Merge pull request 'fix: session_pair_harvester uses role/content format (#91)' (#240) from step35/91-feat-session-transcript-trai into main

- Harvester used old message fields (from/value) but Hermes sessions use role/content
- Import session_reader to normalize conversations properly
- Update extract function to operate on normalized role/content messages
- Change predecessor lookup from "human"/"gpt" to "user"/"assistant"
- Add comprehensive smoke tests (8 tests, all pass)
- Verify extraction from test_sessions: 11 pairs, avg ratio 8.13

scripts/session_pair_harvester.py

@@ -22,114 +22,95 @@ import sys
 from pathlib import Path
 from typing import Optional
 
+from session_reader import extract_conversation, read_session
+
 
 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,
+def extract_pairs_from_conversation(conversation: list, session_id: str, model: str,
+                                min_ratio: float = 1.5,
                                 min_response_words: int = 20) -> list:
-    """Extract terse→rich pairs from a single session object."""
+    """Extract terse→rich pairs from a normalized conversation."""
     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):
+    for i, msg in enumerate(conversation):
-        # Look for assistant/gpt responses
+        # Look for assistant responses
-        if msg.get("from") not in ("gpt", "assistant"):
+        if msg.get('role') != 'assistant':
             continue
 
-        response_text = msg.get("value", "")
+        response_text = msg.get('content', '')
         if not response_text or len(response_text.split()) < min_response_words:
             continue
 
-        # Find the preceding human message
+        # Find the preceding user message
         prompt_text = ""
         for j in range(i - 1, -1, -1):
-            if conversations[j].get("from") == "human":
+            if conversation[j].get('role') == 'user':
-                prompt_text = conversations[j].get("value", "")
+                prompt_text = conversation[j].get('content', '')
                 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]:
+        if prompt_text.startswith('{') and 'output' in prompt_text[:100]:
-            continue  # likely a tool result
+            continue
-        if prompt_text.startswith("# SOUL.md") or prompt_text.startswith("You are"):
+        if prompt_text.startswith('# SOUL.md') or prompt_text.startswith('You are'):
-            continue  # system prompt leak
+            continue
 
         # 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('```')
-        code_blocks = response_text.count("```")
+        if code_blocks >= 4 and len(response_text.replace('```', '').strip()) < 50:
-        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]:
-        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(),
+            'terse': prompt_text.strip(),
-            "rich": clean_response.strip(),
+            'rich': clean_response.strip(),
-            "source": session_id,
+            'source': session_id,
-            "model": model,
+            'model': model,
-            "prompt_words": prompt_words,
+            'prompt_words': prompt_words,
-            "response_words": response_words,
+            'response_words': response_words,
-            "ratio": round(ratio, 2),
+            '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():
+def extract_from_jsonl_file(path: str, **kwargs) -> list:
-        print(f"Warning: {filepath} not found", file=sys.stderr)
+    """Read a session file and extract training pairs using normalized conversation."""
-        return pairs
+    session_messages = read_session(path)
-
+    if not session_messages:
-    content = path.read_text()
+        return []
-    lines = content.strip().split("\n")
+    conversation = extract_conversation(session_messages)
-
+    # Derive session_id and model from first real message metadata
-    for line in lines:
+    first_msg = next((m for m in session_messages if m.get('role') or m.get('from')), {})
-        line = line.strip()
+    session_id = first_msg.get('meta_session_id', Path(path).name)
-        if not line:
+    model = first_msg.get('model', 'unknown')
-            continue
+    return extract_pairs_from_conversation(conversation, session_id, model, **kwargs)
-        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:

scripts/transcript_harvester.py

@@ -1,377 +0,0 @@
-#!/usr/bin/env python3
-"""
-transcript_harvester.py — Rule-based knowledge extraction from Hermes session transcripts.
-
-Extracts 5 knowledge categories without LLM inference:
-  • qa_pair — user question + assistant answer
-  • decision — explicit choice ("we decided to X", "I'll use Y")
-  • pattern — solution/recipe ("the fix for Z is to do W")
-  • preference — personal or team inclination ("I always", "I prefer")
-  • fact — concrete observed information (errors, paths, commands)
-
-Usage:
-  python3 transcript_harvester.py --session ~/.hermes/sessions/session_xxx.jsonl
-  python3 transcript_harvester.py --batch --sessions-dir ~/.hermes/sessions --limit 50
-  python3 transcript_harvester.py --session session.jsonl --output knowledge/transcripts/
-"""
-
-import argparse
-import json
-import re
-import sys
-from datetime import datetime, timezone
-from pathlib import Path
-from typing import Optional
-
-# Import session_reader from the same scripts directory
-SCRIPT_DIR = Path(__file__).parent.absolute()
-sys.path.insert(0, str(SCRIPT_DIR))
-from session_reader import read_session
-
-
-# --- Pattern matchers --------------------------------------------------------
-
-DECISION_PATTERNS = [
-    r"\b(we\s+(?:decided|chose|agreed|will|are going)\s+to\s+.*)",
-    r"\b(I\s+will\s+use|I\s+choose|I\s+am going\s+to)\s+.*",
-    r"\b(let's\s+(?:use|go\s+with|do|try))\s+.*",
-    r"\b(the\s+(?:decision|choice)\s+is)\s+.*",
-    r"\b(I'll\s+implement|I'll\s+deploy|I'll\s+create)\s+.*",
-]
-
-PATTERN_PATTERNS = [
-    r"\b(the\s+fix\s+for\s+.*\s+is\s+to\s+.*)",
-    r"\b(solution:?\s+.*)",
-    r"\b(approach:?\s+.*)",
-    r"\b(procedure:?\s+.*)",
-    r"\b(to\s+resolve\s+this.*?,\s+.*)",
-    r"\b(used\s+.*\s+to\s+.*)",  # "used X to do Y"
-    r"\b(by\s+doing\s+.*\s+we\s+.*)",
-    r"\b(Here's\s+the\s+.*\s+process:?)",  # "Here's the deployment process:"
-    r"\b(The\s+steps\s+are:?)",
-    r"\b(steps\s+to\s+.*:?)",
-    r"\b(Implementation\s+plan:?)",
-    r"\b(\d+\.\s+.*\n\d+\.)",  # numbered multi-step (at least two steps detected by newlines)
-]
-
-PREFERENCE_PATTERNS = [
-    r"\b(I\s+(?:always|never|prefer|usually|typically|generally)\s+.*)",
-    r"\b(I\s+like\s+.*)",
-    r"\b(My\s+preference\s+is\s+.*)",
-    r"\b(Alexander\s+(?:prefers|always|never).*)",
-    r"\b(We\s+always\s+.*)",
-]
-
-ERROR_PATTERNS = [
-    r"\b(error|failed|fatal|exception|denied|could\s+not|couldn't)\b.*",
-]
-
-# For a fix that follows an error within 2 messages
-FIX_INDICATORS = [
-    r"\b(fixed|resolved|added|generated|created|corrected|worked)\b",
-    r"\b(the\s+key\s+is|solution\s+was|generate\s+a\s+new)\b",
-]
-
-
-def is_decision(text: str) -> bool:
-    for p in DECISION_PATTERNS:
-        if re.search(p, text, re.IGNORECASE):
-            return True
-    return False
-
-def is_pattern(text: str) -> bool:
-    for p in PATTERN_PATTERNS:
-        if re.search(p, text, re.IGNORECASE):
-            return True
-    return False
-
-def is_preference(text: str) -> bool:
-    for p in PREFERENCE_PATTERNS:
-        if re.search(p, text, re.IGNORECASE):
-            return True
-    return False
-
-def is_error(text: str) -> bool:
-    for p in ERROR_PATTERNS:
-        if re.search(p, text, re.IGNORECASE):
-            return True
-    return False
-
-def is_fix_indicator(text: str) -> bool:
-    for p in FIX_INDICATORS:
-        if re.search(p, text, re.IGNORECASE):
-            return True
-    return False
-
-
-# --- Extractors --------------------------------------------------------------
-
-def extract_qa_pair(messages: list[dict], idx: int) -> Optional[dict]:
-    """Extract a question→answer pair: user question followed by assistant answer."""
-    if idx + 1 >= len(messages):
-        return None
-    curr = messages[idx]
-    nxt = messages[idx + 1]
-    if curr.get('role') != 'user' or nxt.get('role') != 'assistant':
-        return None
-    question = curr.get('content', '').strip()
-    answer = nxt.get('content', '').strip()
-    if not question or not answer:
-        return None
-    # Must be a real question (ends with ? or starts with WH-)
-    if not (question.endswith('?') or re.match(r'^(how|what|why|when|where|who|which|can|do|is|are)', question, re.IGNORECASE)):
-        return None
-    # Skip very short answers ("OK", "Yes")
-    if len(answer.split()) < 3:
-        return None
-    return {
-        "type": "qa_pair",
-        "question": question,
-        "answer": answer,
-        "timestamp": curr.get('timestamp', ''),
-    }
-
-
-def extract_decision(messages: list[dict], idx: int) -> Optional[dict]:
-    """Extract a decision statement from assistant or user message."""
-    msg = messages[idx]
-    text = msg.get('content', '').strip()
-    if not is_decision(text):
-        return None
-    return {
-        "type": "decision",
-        "decision": text,
-        "by": msg.get('role', 'unknown'),
-        "timestamp": msg.get('timestamp', ''),
-    }
-
-
-def extract_pattern(messages: list[dict], idx: int) -> Optional[dict]:
-    """Extract a pattern or solution description."""
-    msg = messages[idx]
-    text = msg.get('content', '').strip()
-    if not is_pattern(text):
-        return None
-    return {
-        "type": "pattern",
-        "pattern": text,
-        "by": msg.get('role', 'unknown'),
-        "timestamp": msg.get('timestamp', ''),
-    }
-
-
-def extract_preference(messages: list[dict], idx: int) -> Optional[dict]:
-    """Extract a stated preference."""
-    msg = messages[idx]
-    text = msg.get('content', '').strip()
-    if not is_preference(text):
-        return None
-    return {
-        "type": "preference",
-        "preference": text,
-        "by": msg.get('role', 'unknown'),
-        "timestamp": msg.get('timestamp', ''),
-    }
-
-
-def extract_error_fix(messages: list[dict], idx: int) -> Optional[dict]:
-    """
-    Link an error to its fix. Catch two patterns:
-    1. Error statement followed by explicit fix indicator ("fixed", "resolved")
-    2. Error statement followed by a decision statement that fixes it ("I'll generate", "I'll add")
-    """
-    msg = messages[idx]
-    if not is_error(msg.get('content', '')):
-        return None
-    error_text = msg.get('content', '').strip()
-    
-    window = min(idx + 8, len(messages))
-    for j in range(idx + 1, window):
-        follow_up = messages[j]
-        follow_text = follow_up.get('content', '').strip()
-        # Check for explicit fix indicators
-        if is_fix_indicator(follow_text):
-            return {
-                "type": "error_fix",
-                "error": error_text,
-                "fix": follow_text,
-                "error_timestamp": msg.get('timestamp', ''),
-                "fix_timestamp": follow_up.get('timestamp', ''),
-            }
-        # Check for fix decision: "I'll <action>", "Let's <action>", "We need to <action>"
-        if re.match(r"^(I'll|I will|Let's|We (will|should|need to))\s+\w+", follow_text, re.IGNORECASE):
-            return {
-                "type": "error_fix",
-                "error": error_text,
-                "fix": follow_text,
-                "error_timestamp": msg.get('timestamp', ''),
-                "fix_timestamp": follow_up.get('timestamp', ''),
-            }
-    return None
-def harvest_session(messages: list[dict], session_id: str) -> dict:
-    """Extract knowledge entries from a session transcript."""
-    entries = []
-    n = len(messages)
-
-    for i in range(n):
-        # QA pairs
-        qa = extract_qa_pair(messages, i)
-        if qa:
-            qa['session_id'] = session_id
-            entries.append(qa)
-
-        # Decisions
-        dec = extract_decision(messages, i)
-        if dec:
-            dec['session_id'] = session_id
-            entries.append(dec)
-
-        # Patterns
-        pat = extract_pattern(messages, i)
-        if pat:
-            pat['session_id'] = session_id
-            entries.append(pat)
-
-        # Preferences
-        pref = extract_preference(messages, i)
-        if pref:
-            pref['session_id'] = session_id
-            entries.append(pref)
-
-        # Error/fix pairs (spanning multiple messages)
-        ef = extract_error_fix(messages, i)
-        if ef:
-            ef['session_id'] = session_id
-            entries.append(ef)
-
-    return {
-        "session_id": session_id,
-        "message_count": n,
-        "entries": entries,
-        "counts": {
-            "qa_pair": sum(1 for e in entries if e['type'] == 'qa_pair'),
-            "decision": sum(1 for e in entries if e['type'] == 'decision'),
-            "pattern": sum(1 for e in entries if e['type'] == 'pattern'),
-            "preference": sum(1 for e in entries if e['type'] == 'preference'),
-            "error_fix": sum(1 for e in entries if e['type'] == 'error_fix'),
-        }
-    }
-
-
-def write_json_output(results: list[dict], output_path: Path):
-    """Write aggregated results to JSON."""
-    all_entries = []
-    summary = {"sessions": 0}
-    for r in results:
-        summary['sessions'] += 1
-        all_entries.extend(r['entries'])
-    
-    output = {
-        "harvester": "transcript_harvester",
-        "generated_at": datetime.now(timezone.utc).isoformat(),
-        "summary": summary,
-        "total_entries": len(all_entries),
-        "entries": all_entries,
-    }
-    output_path.write_text(json.dumps(output, indent=2, ensure_ascii=False))
-    return output
-
-
-def write_report(results: list[dict], report_path: Path):
-    """Write a human-readable markdown report."""
-    lines = []
-    lines.append("# Transcript Harvester Report")
-    lines.append(f"Generated: {datetime.now(timezone.utc).isoformat()}")
-    lines.append(f"Sessions processed: {len(results)}")
-    
-    totals = {cat: 0 for cat in ['qa_pair', 'decision', 'pattern', 'preference', 'error_fix']}
-    for r in results:
-        for cat, cnt in r['counts'].items():
-            totals[cat] += cnt  # BUG: should be += cnt
-    
-    lines.append("\n## Extracted Knowledge by Category\n")
-    for cat, cnt in totals.items():
-        lines.append(f"- **{cat}**: {cnt}")
-    
-    lines.append("\n## Sample Entries\n")
-    for r in results:
-        for entry in r['entries'][:3]:
-            lines.append(f"\n### {entry['type'].upper()} ({r['session_id']})\n")
-            if entry['type'] == 'qa_pair':
-                lines.append(f"**Q:** {entry['question']}\n")
-                lines.append(f"**A:** {entry['answer']}\n")
-            elif entry['type'] == 'decision':
-                lines.append(f"**Decision:** {entry['decision']}\n")
-                lines.append(f"By: {entry['by']}\n")
-            elif entry['type'] == 'pattern':
-                lines.append(f"**Pattern:** {entry['pattern']}\n")
-            elif entry['type'] == 'preference':
-                lines.append(f"**Preference:** {entry['preference']}\n")
-            elif entry['type'] == 'error_fix':
-                lines.append(f"**Error:** {entry['error']}\n")
-                lines.append(f"**Fixed by:** {entry['fix']}\n")
-    
-    report_path.write_text("\n".join(lines))
-
-
-def find_recent_sessions(sessions_dir: Path, limit: int = 50) -> list[Path]:
-    """Find up to `limit` most recent .jsonl session files."""
-    sessions = sorted(sessions_dir.glob("*.jsonl"), reverse=True)
-    return sessions[:limit] if limit > 0 else sessions
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Harvest knowledge from session transcripts")
-    parser.add_argument('--session', help='Single session JSONL file')
-    parser.add_argument('--batch', action='store_true', help='Batch mode')
-    parser.add_argument('--sessions-dir', default=str(Path.home() / '.hermes' / 'sessions'),
-                        help='Directory of session files')
-    parser.add_argument('--output', default='knowledge/transcripts',
-                        help='Output directory (default: knowledge/transcripts)')
-    parser.add_argument('--limit', type=int, default=50,
-                        help='Max sessions to process in batch (default: 50)')
-    
-    args = parser.parse_args()
-    output_dir = Path(args.output)
-    output_dir.mkdir(parents=True, exist_ok=True)
-    
-    results = []
-    
-    if args.session:
-        messages = read_session(args.session)
-        session_id = Path(args.session).stem
-        results.append(harvest_session(messages, session_id))
-    elif args.batch:
-        sessions_dir = Path(args.sessions_dir)
-        sessions = find_recent_sessions(sessions_dir, args.limit)
-        print(f"Processing {len(sessions)} sessions...")
-        for sf in sessions:
-            messages = read_session(str(sf))
-            results.append(harvest_session(messages, sf.stem))
-    else:
-        parser.print_help()
-        sys.exit(1)
-    
-    # Write outputs
-    json_path = output_dir / "transcript_knowledge.json"
-    report_path = output_dir / "transcript_report.md"
-    
-    output = write_json_output(results, json_path)
-    write_report(results, report_path)
-    
-    print(f"\nDone: {output['total_entries']} entries from {len(results)} sessions")
-    print(f"Output: {json_path}")
-    print(f"Report: {report_path}")
-    
-    # Print category totals
-    totals = {}
-    for r in results:
-        for cat, cnt in r['counts'].items():
-            totals[cat] = totals.get(cat, 0) + cnt
-    print("\nCategory counts:")
-    for cat, cnt in sorted(totals.items()):
-        print(f"  {cat}: {cnt}")
-
-
-if __name__ == '__main__':
-    main()

tests/test_session_pair_harvester.py

@@ -0,0 +1,118 @@
+"""
+Tests for session_pair_harvester — training pair extraction from sessions.
+"""
+
+import json
+import tempfile
+import unittest
+from pathlib import Path
+
+import sys
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).parent.parent / "scripts"))
+from session_pair_harvester import (
+    extract_pairs_from_conversation,
+    extract_from_jsonl_file,
+    deduplicate_pairs,
+    compute_hash,
+)
+
+
+class TestSessionPairHarvester(unittest.TestCase):
+    def test_compute_hash_consistent(self):
+        h1 = compute_hash("hello world")
+        h2 = compute_hash("hello world")
+        self.assertEqual(h1, h2)
+        self.assertEqual(len(h1), 16)
+
+    def test_extract_simple_qa_pair(self):
+        """A simple user→assistant exchange produces one pair."""
+        conversation = [
+            {"role": "user", "content": "What is the capital of France?"},
+            {"role": "assistant", "content": "The capital of France is Paris. It is a major European city renowned for its art, fashion, gastronomy, cultural heritage, and historical significance. The city attracts millions of tourists annually."},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "test_session", "test-model")
+        self.assertEqual(len(pairs), 1)
+        self.assertEqual(pairs[0]["terse"], "What is the capital of France?")
+        self.assertIn("Paris", pairs[0]["rich"])
+        self.assertEqual(pairs[0]["source"], "test_session")
+
+    def test_min_ratio_filter(self):
+        """Very short responses are filtered out."""
+        conversation = [
+            {"role": "user", "content": "Yes"},
+            {"role": "assistant", "content": "No."},
+        ]
+        # Default min_ratio = 1.5, min_words = 20 for response
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=3)
+        self.assertEqual(len(pairs), 0)
+
+    def test_min_words_filter(self):
+        """Assistant responses below min word count are skipped."""
+        conversation = [
+            {"role": "user", "content": "Explain the project architecture in detail"},
+            {"role": "assistant", "content": "OK."},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=5)
+        self.assertEqual(len(pairs), 0)
+
+    def test_skip_non_assistant_messages(self):
+        """System and tool messages are ignored."""
+        conversation = [
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": "Hello"},
+            {"role": "assistant", "content": "Hi there! How can I help you today?"},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=3)
+        self.assertEqual(len(pairs), 1)
+        self.assertEqual(pairs[0]["terse"], "Hello")
+
+    def test_multiple_pairs_from_one_session(self):
+        """A conversation with several Q&A turns yields multiple pairs."""
+        conversation = [
+            {"role": "user", "content": "First question?"},
+            {"role": "assistant", "content": "Here is a detailed and comprehensive answer that thoroughly explores multiple aspects of the subject. It provides background context and practical implications for the reader."},
+            {"role": "user", "content": "Second?"},
+            {"role": "assistant", "content": "Another comprehensive response with detailed examples. This includes practical code blocks and thorough explanations to ensure deep understanding of the topic at hand."},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_ratio=1.0)
+        self.assertEqual(len(pairs), 2)
+
+    def test_deduplication_removes_duplicates(self):
+        """Identical pairs across sessions are deduplicated."""
+        pairs = [
+            {"terse": "q1", "rich": "a1", "source": "s1", "model": "m"},
+            {"terse": "q1", "rich": "a1", "source": "s2", "model": "m"},
+            {"terse": "q2", "rich": "a2", "source": "s1", "model": "m"},
+        ]
+        unique = deduplicate_pairs(pairs)
+        self.assertEqual(len(unique), 2)
+        sources = {p["source"] for p in unique}
+        # First unique pair can be from either s1 or s2
+        self.assertIn("s1", sources)
+
+    def test_integration_with_test_sessions(self):
+        """Harvester finds pairs in real test session files."""
+        repo_root = Path(__file__).parent.parent
+        test_sessions_dir = repo_root / "test_sessions"
+        if not test_sessions_dir.exists():
+            self.skipTest("test_sessions not found")
+
+        pairs = []
+        for jsonl_file in sorted(test_sessions_dir.glob("*.jsonl")):
+            pairs.extend(extract_from_jsonl_file(str(jsonl_file)))
+
+        self.assertGreater(len(pairs), 0, "Should extract at least one pair from test_sessions")
+        for p in pairs:
+            self.assertIn("terse", p)
+            self.assertIn("rich", p)
+            self.assertIn("source", p)
+            self.assertIn("model", p)
+            # Verify content exists
+            self.assertGreater(len(p["terse"]), 0)
+            self.assertGreater(len(p["rich"]), 0)
+
+
+if __name__ == "__main__":
+    unittest.main()
+