feat: add sampler.py — session value scorer (#17)

scripts/sampler.py

@@ -0,0 +1,353 @@
+#!/usr/bin/env python3
+"""
+sampler.py — Score and rank sessions by harvest value.
+
+With 20k+ sessions on disk, we can't harvest all at once. This script
+scores each session by how likely it is to contain valuable knowledge,
+so the harvester processes the best ones first.
+
+Scoring strategy:
+  - Recency: last 7d=3pts, last 30d=2pts, older=1pt
+  - Length: >50 messages=3pts, >20=2pts, <20=1pt
+  - Repo uniqueness: first session for a repo=5pts, otherwise=1pt
+  - Outcome: failure=3pts (most to learn), success=2pts, unknown=1pt
+  - Tool calls: >10 tool invocations=2pts (complex sessions)
+
+Usage:
+    python3 sampler.py --count 100                          # Top 100 sessions
+    python3 sampler.py --repo the-nexus --count 20          # Top 20 for a repo
+    python3 sampler.py --since 2026-04-01                   # All sessions since date
+    python3 sampler.py --count 50 --min-score 8             # Only high-value sessions
+    python3 sampler.py --count 100 --output sample.json     # Save to file
+"""
+
+import argparse
+import json
+import os
+import sys
+import time
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+from typing import Optional
+
+
+# --- Fast session scanning (no full parse) ---
+
+def scan_session_fast(path: str) -> dict:
+    """Extract scoring metadata from a session without parsing the full JSONL.
+    
+    Reads only: first line, last ~20 lines, and line count. This processes
+    20k sessions in seconds instead of minutes.
+    """
+    meta = {
+        'path': path,
+        'message_count': 0,
+        'has_tool_calls': False,
+        'tool_call_count': 0,
+        'first_timestamp': '',
+        'last_timestamp': '',
+        'is_failure': False,
+        'repos_mentioned': set(),
+        'first_role': '',
+        'last_content_preview': '',
+    }
+    
+    try:
+        file_size = os.path.getsize(path)
+        if file_size == 0:
+            return meta
+        
+        with open(path, 'r', encoding='utf-8', errors='replace') as f:
+            # Read first line for timestamp + role
+            first_line = f.readline().strip()
+            if first_line:
+                try:
+                    first_msg = json.loads(first_line)
+                    meta['first_timestamp'] = first_msg.get('timestamp', '')
+                    meta['first_role'] = first_msg.get('role', '')
+                except json.JSONDecodeError:
+                    pass
+            
+            # Fast line count + collect tail lines
+            # For the tail, seek to near end of file
+            tail_lines = []
+            line_count = 1  # already read first
+            
+            if file_size > 8192:
+                # Seek to last 8KB for tail sampling
+                f.seek(max(0, file_size - 8192))
+                f.readline()  # skip partial line
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        tail_lines.append(line)
+                    line_count += 1
+                # We lost the exact count for big files — estimate from file size
+                # Average JSONL line is ~500 bytes
+                if line_count < 100:
+                    line_count = max(line_count, file_size // 500)
+            else:
+                # Small file — read all
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        tail_lines.append(line)
+                    line_count += 1
+            
+            meta['message_count'] = line_count
+            
+            # Parse tail lines for outcome, tool calls, repos
+            for line in tail_lines[-30:]:  # last 30 non-empty lines
+                try:
+                    msg = json.loads(line)
+                    
+                    # Track last timestamp
+                    ts = msg.get('timestamp', '')
+                    if ts:
+                        meta['last_timestamp'] = ts
+                    
+                    # Count tool calls
+                    if msg.get('tool_calls'):
+                        meta['has_tool_calls'] = True
+                        meta['tool_call_count'] += len(msg['tool_calls'])
+                    
+                    # Detect failure signals in content
+                    content = ''
+                    if isinstance(msg.get('content'), str):
+                        content = msg['content'].lower()
+                    elif isinstance(msg.get('content'), list):
+                        for part in msg['content']:
+                            if isinstance(part, dict) and part.get('type') == 'text':
+                                content += part.get('text', '').lower()
+                    
+                    if content:
+                        meta['last_content_preview'] = content[:200]
+                        failure_signals = ['error', 'failed', 'cannot', 'unable', 
+                                          'exception', 'traceback', 'rejected', 'denied']
+                        if any(sig in content for sig in failure_signals):
+                            meta['is_failure'] = True
+                    
+                    # Extract repo references from tool call arguments
+                    if msg.get('tool_calls'):
+                        for tc in msg['tool_calls']:
+                            args = tc.get('function', {}).get('arguments', '')
+                            if isinstance(args, str):
+                                # Look for repo patterns
+                                for pattern in ['Timmy_Foundation/', 'Rockachopa/', 'compounding-intelligence', 'the-nexus', 'timmy-home', 'hermes-agent', 'the-beacon', 'the-door']:
+                                    if pattern in args:
+                                        repo = pattern.rstrip('/')
+                                        meta['repos_mentioned'].add(repo)
+                
+                except json.JSONDecodeError:
+                    continue
+    
+    except (IOError, OSError):
+        pass
+    
+    meta['repos_mentioned'] = list(meta['repos_mentioned'])
+    return meta
+
+
+# --- Filename timestamp parsing ---
+
+def parse_session_timestamp(filename: str) -> Optional[datetime]:
+    """Parse timestamp from session filename.
+    
+    Common formats:
+      session_20260413_123456_hash.jsonl
+      20260413_123456_hash.jsonl
+    """
+    stem = Path(filename).stem
+    parts = stem.split('_')
+    
+    # Try session_YYYYMMDD_HHMMSS format
+    for i, part in enumerate(parts):
+        if len(part) == 8 and part.isdigit():
+            date_part = part
+            time_part = parts[i + 1] if i + 1 < len(parts) and len(parts[i + 1]) == 6 else '000000'
+            try:
+                return datetime.strptime(f"{date_part}_{time_part}", '%Y%m%d_%H%M%S').replace(tzinfo=timezone.utc)
+            except ValueError:
+                continue
+    
+    # Fallback: use file modification time
+    return None
+
+
+# --- Scoring ---
+
+def score_session(meta: dict, now: datetime, seen_repos: set) -> tuple[int, dict]:
+    """Score a session for harvest value. Returns (score, breakdown)."""
+    score = 0
+    breakdown = {}
+    
+    # 1. Recency
+    ts = parse_session_timestamp(os.path.basename(meta['path']))
+    if ts is None:
+        # Fallback to mtime
+        try:
+            ts = datetime.fromtimestamp(os.path.getmtime(meta['path']), tz=timezone.utc)
+        except OSError:
+            ts = now - timedelta(days=365)
+    
+    age_days = (now - ts).days
+    if age_days <= 7:
+        recency = 3
+    elif age_days <= 30:
+        recency = 2
+    else:
+        recency = 1
+    score += recency
+    breakdown['recency'] = recency
+    
+    # 2. Length
+    count = meta['message_count']
+    if count > 50:
+        length = 3
+    elif count > 20:
+        length = 2
+    else:
+        length = 1
+    score += length
+    breakdown['length'] = length
+    
+    # 3. Repo uniqueness (first session mentioning a repo gets bonus)
+    repo_score = 0
+    for repo in meta.get('repos_mentioned', []):
+        if repo not in seen_repos:
+            seen_repos.add(repo)
+            repo_score = max(repo_score, 5)
+        else:
+            repo_score = max(repo_score, 1)
+    score += repo_score
+    breakdown['repo_unique'] = repo_score
+    
+    # 4. Outcome
+    if meta.get('is_failure'):
+        outcome = 3
+    elif meta.get('last_content_preview', '').strip():
+        outcome = 2  # has some content = likely completed
+    else:
+        outcome = 1
+    score += outcome
+    breakdown['outcome'] = outcome
+    
+    # 5. Tool calls
+    if meta.get('tool_call_count', 0) > 10:
+        tool = 2
+    else:
+        tool = 0
+    score += tool
+    breakdown['tool_calls'] = tool
+    
+    return score, breakdown
+
+
+# --- Main ---
+
+def main():
+    parser = argparse.ArgumentParser(description="Score and rank sessions for harvesting")
+    parser.add_argument('--sessions-dir', default=os.path.expanduser('~/.hermes/sessions'),
+                        help='Directory containing session files')
+    parser.add_argument('--count', type=int, default=100, help='Number of top sessions to return')
+    parser.add_argument('--repo', default='', help='Filter to sessions mentioning this repo')
+    parser.add_argument('--since', default='', help='Only score sessions after this date (YYYY-MM-DD)')
+    parser.add_argument('--min-score', type=int, default=0, help='Minimum score threshold')
+    parser.add_argument('--output', default='', help='Output file (JSON). Default: stdout')
+    parser.add_argument('--format', choices=['json', 'paths', 'table'], default='table',
+                        help='Output format: json (full), paths (one per line), table (human)')
+    parser.add_argument('--top-percent', type=float, default=0, help='Return top N%% instead of --count')
+    
+    args = parser.parse_args()
+    
+    sessions_dir = Path(args.sessions_dir)
+    if not sessions_dir.is_dir():
+        print(f"ERROR: Sessions directory not found: {sessions_dir}", file=sys.stderr)
+        sys.exit(1)
+    
+    # Find all JSONL files
+    print(f"Scanning {sessions_dir}...", file=sys.stderr)
+    t0 = time.time()
+    
+    session_files = list(sessions_dir.glob('*.jsonl'))
+    total = len(session_files)
+    print(f"Found {total} session files", file=sys.stderr)
+    
+    # Parse since date
+    since_dt = None
+    if args.since:
+        since_dt = datetime.strptime(args.since, '%Y-%m-%d').replace(tzinfo=timezone.utc)
+    
+    # Score all sessions
+    now = datetime.now(timezone.utc)
+    seen_repos = set()  # Track repos for uniqueness scoring
+    scored = []
+    
+    for i, sf in enumerate(session_files):
+        # Date filter (fast path: check filename first)
+        if since_dt:
+            ts = parse_session_timestamp(sf.name)
+            if ts and ts < since_dt:
+                continue
+        
+        meta = scan_session_fast(str(sf))
+        
+        # Repo filter
+        if args.repo:
+            repos = meta.get('repos_mentioned', [])
+            if args.repo.lower() not in [r.lower() for r in repos]:
+                # Also check filename
+                if args.repo.lower() not in sf.name.lower():
+                    continue
+        
+        score, breakdown = score_session(meta, now, seen_repos)
+        
+        if score >= args.min_score:
+            scored.append({
+                'path': str(sf),
+                'filename': sf.name,
+                'score': score,
+                'breakdown': breakdown,
+                'message_count': meta['message_count'],
+                'repos': meta['repos_mentioned'],
+                'is_failure': meta['is_failure'],
+            })
+        
+        if (i + 1) % 5000 == 0:
+            elapsed = time.time() - t0
+            print(f"  Scanned {i + 1}/{total} ({elapsed:.1f}s)", file=sys.stderr)
+    
+    elapsed = time.time() - t0
+    print(f"Scored {len(scored)} sessions in {elapsed:.1f}s", file=sys.stderr)
+    
+    # Sort by score descending
+    scored.sort(key=lambda x: x['score'], reverse=True)
+    
+    # Apply count or percent
+    if args.top_percent > 0:
+        count = max(1, int(len(scored) * args.top_percent / 100))
+    else:
+        count = args.count
+    scored = scored[:count]
+    
+    # Output
+    if args.output:
+        with open(args.output, 'w', encoding='utf-8') as f:
+            json.dump(scored, f, indent=2)
+        print(f"Wrote {len(scored)} sessions to {args.output}", file=sys.stderr)
+    elif args.format == 'json':
+        json.dump(scored, sys.stdout, indent=2)
+    elif args.format == 'paths':
+        for s in scored:
+            print(s['path'])
+    else:  # table
+        print(f"{'SCORE':>5}  {'MSGS':>5}  {'REPOS':<25}  {'FILE'}")
+        print(f"{'-'*5}  {'-'*5}  {'-'*25}  {'-'*40}")
+        for s in scored:
+            repos = ', '.join(s['repos'][:2]) if s['repos'] else '-'
+            fail = ' FAIL' if s['is_failure'] else ''
+            print(f"{s['score']:>5}  {s['message_count']:>5}  {repos:<25}  {s['filename'][:40]}{fail}")
+
+
+if __name__ == '__main__':
+    main()

scripts/session_pair_harvester.py

@@ -1,234 +0,0 @@
-#!/usr/bin/env python3
-"""
-Session Transcript → Training Pair Harvester
-
-Scans Hermes session JSONL files for Q&A patterns and extracts
-terse→rich training pairs. Outputs JSONL matching the timmy-config
-training pairs spec.
-
-Usage:
-  python3 scripts/session_pair_harvester.py ~/.hermes/sessions/
-  python3 scripts/session_pair_harvester.py session.jsonl --output pairs.jsonl
-  python3 scripts/session_pair_harvester.py --dir ~/.hermes/sessions/ --min-ratio 2.0
-
-Output format:
-  {"terse": "user short prompt", "rich": "ai detailed response", "source": "session_id", "model": "..."}
-"""
-
-import argparse
-import hashlib
-import json
-import sys
-from pathlib import Path
-from typing import Optional
-
-
-def compute_hash(text: str) -> str:
-    """Content hash for deduplication."""
-    return hashlib.sha256(text.encode()).hexdigest()[:16]
-
-
-def extract_pairs_from_session(session_data: dict, min_ratio: float = 1.5,
-                                min_response_words: int = 20) -> list:
-    """Extract terse→rich pairs from a single session object."""
-    pairs = []
-    conversations = session_data.get("conversations", [])
-    session_id = session_data.get("id", "unknown")
-    model = session_data.get("model", "unknown")
-
-    seen_hashes = set()
-
-    for i, msg in enumerate(conversations):
-        # Look for assistant/gpt responses
-        if msg.get("from") not in ("gpt", "assistant"):
-            continue
-
-        response_text = msg.get("value", "")
-        if not response_text or len(response_text.split()) < min_response_words:
-            continue
-
-        # Find the preceding human message
-        prompt_text = ""
-        for j in range(i - 1, -1, -1):
-            if conversations[j].get("from") == "human":
-                prompt_text = conversations[j].get("value", "")
-                break
-
-        if not prompt_text:
-            continue
-
-        # Filter: skip tool results, system messages embedded as human
-        if prompt_text.startswith("{") and "output" in prompt_text[:100]:
-            continue  # likely a tool result
-        if prompt_text.startswith("# SOUL.md") or prompt_text.startswith("You are"):
-            continue  # system prompt leak
-
-        # Quality filters
-        prompt_words = len(prompt_text.split())
-        response_words = len(response_text.split())
-
-        # Must have meaningful length ratio
-        if prompt_words == 0 or response_words == 0:
-            continue
-        ratio = response_words / prompt_words
-        if ratio < min_ratio:
-            continue
-
-        # Skip responses that are mostly code
-        code_blocks = response_text.count("```")
-        if code_blocks >= 4 and len(response_text.replace("```", "").strip()) < 50:
-            continue
-
-        # Skip responses with tool call artifacts
-        if "tool_call" in response_text[:100] or "function_call" in response_text[:100]:
-            continue
-
-        # Deduplicate by content hash
-        content_hash = compute_hash(prompt_text + response_text[:200])
-        if content_hash in seen_hashes:
-            continue
-        seen_hashes.add(content_hash)
-
-        # Clean up response: remove markdown headers if too many
-        clean_response = response_text
-
-        pairs.append({
-            "terse": prompt_text.strip(),
-            "rich": clean_response.strip(),
-            "source": session_id,
-            "model": model,
-            "prompt_words": prompt_words,
-            "response_words": response_words,
-            "ratio": round(ratio, 2),
-        })
-
-    return pairs
-
-
-def extract_from_jsonl_file(filepath: str, **kwargs) -> list:
-    """Extract pairs from a session JSONL file."""
-    pairs = []
-    path = Path(filepath)
-
-    if not path.exists():
-        print(f"Warning: {filepath} not found", file=sys.stderr)
-        return pairs
-
-    content = path.read_text()
-    lines = content.strip().split("\n")
-
-    for line in lines:
-        line = line.strip()
-        if not line:
-            continue
-        try:
-            session = json.loads(line)
-        except json.JSONDecodeError:
-            continue
-
-        session_pairs = extract_pairs_from_session(session, **kwargs)
-        pairs.extend(session_pairs)
-
-    return pairs
-
-
-def deduplicate_pairs(pairs: list) -> list:
-    """Remove duplicate pairs across files."""
-    seen = set()
-    unique = []
-    for pair in pairs:
-        key = compute_hash(pair["terse"] + pair["rich"][:200])
-        if key not in seen:
-            seen.add(key)
-            unique.append(pair)
-    return unique
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Harvest training pairs from session transcripts")
-    parser.add_argument("input", nargs="?", help="Session JSONL file or directory")
-    parser.add_argument("--dir", "-d", help="Directory to scan for session files")
-    parser.add_argument("--output", "-o", default="harvested_pairs.jsonl", help="Output file")
-    parser.add_argument("--min-ratio", type=float, default=1.5, help="Min response/prompt word ratio")
-    parser.add_argument("--min-words", type=int, default=20, help="Min response word count")
-    parser.add_argument("--dry-run", action="store_true", help="Print stats without writing")
-    args = parser.parse_args()
-
-    all_pairs = []
-    files_scanned = 0
-
-    scan_dir = args.dir or args.input
-    if not scan_dir:
-        parser.print_help()
-        sys.exit(1)
-
-    scan_path = Path(scan_dir)
-    if scan_path.is_dir():
-        jsonl_files = sorted(scan_path.rglob("*.jsonl"))
-        print(f"Scanning {len(jsonl_files)} files in {scan_dir}...", file=sys.stderr)
-        for fpath in jsonl_files:
-            pairs = extract_from_jsonl_file(
-                str(fpath),
-                min_ratio=args.min_ratio,
-                min_response_words=args.min_words
-            )
-            all_pairs.extend(pairs)
-            files_scanned += 1
-    else:
-        pairs = extract_from_jsonl_file(
-            str(scan_path),
-            min_ratio=args.min_ratio,
-            min_response_words=args.min_words
-        )
-        all_pairs.extend(pairs)
-        files_scanned = 1
-
-    # Deduplicate
-    unique_pairs = deduplicate_pairs(all_pairs)
-
-    # Stats
-    if unique_pairs:
-        avg_prompt = sum(p["prompt_words"] for p in unique_pairs) / len(unique_pairs)
-        avg_response = sum(p["response_words"] for p in unique_pairs) / len(unique_pairs)
-        avg_ratio = sum(p["ratio"] for p in unique_pairs) / len(unique_pairs)
-    else:
-        avg_prompt = avg_response = avg_ratio = 0
-
-    stats = {
-        "files_scanned": files_scanned,
-        "raw_pairs": len(all_pairs),
-        "unique_pairs": len(unique_pairs),
-        "duplicates_removed": len(all_pairs) - len(unique_pairs),
-        "avg_prompt_words": round(avg_prompt, 1),
-        "avg_response_words": round(avg_response, 1),
-        "avg_ratio": round(avg_ratio, 2),
-    }
-
-    print(json.dumps(stats, indent=2), file=sys.stderr)
-
-    if args.dry_run:
-        # Print sample pairs
-        for pair in unique_pairs[:3]:
-            print(f"\n--- Source: {pair['source']} (ratio: {pair['ratio']}) ---", file=sys.stderr)
-            print(f"TERSE: {pair['terse'][:100]}...", file=sys.stderr)
-            print(f"RICH: {pair['rich'][:150]}...", file=sys.stderr)
-        return
-
-    # Write output
-    output_path = Path(args.output)
-    with open(output_path, "w") as f:
-        for pair in unique_pairs:
-            # Strip internal fields for output
-            output = {
-                "terse": pair["terse"],
-                "rich": pair["rich"],
-                "source": pair["source"],
-                "model": pair["model"],
-            }
-            f.write(json.dumps(output) + "\n")
-
-    print(f"\nWrote {len(unique_pairs)} pairs to {output_path}", file=sys.stderr)
-
-
-if __name__ == "__main__":
-    main()

scripts/test_session_pair_harvester.py

@@ -1,90 +0,0 @@
-#!/usr/bin/env python3
-"""Tests for session_pair_harvester."""
-
-import json
-import sys
-import os
-import tempfile
-
-sys.path.insert(0, os.path.dirname(__file__))
-from session_pair_harvester import extract_pairs_from_session, deduplicate_pairs, compute_hash
-
-
-def test_basic_extraction():
-    session = {
-        "id": "test_001",
-        "model": "test-model",
-        "conversations": [
-            {"from": "system", "value": "You are helpful."},
-            {"from": "human", "value": "What is Python?"},
-            {"from": "gpt", "value": "Python is a high-level programming language known for its readability and versatility. It supports multiple paradigms including procedural, object-oriented, and functional programming. Python is widely used in web development, data science, machine learning, and automation."},
-        ]
-    }
-    pairs = extract_pairs_from_session(session, min_ratio=1.5, min_response_words=10)
-    assert len(pairs) == 1
-    assert pairs[0]["terse"] == "What is Python?"
-    assert "programming language" in pairs[0]["rich"]
-    assert pairs[0]["source"] == "test_001"
-    print("PASS: test_basic_extraction")
-
-
-def test_filters_short_responses():
-    session = {
-        "id": "test_002",
-        "model": "test",
-        "conversations": [
-            {"from": "human", "value": "Hi"},
-            {"from": "gpt", "value": "Hello!"},
-        ]
-    }
-    pairs = extract_pairs_from_session(session, min_ratio=1.5, min_response_words=20)
-    assert len(pairs) == 0
-    print("PASS: test_filters_short_responses")
-
-
-def test_skips_tool_results():
-    session = {
-        "id": "test_003",
-        "model": "test",
-        "conversations": [
-            {"from": "human", "value": '{"output": "file content", "exit_code": 0}'},
-            {"from": "gpt", "value": "The file was read successfully. Now let me analyze the content and provide a detailed summary of what was found in the file system."},
-        ]
-    }
-    pairs = extract_pairs_from_session(session, min_ratio=1.5, min_response_words=10)
-    assert len(pairs) == 0
-    print("PASS: test_skips_tool_results")
-
-
-def test_deduplication():
-    pairs = [
-        {"terse": "What is X?", "rich": "X is Y.", "source": "s1", "model": "m"},
-        {"terse": "What is X?", "rich": "X is Y.", "source": "s2", "model": "m"},
-        {"terse": "What is Z?", "rich": "Z is W.", "source": "s1", "model": "m"},
-    ]
-    unique = deduplicate_pairs(pairs)
-    assert len(unique) == 2
-    print("PASS: test_deduplication")
-
-
-def test_ratio_filter():
-    session = {
-        "id": "test_005",
-        "model": "test",
-        "conversations": [
-            {"from": "human", "value": "Explain quantum computing in detail with examples and applications"},
-            {"from": "gpt", "value": "OK."},
-        ]
-    }
-    pairs = extract_pairs_from_session(session, min_ratio=1.5, min_response_words=10)
-    assert len(pairs) == 0  # response too short relative to prompt
-    print("PASS: test_ratio_filter")
-
-
-if __name__ == "__main__":
-    test_basic_extraction()
-    test_filters_short_responses()
-    test_skips_tool_results()
-    test_deduplication()
-    test_ratio_filter()
-    print("\nAll tests passed.")