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

scripts/gitea_issue_parser.py

@@ -1,131 +0,0 @@
-#!/usr/bin/env python3
-"""
-Gitea Issue Body Parser — Extract structured data from markdown issue bodies.
-
-Usage:
-    cat issue_body.txt | python3 scripts/gitea_issue_parser.py --stdin --pretty
-    python3 scripts/gitea_issue_parser.py --url https://forge.../api/v1/repos/.../issues/123 --pretty
-    python3 scripts/gitea_issue_parser.py body.txt --title "Fix thing (#42)" --labels pipeline extraction
-"""
-
-import argparse
-import json
-import re
-import sys
-from typing import Dict, List, Any, Optional
-
-
-def parse_issue_body(body: str, title: str = "", labels: List[str] = None) -> Dict[str, Any]:
-    """Parse a Gitea issue markdown body into structured JSON.
-    
-    Extracted fields:
-    - title: Issue title
-    - context: Background/description section
-    - criteria[]: Acceptance criteria (checkboxes or numbered lists)
-    - labels[]: Issue labels
-    - epic_ref: Parent/epic issue reference (from "Closes #N" or title)
-    - sections{}: All ## sections as key-value pairs
-    """
-    result = {
-        "title": title,
-        "context": "",
-        "criteria": [],
-        "labels": labels or [],
-        "epic_ref": None,
-        "sections": {},
-    }
-    
-    if not body:
-        return result
-    
-    # Extract epic reference from title or body
-    epic_patterns = [
-        r"(?:closes|fixes|addresses|refs?)\s+#(\d+)",
-        r"#(\d+)",
-    ]
-    for pattern in epic_patterns:
-        match = re.search(pattern, (title + " " + body).lower())
-        if match:
-            result["epic_ref"] = int(match.group(1))
-            break
-    
-    # Parse ## sections
-    section_pattern = r"^##\s+(.+?)$\n((?:^(?!##\s).*$\n?)*)"
-    for match in re.finditer(section_pattern, body, re.MULTILINE):
-        section_name = match.group(1).strip().lower().replace(" ", "_")
-        section_content = match.group(2).strip()
-        result["sections"][section_name] = section_content
-    
-    # Extract acceptance criteria (checkboxes)
-    checkbox_pattern = r"^\s*-\s*\[([ xX])\]\s*(.+)$"
-    for match in re.finditer(checkbox_pattern, body, re.MULTILINE):
-        checked = match.group(1).lower() == "x"
-        text = match.group(2).strip()
-        result["criteria"].append({"text": text, "checked": checked})
-    
-    # If no checkboxes, try numbered lists in "Acceptance Criteria" or "Criteria" section
-    if not result["criteria"]:
-        for section_name in ["acceptance_criteria", "criteria", "acceptance criteria"]:
-            if section_name in result["sections"]:
-                numbered = r"^\s*\d+\.\s*(.+)$"
-                for match in re.finditer(numbered, result["sections"][section_name], re.MULTILINE):
-                    result["criteria"].append({"text": match.group(1).strip(), "checked": False})
-                break
-    
-    # Extract context (first section or first paragraph before any ## heading)
-    first_heading = body.find("## ")
-    if first_heading > 0:
-        context_text = body[:first_heading].strip()
-    else:
-        context_text = body.split("\n\n")[0].strip()
-    # Clean up: remove "## Context" or "## Problem" header if present
-    context_text = re.sub(r"^#+\s*\w+\s*\n?", "", context_text).strip()
-    result["context"] = context_text[:500]  # Cap at 500 chars
-    
-    return result
-
-
-def fetch_issue_from_url(url: str) -> Dict[str, Any]:
-    """Fetch an issue from a Gitea API URL and parse it."""
-    import urllib.request
-    req = urllib.request.Request(url, headers={"Accept": "application/json"})
-    with urllib.request.urlopen(req) as resp:
-        data = json.loads(resp.read())
-    
-    return parse_issue_body(
-        body=data.get("body", ""),
-        title=data.get("title", ""),
-        labels=[l["name"] for l in data.get("labels", [])]
-    )
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Parse Gitea issue markdown into structured JSON")
-    parser.add_argument("file", nargs="?", help="Issue body file (or use --stdin)")
-    parser.add_argument("--stdin", action="store_true", help="Read from stdin")
-    parser.add_argument("--url", help="Gitea API URL to fetch issue from")
-    parser.add_argument("--title", default="", help="Issue title")
-    parser.add_argument("--labels", nargs="*", default=[], help="Issue labels")
-    parser.add_argument("--pretty", action="store_true", help="Pretty-print JSON output")
-    
-    args = parser.parse_args()
-    
-    if args.url:
-        result = fetch_issue_from_url(args.url)
-    elif args.stdin:
-        body = sys.stdin.read()
-        result = parse_issue_body(body, args.title, args.labels)
-    elif args.file:
-        with open(args.file) as f:
-            body = f.read()
-        result = parse_issue_body(body, args.title, args.labels)
-    else:
-        parser.print_help()
-        sys.exit(1)
-    
-    indent = 2 if args.pretty else None
-    print(json.dumps(result, indent=indent))
-
-
-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_gitea_issue_parser.py

@@ -1,109 +0,0 @@
-#!/usr/bin/env python3
-"""Tests for scripts/gitea_issue_parser.py"""
-
-import sys
-import os
-sys.path.insert(0, os.path.dirname(__file__) or ".")
-
-# Import from sibling
-import importlib.util
-spec = importlib.util.spec_from_file_location("parser", os.path.join(os.path.dirname(__file__) or ".", "gitea_issue_parser.py"))
-mod = importlib.util.module_from_spec(spec)
-spec.loader.exec_module(mod)
-parse_issue_body = mod.parse_issue_body
-
-
-def test_basic_parsing():
-    body = """## Context
-
-This is the background info.
-
-## Acceptance Criteria
-
-- [ ] First criterion
-- [x] Second criterion (done)
-
-## What to build
-
-Some description.
-"""
-    result = parse_issue_body(body, title="Test (#42)", labels=["bug"])
-    assert result["title"] == "Test (#42)"
-    assert result["labels"] == ["bug"]
-    assert result["epic_ref"] == 42
-    assert len(result["criteria"]) == 2
-    assert result["criteria"][0]["text"] == "First criterion"
-    assert result["criteria"][0]["checked"] == False
-    assert result["criteria"][1]["checked"] == True
-    assert "context" in result["sections"]
-    print("PASS: test_basic_parsing")
-
-
-def test_numbered_criteria():
-    body = """## Acceptance Criteria
-
-1. First item
-2. Second item
-3. Third item
-"""
-    result = parse_issue_body(body)
-    assert len(result["criteria"]) == 3
-    assert result["criteria"][0]["text"] == "First item"
-    print("PASS: test_numbered_criteria")
-
-
-def test_epic_ref_from_body():
-    body = "Closes #123\n\nSome description."
-    result = parse_issue_body(body)
-    assert result["epic_ref"] == 123
-    print("PASS: test_epic_ref_from_body")
-
-
-def test_empty_body():
-    result = parse_issue_body("")
-    assert result["criteria"] == []
-    assert result["context"] == ""
-    assert result["sections"] == {}
-    print("PASS: test_empty_body")
-
-
-def test_no_sections():
-    body = "Just a plain issue body with no headings."
-    result = parse_issue_body(body)
-    assert result["context"] == "Just a plain issue body with no headings."
-    print("PASS: test_no_sections")
-
-
-def test_multiple_sections():
-    body = """## Problem
-
-Something is broken.
-
-## Fix
-
-Do this instead.
-
-## Notes
-
-Additional info.
-"""
-    result = parse_issue_body(body)
-    assert "problem" in result["sections"]
-    assert "fix" in result["sections"]
-    assert "notes" in result["sections"]
-    assert "Something is broken" in result["sections"]["problem"]
-    print("PASS: test_multiple_sections")
-
-
-def run_all():
-    test_basic_parsing()
-    test_numbered_criteria()
-    test_epic_ref_from_body()
-    test_empty_body()
-    test_no_sections()
-    test_multiple_sections()
-    print("\nAll 6 tests passed!")
-
-
-if __name__ == "__main__":
-    run_all()