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

scripts/knowledge_staleness_check.py

@@ -1,131 +0,0 @@
-#!/usr/bin/env python3
-"""
-Knowledge Store Staleness Detector — Detect stale knowledge entries by comparing source file hashes.
-
-Usage:
-    python3 scripts/knowledge_staleness_check.py --index knowledge/index.json
-    python3 scripts/knowledge_staleness_check.py --index knowledge/index.json --json
-    python3 scripts/knowledge_staleness_check.py --index knowledge/index.json --fix
-"""
-
-import argparse
-import hashlib
-import json
-import os
-import sys
-from datetime import datetime, timezone
-from pathlib import Path
-from typing import Dict, List, Any, Optional
-
-
-def compute_file_hash(filepath: str) -> Optional[str]:
-    """Compute SHA-256 hash of a file. Returns None if file doesn't exist."""
-    try:
-        with open(filepath, "rb") as f:
-            return "sha256:" + hashlib.sha256(f.read()).hexdigest()
-    except (FileNotFoundError, IsADirectoryError, PermissionError):
-        return None
-
-
-def check_staleness(index_path: str, repo_root: str = ".") -> List[Dict[str, Any]]:
-    """Check all entries in knowledge index for staleness.
-    
-    Returns list of entries with staleness info:
-    - status: "fresh" | "stale" | "missing_source" | "no_hash"
-    - current_hash: computed hash (if source exists)
-    - stored_hash: hash from index
-    """
-    with open(index_path) as f:
-        data = json.load(f)
-
-    facts = data.get("facts", [])
-    results = []
-
-    for entry in facts:
-        source_file = entry.get("source_file")
-        stored_hash = entry.get("source_hash")
-
-        if not source_file:
-            results.append({**entry, "status": "no_source", "current_hash": None})
-            continue
-
-        full_path = os.path.join(repo_root, source_file)
-        current_hash = compute_file_hash(full_path)
-
-        if current_hash is None:
-            results.append({**entry, "status": "missing_source", "current_hash": None})
-        elif not stored_hash:
-            results.append({**entry, "status": "no_hash", "current_hash": current_hash})
-        elif current_hash != stored_hash:
-            results.append({**entry, "status": "stale", "current_hash": current_hash})
-        else:
-            results.append({**entry, "status": "fresh", "current_hash": current_hash})
-
-    return results
-
-
-def fix_hashes(index_path: str, repo_root: str = ".") -> int:
-    """Add hashes to entries missing them. Returns count of fixed entries."""
-    with open(index_path) as f:
-        data = json.load(f)
-
-    fixed = 0
-    for entry in data.get("facts", []):
-        if entry.get("source_hash"):
-            continue
-        source_file = entry.get("source_file")
-        if not source_file:
-            continue
-        full_path = os.path.join(repo_root, source_file)
-        h = compute_file_hash(full_path)
-        if h:
-            entry["source_hash"] = h
-            fixed += 1
-
-    with open(index_path, "w") as f:
-        json.dump(data, f, indent=2)
-
-    return fixed
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Check knowledge store staleness")
-    parser.add_argument("--index", required=True, help="Path to knowledge/index.json")
-    parser.add_argument("--repo", default=".", help="Repo root for source file resolution")
-    parser.add_argument("--json", action="store_true", help="Output as JSON")
-    parser.add_argument("--fix", action="store_true", help="Add hashes to entries missing them")
-
-    args = parser.parse_args()
-
-    if args.fix:
-        fixed = fix_hashes(args.index, args.repo)
-        print(f"Fixed {fixed} entries with missing hashes.")
-        return
-
-    results = check_staleness(args.index, args.repo)
-
-    if args.json:
-        print(json.dumps(results, indent=2))
-    else:
-        stale = [r for r in results if r["status"] != "fresh"]
-        fresh = [r for r in results if r["status"] == "fresh"]
-
-        print(f"Knowledge Store Staleness Check")
-        print(f"  Total entries: {len(results)}")
-        print(f"  Fresh: {len(fresh)}")
-        print(f"  Stale/Issues: {len(stale)}")
-        print()
-
-        if stale:
-            print("Issues found:")
-            for r in stale:
-                status = r["status"]
-                fact = r.get("fact", "?")[:60]
-                source = r.get("source_file", "?")
-                print(f"  [{status}] {source}: {fact}")
-        else:
-            print("All entries are fresh!")
-
-
-if __name__ == "__main__":
-    main()

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/test_knowledge_staleness.py

@@ -1,129 +0,0 @@
-#!/usr/bin/env python3
-"""Tests for scripts/knowledge_staleness_check.py — 8 tests."""
-
-import json
-import os
-import sys
-import tempfile
-
-sys.path.insert(0, os.path.dirname(__file__) or ".")
-import importlib.util
-spec = importlib.util.spec_from_file_location("ks", os.path.join(os.path.dirname(__file__) or ".", "knowledge_staleness_check.py"))
-mod = importlib.util.module_from_spec(spec)
-spec.loader.exec_module(mod)
-check_staleness = mod.check_staleness
-fix_hashes = mod.fix_hashes
-compute_file_hash = mod.compute_file_hash
-
-
-def test_fresh_entry():
-    with tempfile.TemporaryDirectory() as tmpdir:
-        src = os.path.join(tmpdir, "source.py")
-        with open(src, "w") as f:
-            f.write("print('hello')")
-        h = compute_file_hash(src)
-        idx = os.path.join(tmpdir, "index.json")
-        with open(idx, "w") as f:
-            json.dump({"facts": [{"fact": "hello", "source_file": "source.py", "source_hash": h}]}, f)
-        results = check_staleness(idx, tmpdir)
-        assert results[0]["status"] == "fresh"
-        print("PASS: test_fresh_entry")
-
-
-def test_stale_entry():
-    with tempfile.TemporaryDirectory() as tmpdir:
-        src = os.path.join(tmpdir, "source.py")
-        with open(src, "w") as f:
-            f.write("original content")
-        idx = os.path.join(tmpdir, "index.json")
-        with open(idx, "w") as f:
-            json.dump({"facts": [{"fact": "old", "source_file": "source.py", "source_hash": "sha256:wrong"}]}, f)
-        # Now change the source
-        with open(src, "w") as f:
-            f.write("modified content")
-        results = check_staleness(idx, tmpdir)
-        assert results[0]["status"] == "stale"
-        print("PASS: test_stale_entry")
-
-
-def test_missing_source():
-    with tempfile.TemporaryDirectory() as tmpdir:
-        idx = os.path.join(tmpdir, "index.json")
-        with open(idx, "w") as f:
-            json.dump({"facts": [{"fact": "gone", "source_file": "nonexistent.py", "source_hash": "sha256:abc"}]}, f)
-        results = check_staleness(idx, tmpdir)
-        assert results[0]["status"] == "missing_source"
-        print("PASS: test_missing_source")
-
-
-def test_no_hash():
-    with tempfile.TemporaryDirectory() as tmpdir:
-        src = os.path.join(tmpdir, "source.py")
-        with open(src, "w") as f:
-            f.write("content")
-        idx = os.path.join(tmpdir, "index.json")
-        with open(idx, "w") as f:
-            json.dump({"facts": [{"fact": "no hash", "source_file": "source.py"}]}, f)
-        results = check_staleness(idx, tmpdir)
-        assert results[0]["status"] == "no_hash"
-        assert results[0]["current_hash"].startswith("sha256:")
-        print("PASS: test_no_hash")
-
-
-def test_no_source_field():
-    with tempfile.TemporaryDirectory() as tmpdir:
-        idx = os.path.join(tmpdir, "index.json")
-        with open(idx, "w") as f:
-            json.dump({"facts": [{"fact": "orphan"}]}, f)
-        results = check_staleness(idx, tmpdir)
-        assert results[0]["status"] == "no_source"
-        print("PASS: test_no_source_field")
-
-
-def test_fix_hashes():
-    with tempfile.TemporaryDirectory() as tmpdir:
-        src = os.path.join(tmpdir, "source.py")
-        with open(src, "w") as f:
-            f.write("content for hashing")
-        idx = os.path.join(tmpdir, "index.json")
-        with open(idx, "w") as f:
-            json.dump({"facts": [{"fact": "needs hash", "source_file": "source.py"}]}, f)
-        fixed = fix_hashes(idx, tmpdir)
-        assert fixed == 1
-        # Verify hash was added
-        with open(idx) as f:
-            data = json.load(f)
-        assert data["facts"][0]["source_hash"].startswith("sha256:")
-        print("PASS: test_fix_hashes")
-
-
-def test_empty_index():
-    with tempfile.TemporaryDirectory() as tmpdir:
-        idx = os.path.join(tmpdir, "index.json")
-        with open(idx, "w") as f:
-            json.dump({"facts": []}, f)
-        results = check_staleness(idx, tmpdir)
-        assert results == []
-        print("PASS: test_empty_index")
-
-
-def test_compute_hash_nonexistent():
-    h = compute_file_hash("/nonexistent/path/file.py")
-    assert h is None
-    print("PASS: test_compute_hash_nonexistent")
-
-
-def run_all():
-    test_fresh_entry()
-    test_stale_entry()
-    test_missing_source()
-    test_no_hash()
-    test_no_source_field()
-    test_fix_hashes()
-    test_empty_index()
-    test_compute_hash_nonexistent()
-    print("\nAll 8 tests passed!")
-
-
-if __name__ == "__main__":
-    run_all()