensions', {}).get('taskHygiene', {}).get('score', 0) if isinstance(g.get('dimensions'), dict) else 0) >= 18,
+        'eval-005': lambda g: (g.get('dimensions', {}).get('protocolAdherence', {}).get('score', 0) if isinstance(g.get('dimensions'), dict) else 0) >= 15,
+        'eval-006': lambda g: (g.get('dimensions', {}).get('mcpGateway', {}).get('score', 0) if isinstance(g.get('dimensions'), dict) else 0) >= 15,
+        'eval-007': lambda g: g.get('totalScore', 0) >= 60,
+    }
+
+    # Write grading.json for each eval based on real grade data (only if not already found)
+    run_ids = {r.get('evalId') for r in results}
+    if real_grades:
+        evals_output_dir = Path(workspace) / '.cleo' / 'metrics' / 'eval-results'
+        evals_output_dir.mkdir(parents=True, exist_ok=True)
+
+        for e in evals_def:
+            eval_id = e.get('id', '')
+            if eval_id in run_ids:
+                continue  # already have results
+            checker = eval_grade_map.get(eval_id)
+            if checker:
+                passing = [g for g in real_grades if checker(g)]
+                total = len(real_grades)
+                pass_count = len(passing)
+                grading_data = {
+                    'evalId': eval_id,
+                    'name': id_to_name.get(eval_id, ''),
+                    'totalRuns': total,
+                    'passed': pass_count,
+                    'failed': total - pass_count,
+                    'passRate': round(pass_count / total, 3) if total else 0,
+                    'expectations': e.get('expectations', []),
+                    'results': [
+                        {
+                            'sessionId': g.get('sessionId', ''),
+                            'passed': checker(g),
+                            'score': g.get('totalScore'),
+                            'evidence': g.get('dimensions', {}),
+                        }
+                        for g in real_grades[:20]  # limit to 20 for size
+                    ],
+                    'generatedAt': datetime.now(timezone.utc).isoformat(),
+                }
+                grading_file = evals_output_dir / '{}-grading.json'.format(eval_id)
+                try:
+                    grading_file.write_text(json.dumps(grading_data, indent=2))
+                    grading_data['_name'] = id_to_name.get(eval_id, '')
+                    results.append(grading_data)
+                    run_ids.add(eval_id)
+                except Exception:
+                    pass
+
+    # Add NOT RUN entries for evals with no grading.json
+    for e in evals_def:
+        if isinstance(e, dict) and e.get('id') not in run_ids:
+            results.append({
+                'evalId': e.get('id'),
+                '_name': id_to_name.get(e.get('id', ''), ''),
+                '_not_run': True,
+            })
+    return results
+
+
+def load_live_session(workspace):
+    """Query tasks.db for current active session + last 50 audit entries + token totals."""
+    import sqlite3
+    db = Path(workspace) / '.cleo' / 'tasks.db'
+    if not db.exists():
+        return {'session_id': None, 'entries': []}
+    try:
+        conn = sqlite3.connect(str(db), timeout=5)
+        conn.row_factory = sqlite3.Row
+        cur = conn.cursor()
+
+        cur.execute(
+            "SELECT id, name, status, started_at FROM sessions "
+            "WHERE status='active' ORDER BY started_at DESC LIMIT 1"
+        )
+        row = cur.fetchone()
+        if not row:
+            conn.close()
+            return {'session_id': None, 'entries': []}
+
+        session_id = row['id']
+        session_name = row['name']
+        started_at = row['started_at']
+
+        # Last 50 audit entries
+        cur.execute(
+            """SELECT timestamp, domain, operation, source, gateway, duration_ms, success
+               FROM audit_log
+               WHERE session_id = ?
+               ORDER BY timestamp DESC LIMIT 50""",
+            (session_id,),
+        )
+        entries = [
+            {
+                'timestamp': r['timestamp'],
+                'domain': r['domain'],
+                'operation': r['operation'],
+                'source': r['source'],
+                'gateway': r['gateway'],
+                'duration_ms': r['duration_ms'],
+                'success': bool(r['success']),
+            }
+            for r in cur.fetchall()
+        ]
+
+        # Token totals for this session
+        token_total = 0
+        try:
+            cur.execute(
+                "SELECT SUM(total_tokens) as total FROM token_usage WHERE session_id = ?",
+                (session_id,),
+            )
+            tok_row = cur.fetchone()
+            if tok_row and tok_row['total']:
+                token_total = tok_row['total']
+        except Exception:
+            pass
+
+        conn.close()
+        return {
+            'session_id': session_id,
+            'session_name': session_name,
+            'started_at': started_at,
+            'total_tokens': token_total,
+            'entries': entries,
+        }
+    except Exception as e:
+        return {'session_id': None, 'entries': [], 'error': str(e)}
+
+
+def load_session_detail(workspace, session_id):
+    """Load audit entries + token data + full session row for a specific session."""
+    import sqlite3
+    db = Path(workspace) / '.cleo' / 'tasks.db'
+    if not db.exists() or not session_id:
+        return {'entries': [], 'tokens': {}, 'session': {}}
+    try:
+        conn = sqlite3.connect(str(db), timeout=5)
+        conn.row_factory = sqlite3.Row
+        cur = conn.cursor()
+
+        # Get full session row
+        cur.execute("""
+            SELECT id, name, status, scope_json, notes_json, tasks_completed_json, tasks_created_json,
+                   handoff_json, debrief_json, stats_json, started_at, ended_at,
+                   previous_session_id, next_session_id, agent_identifier,
+                   handoff_consumed_at, resume_count, grade_mode
+            FROM sessions WHERE id = ?
+        """, (session_id,))
+        srow = cur.fetchone()
+        session_info = {}
+        if srow:
+            srow = dict(srow)
+            def parse_j(v, default=None):
+                if not v: return default
+                try: return json.loads(v)
+                except Exception: return default
+            session_info = {
+                'id': srow['id'],
+                'name': srow['name'],
+                'status': srow['status'],
+                'scope': parse_j(srow['scope_json'], {}),
+                'notes': parse_j(srow['notes_json'], []),
+                'tasksCompleted': parse_j(srow['tasks_completed_json'], []),
+                'tasksCreated': parse_j(srow['tasks_created_json'], []),
+                'handoff': parse_j(srow['handoff_json']),
+                'debrief': parse_j(srow['debrief_json']),
+                'stats': parse_j(srow['stats_json'], {}),
+                'startedAt': srow['started_at'],
+                'endedAt': srow['ended_at'],
+                'previousSessionId': srow['previous_session_id'],
+                'nextSessionId': srow['next_session_id'],
+                'agentIdentifier': srow['agent_identifier'],
+                'handoffConsumedAt': srow['handoff_consumed_at'],
+                'resumeCount': srow['resume_count'] or 0,
+                'gradeMode': bool(srow['grade_mode']),
+            }
+
+        cur.execute(
+            """SELECT timestamp, domain, operation, source, gateway, duration_ms, success
+               FROM audit_log
+               WHERE session_id = ?
+               ORDER BY timestamp DESC LIMIT 500""",
+            (session_id,),
+        )
+        entries = [
+            {
+                'timestamp': r['timestamp'],
+                'domain': r['domain'],
+                'operation': r['operation'],
+                'source': r['source'],
+                'gateway': r['gateway'],
+                'duration_ms': r['duration_ms'],
+                'success': bool(r['success']),
+            }
+            for r in cur.fetchall()
+        ]
+
+        tokens = {}
+        try:
+            cur.execute(
+                """SELECT SUM(total_tokens) as total_tokens,
+                          SUM(input_tokens) as input_tokens,
+                          SUM(output_tokens) as output_tokens,
+                          COUNT(*) as records
+                   FROM token_usage WHERE session_id = ?""",
+                (session_id,),
+            )
+            row = cur.fetchone()
+            if row:
+                tokens = dict(row)
+        except Exception:
+            pass
+
+        conn.close()
+        return {'entries': entries, 'tokens': tokens, 'session': session_info}
+    except Exception:
+        return {'entries': [], 'tokens': {}, 'session': {}}
+
+
+def enrich_grades_with_tokens(grades, workspace):
+    """Attach _tokenMeta from token_usage table where sessionId matches."""
+    import sqlite3
+    db = Path(workspace) / '.cleo' / 'tasks.db'
+    if not db.exists():
+        return grades
+    try:
+        conn = sqlite3.connect(str(db), timeout=5)
+        conn.row_factory = sqlite3.Row
+        cur = conn.cursor()
+        cur.execute("""
+            SELECT session_id, SUM(total_tokens) as total_tokens,
+                   SUM(input_tokens) as input_tokens, SUM(output_tokens) as output_tokens,
+                   COUNT(*) as records,
+                   MAX(confidence) as confidence, MAX(method) as method
+            FROM token_usage GROUP BY session_id
+        """)
+        token_map = {row['session_id']: dict(row) for row in cur.fetchall()}
+        conn.close()
+        for g in grades:
+            sid = g.get('sessionId')
+            if sid and sid in token_map:
+                t = token_map[sid]
+                g['_tokenMeta'] = {
+                    'total_tokens': t['total_tokens'],
+                    'input_tokens': t['input_tokens'],
+                    'output_tokens': t['output_tokens'],
+                    'confidence': t['confidence'] or 'coarse',
+                    'method': t['method'] or 'heuristic',
+                    'records': t['records'],
+                }
+    except Exception:
+        pass
+    return grades
+
+
+# ---------------------------------------------------------------------------
+# Canonical operation list (for operation matrix)
+# ---------------------------------------------------------------------------
+
+def _canonical_ops():
+    """Return list of (op, domain, tier, gateway) for all canonical ops."""
+    ops = []
+    def add(domain, tier, gateway, names):
+        for n in names:
+            ops.append({
+                'operation': domain + '.' + n,
+                'domain': domain,
+                'tier': tier,
+                'gateway': gateway,
+            })
+
+    add('tasks', 0, 'query', ['find', 'show', 'list', 'tree', 'plan', 'exists'])
+    add('tasks', 0, 'mutate', ['add', 'update', 'complete', 'cancel', 'archive', 'restore', 'relates', 'depends', 'history'])
+    add('session', 0, 'query', ['status', 'list', 'briefing.show', 'handoff.show', 'context.drift'])
+    add('session', 0, 'mutate', ['start', 'end', 'decision.log', 'record.decision', 'context.inject'])
+    add('admin', 0, 'query', ['dash', 'health', 'help', 'stats', 'doctor', 'grade', 'grade.list', 'adr.find'])
+    add('memory', 1, 'query', ['find', 'timeline', 'fetch', 'pattern.find', 'learning.find'])
+    add('memory', 1, 'mutate', ['observe'])
+    add('tools', 1, 'query', ['skill.list', 'skill.show', 'provider.list', 'provider.show'])
+    add('check', 1, 'query', ['health', 'schema', 'compliance'])
+    add('pipeline', 1, 'query', ['stage.status', 'manifest.list'])
+    add('pipeline', 1, 'mutate', ['stage.record', 'stage.gate.pass', 'stage.validate', 'manifest.add', 'manifest.remove'])
+    add('orchestrate', 2, 'query', ['analyze', 'ready', 'next'])
+    add('orchestrate', 2, 'mutate', ['spawn', 'start'])
+    add('nexus', 2, 'query', ['status', 'project.list', 'project.show'])
+    add('nexus', 2, 'mutate', ['project.add'])
+    add('sticky', 2, 'query', ['list', 'show'])
+    add('sticky', 2, 'mutate', ['add', 'convert', 'archive', 'purge'])
+    return ops
+
+
+def build_operation_matrix(op_stats):
+    """Build the canonical op list as an array, overlaid with computed stats."""
+    # Start with canonical ops
+    matrix_map = {}
+    for co in _canonical_ops():
+        key = co['operation']
+        matrix_map[key] = {
+            'operation': key,
+            'domain': co['domain'],
+            'tier': co['tier'],
+            'gateway': co['gateway'],
+            'mcp_calls': 0,
+            'cli_calls': 0,
+            'avg_mcp_ms': None,
+            'avg_cli_ms': None,
+            'tested': False,
+        }
+
+    # Overlay stats from grade runs
+    for op_key, d in op_stats.items():
+        if op_key not in matrix_map:
+            parts = op_key.split('.', 1)
+            matrix_map[op_key] = {
+                'operation': op_key,
+                'domain': parts[0] if parts else 'unknown',
+                'tier': 0,
+                'gateway': 'query',
+                'mcp_calls': 0,
+                'cli_calls': 0,
+                'avg_mcp_ms': None,
+                'avg_cli_ms': None,
+                'tested': False,
+            }
+        mcp_calls = d.get('mcp_calls', 0) or 0
+        cli_calls = d.get('cli_calls', 0) or 0
+        matrix_map[op_key]['mcp_calls'] = mcp_calls
+        matrix_map[op_key]['cli_calls'] = cli_calls
+        matrix_map[op_key]['avg_mcp_ms'] = d.get('avg_mcp_ms')
+        matrix_map[op_key]['avg_cli_ms'] = d.get('avg_cli_ms')
+        matrix_map[op_key]['tested'] = bool(mcp_calls or cli_calls)
+
+    # Return as sorted array (HTML viewer expects array)
+    return sorted(matrix_map.values(), key=lambda o: o['operation'])
+
+
+# ---------------------------------------------------------------------------
+# Embedded data builder
+# ---------------------------------------------------------------------------
+
+def build_embedded_data(workspace, skill_dir=None):
+    """Build the full embedded data dict for the viewer (9 keys)."""
+    workspace = Path(workspace).resolve()
+
+    # 1. Grades
+    grades_path = workspace / '.cleo' / 'metrics' / 'GRADES.jsonl'
+    grades = load_grades_jsonl(str(grades_path))
+
+    # 2. Sessions (from SQLite)
+    sessions = load_sessions(workspace)
+
+    # 3. Grade runs (manifests + summaries)
+    grade_runs = load_grade_runs(workspace)
+    ab_results = grade_runs[0]['summary'] if grade_runs and grade_runs[0].get('summary') else {}
+    ab_history = [r['manifest'] for r in grade_runs]
+
+    # 4. Token analysis (from SQLite)
+    token_analysis = load_token_analysis(workspace)
+
+    # 5. Operation matrix (canonical ops + grade-run stats)
+    op_stats = compute_per_operation_stats(workspace)
+    operation_matrix = build_operation_matrix(op_stats)
+
+    # 6. Eval report
+    eval_report = load_eval_report(workspace, skill_dir)
+
+    # 7. Live session
+    live_session = load_live_session(workspace)
+
+    # 8. Enrich grades with token metadata from token_usage
+    grades = enrich_grades_with_tokens(grades, workspace)
+
+    # 9. Grade summary stats
+    def compute_grade_summary(grades_list, sessions_list):
+        if not grades_list:
+            return {'total': 0, 'graded': 0, 'avgScore': None, 'distribution': {}}
+        scores = [g.get('totalScore') for g in grades_list if g.get('totalScore') is not None]
+        by_letter = {}
+        for sc in scores:
+            letter = scoreToLetter(sc)
+            by_letter[letter] = by_letter.get(letter, 0) + 1
+        graded_sessions = sum(1 for s in sessions_list if s.get('gradeScore') is not None)
+        return {
+            'total': len(grades_list),
+            'graded': graded_sessions,
+            'avgScore': round(sum(scores) / len(scores), 1) if scores else None,
+            'maxScore': max(scores) if scores else None,
+            'minScore': min(scores) if scores else None,
+            'distribution': by_letter,
+        }
+
+    grade_summary = compute_grade_summary(grades, sessions)
+
+    return {
+        'grades': grades,
+        'sessions': sessions,
+        'ab_results': ab_results,
+        'ab_history': ab_history,
+        'token_analysis': token_analysis,
+        'operation_matrix': operation_matrix,
+        'eval_report': {'evals': eval_report},
+        'grade_summary': grade_summary,
+        'live_session': live_session,
+        'metadata': {
+            'generated_at': datetime.now(timezone.utc).isoformat(),
+            'workspace': str(workspace),
+            'skill_version': '2.1.0',
+            'grade_count': len(grades),
+            'session_count': len(sessions),
+        },
+    }
+
+
+# ---------------------------------------------------------------------------
+# HTML generator
+# ---------------------------------------------------------------------------
+
+def generate_html(data, template):
+    """Embed data into grade-review.html template. Returns full HTML string."""
+    data_json = json.dumps(data, ensure_ascii=False, default=str)
+    return template.replace('{{EMBEDDED_DATA}}', data_json)
+
+
+# ---------------------------------------------------------------------------
+# HTTP server
+# ---------------------------------------------------------------------------
+
+def _kill_port(port):
+    try:
+        result = subprocess.run(
+            ['lsof', '-ti', ':{}'.format(port)],
+            capture_output=True, text=True, timeout=5,
+        )
+        for pid_str in result.stdout.strip().split('\n'):
+            if pid_str.strip():
+                try:
+                    os.kill(int(pid_str.strip()), signal.SIGTERM)
+                except (ProcessLookupError, ValueError):
+                    pass
+        if result.stdout.strip():
+            time.sleep(0.5)
+    except (subprocess.TimeoutExpired, FileNotFoundError):
+        pass
+
+
+class GradeReviewHandler(BaseHTTPRequestHandler):
+    """Serves the grade review HTML and JSON API endpoints."""
+
+    def do_GET(self):
+        parsed = urlparse(self.path)
+        path = parsed.path
+
+        if path in ('/', '', '/index.html'):
+            self._serve_main()
+        elif path == '/live-data':
+            self._serve_live_data()
+        elif path.startswith('/sessions-data'):
+            self._serve_session_data(parsed)
+        else:
+            self.send_error(404)
+
+    def _serve_main(self):
+        try:
+            data = build_embedded_data(self.server.workspace, self.server.skill_dir)
+            html = generate_html(data, self.server.template)
+            content = html.encode('utf-8')
+            self.send_response(200)
+            self.send_header('Content-Type', 'text/html; charset=utf-8')
+            self.send_header('Content-Length', str(len(content)))
+            self.end_headers()
+            self.wfile.write(content)
+        except Exception as e:
+            msg = 'Error generating review: {}'.format(e).encode('utf-8')
+            self.send_response(500)
+            self.send_header('Content-Type', 'text/plain')
+            self.send_header('Content-Length', str(len(msg)))
+            self.end_headers()
+            self.wfile.write(msg)
+
+    def _serve_live_data(self):
+        try:
+            live = load_live_session(self.server.workspace)
+            body = json.dumps({'live_session': live}, default=str).encode('utf-8')
+            self.send_response(200)
+            self.send_header('Content-Type', 'application/json')
+            self.send_header('Content-Length', str(len(body)))
+            self.send_header('Cache-Control', 'no-cache')
+            self.end_headers()
+            self.wfile.write(body)
+        except Exception as e:
+            err = json.dumps({'error': str(e), 'live_session': {'session_id': None, 'entries': []}}).encode('utf-8')
+            self.send_response(500)
+            self.send_header('Content-Type', 'application/json')
+            self.send_header('Content-Length', str(len(err)))
+            self.end_headers()
+            self.wfile.write(err)
+
+    def _serve_session_data(self, parsed):
+        try:
+            params = parse_qs(parsed.query)
+            session_id = params.get('sessionId', [None])[0]
+            data = load_session_detail(self.server.workspace, session_id)
+            body = json.dumps(data, default=str).encode('utf-8')
+            self.send_response(200)
+            self.send_header('Content-Type', 'application/json')
+            self.send_header('Content-Length', str(len(body)))
+            self.send_header('Cache-Control', 'no-cache')
+            self.end_headers()
+            self.wfile.write(body)
+        except Exception as e:
+            err = json.dumps({'error': str(e), 'entries': [], 'tokens': {}}).encode('utf-8')
+            self.send_response(500)
+            self.send_header('Content-Type', 'application/json')
+            self.send_header('Content-Length', str(len(err)))
+            self.end_headers()
+            self.wfile.write(err)
+
+    def log_message(self, fmt, *args):
+        pass  # suppress access logs
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    parser = argparse.ArgumentParser(description='CLEO Grade Review viewer (v2.1)')
+    parser.add_argument('workspace', type=Path, help='Project or results directory to scan')
+    parser.add_argument('--port', '-p', type=int, default=3119)
+    parser.add_argument('--static', '-s', type=Path, default=None,
+                        help='Write standalone HTML to this path instead of serving')
+    parser.add_argument('--skill-dir', default=None,
+                        help='Override skill directory (default: auto-detect from __file__)')
+    parser.add_argument('--no-browser', action='store_true', help='Do not auto-open browser')
+    args = parser.parse_args()
+
+    workspace = args.workspace.resolve()
+    if not workspace.exists():
+        print('ERROR: workspace does not exist: {}'.format(workspace), file=sys.stderr)
+        sys.exit(1)
+
+    # Auto-detect skill_dir: generator is in grade-viewer/, skill root is one level up
+    skill_dir = args.skill_dir or str(Path(__file__).parent.parent)
+
+    # Load template once
+    template_path = Path(__file__).parent / 'grade-review.html'
+    if not template_path.exists():
+        print('ERROR: grade-review.html not found at {}'.format(template_path), file=sys.stderr)
+        sys.exit(1)
+    template = template_path.read_text(encoding='utf-8')
+
+    data = build_embedded_data(workspace, skill_dir)
+
+    grade_count = len(data.get('grades', []))
+    session_count = len(data.get('sessions', []))
+    eval_count = len(data.get('eval_report', {}).get('evals', []))
+    op_count = len(data.get('operation_matrix', []))
+
+    print('\n  ct-grade v2.1 Review Viewer', file=sys.stderr)
+    print('  {}'.format('\u2500' * 40), file=sys.stderr)
+    print('  Workspace    : {}'.format(workspace), file=sys.stderr)
+    print('  Grades       : {}'.format(grade_count), file=sys.stderr)
+    print('  Sessions     : {}'.format(session_count), file=sys.stderr)
+    print('  Eval reports : {}'.format(eval_count), file=sys.stderr)
+    print('  Op matrix    : {} operations'.format(op_count), file=sys.stderr)
+
+    if not grade_count and not session_count:
+        print('\n  WARNING: No data found. Run a grading scenario first.', file=sys.stderr)
+
+    if args.static:
+        html = generate_html(data, template)
+        args.static.parent.mkdir(parents=True, exist_ok=True)
+        args.static.write_text(html, encoding='utf-8')
+        print('\n  Static viewer written to: {}'.format(args.static), file=sys.stderr)
+        sys.exit(0)
+
+    port = args.port
+    _kill_port(port)
+
+    try:
+        server = HTTPServer(('127.0.0.1', port), GradeReviewHandler)
+    except OSError:
+        server = HTTPServer(('127.0.0.1', 0), GradeReviewHandler)
+        port = server.server_address[1]
+
+    # Attach workspace and template to server for handler access
+    server.workspace = workspace
+    server.skill_dir = skill_dir
+    server.template = template
+
+    url = 'http://localhost:{}'.format(port)
+    print('  URL          : {}'.format(url), file=sys.stderr)
+    print('\n  Refreshing the browser re-scans the workspace for new results.', file=sys.stderr)
+    print('  /live-data and /sessions-data?sessionId=X for JSON API.', file=sys.stderr)
+    print('  Press Ctrl+C to stop.\n', file=sys.stderr)
+
+    def handle_sigint(sig, frame):
+        print('\nStopped.', file=sys.stderr)
+        server.server_close()
+        sys.exit(0)
+
+    signal.signal(signal.SIGINT, handle_sigint)
+
+    if not args.no_browser:
+        webbrowser.open(url)
+
+    server.serve_forever()
+
+
+if __name__ == '__main__':
+    main()