test: add tests for session pair harvester (#91)

scripts/improvement_proposals.py

@@ -1,451 +0,0 @@
-#!/usr/bin/env python3
-"""
-Improvement Proposal Generator for compounding-intelligence.
-
-Analyzes fleet session data to identify waste patterns and generates
-concrete improvement proposals with ROI estimates.
-
-Input: Session analytics JSON (from fleet metrics or session database)
-Output: Markdown proposal document + JSON proposals
-
-Usage:
-    python3 scripts/improvement_proposals.py --input analytics.json
-    python3 scripts/improvement_proposals.py --input analytics.json --format json
-    python3 scripts/improvement_proposals.py --input analytics.json --output proposals.md
-    python3 scripts/improvement_proposals.py --input analytics.json --threshold 5.0
-"""
-
-import argparse
-import json
-import os
-import sys
-from collections import defaultdict
-from dataclasses import dataclass, field, asdict
-from datetime import datetime, timezone
-from typing import Any, Dict, List, Optional
-
-
-# ── Data types ─────────────────────────────────────────────────
-
-@dataclass
-class WastePattern:
-    """A detected waste pattern in fleet sessions."""
-    pattern_type: str       # "repeated_error", "manual_process", "slow_tool", "failed_retry"
-    description: str
-    occurrences: int
-    total_time_hours: float
-    affected_repos: List[str] = field(default_factory=list)
-    example_sessions: List[str] = field(default_factory=list)
-
-
-@dataclass
-class ImprovementProposal:
-    """A concrete improvement proposal with ROI estimate."""
-    title: str
-    problem: str
-    proposed_solution: str
-    estimated_weekly_hours_saved: float
-    estimated_monthly_hours_saved: float
-    implementation_hours: float
-    roi_weeks: float           # weeks to break even
-    priority: str              # "critical", "high", "medium", "low"
-    affected_area: str
-    supporting_evidence: List[str] = field(default_factory=list)
-
-
-# ── Session analysis ──────────────────────────────────────────
-
-def analyze_sessions(sessions: List[dict]) -> List[WastePattern]:
-    """
-    Analyze session data to find waste patterns.
-
-    Looks for:
-    - Repeated errors (same error across sessions)
-    - Manual processes (long sequences of similar actions)
-    - Slow tools (tools with high latency)
-    - Failed retries (multiple attempts at same task)
-    """
-    patterns = []
-
-    # Track error frequency across sessions
-    error_counts: Dict[str, List[dict]] = defaultdict(list)
-    tool_latencies: Dict[str, List[float]] = defaultdict(list)
-    retry_counts: Dict[str, int] = defaultdict(int)
-    manual_sequences: List[dict] = []
-
-    for session in sessions:
-        session_id = session.get("session_id", "unknown")
-        repo = session.get("repo", "global")
-        messages = session.get("messages", [])
-        errors = session.get("errors", [])
-        tool_calls = session.get("tool_calls", [])
-        duration_min = session.get("duration_minutes", 0)
-
-        # 1. Repeated errors
-        for err in errors:
-            err_key = _normalize_error(err.get("message", ""))
-            if err_key:
-                error_counts[err_key].append({
-                    "session_id": session_id,
-                    "repo": repo,
-                })
-
-        # 2. Tool latency tracking
-        for tc in tool_calls:
-            tool_name = tc.get("tool", "unknown")
-            latency = tc.get("latency_ms", 0)
-            if latency > 0:
-                tool_latencies[tool_name].append(latency)
-
-        # 3. Failed retries (same tool called 3+ times in sequence)
-        prev_tool = None
-        streak = 0
-        for tc in tool_calls:
-            tool_name = tc.get("tool", "unknown")
-            if tool_name == prev_tool:
-                streak += 1
-            else:
-                if streak >= 3:
-                    retry_counts[prev_tool] += 1
-                streak = 1
-            prev_tool = tool_name
-        if streak >= 3:
-            retry_counts[prev_tool] += 1
-
-        # 4. Manual processes (10+ sequential tool calls with no automation)
-        if len(tool_calls) > 10:
-            tool_sequence = [tc.get("tool") for tc in tool_calls]
-            unique_tools = set(tool_sequence)
-            if len(unique_tools) <= 3 and len(tool_calls) > 10:
-                manual_sequences.append({
-                    "session_id": session_id,
-                    "repo": repo,
-                    "tool_count": len(tool_calls),
-                    "unique_tools": list(unique_tools),
-                    "duration_min": duration_min,
-                })
-
-    # Generate patterns from collected data
-
-    # Repeated errors (appearing in 3+ sessions)
-    for err_key, occurrences in error_counts.items():
-        if len(occurrences) >= 3:
-            repos = list(set(o["repo"] for o in occurrences))
-            sessions_list = [o["session_id"] for o in occurrences[:5]]
-            # Estimate time wasted: 5 min per error occurrence
-            hours = len(occurrences) * 5 / 60
-            patterns.append(WastePattern(
-                pattern_type="repeated_error",
-                description=f"Error: {err_key[:100]}",
-                occurrences=len(occurrences),
-                total_time_hours=round(hours, 1),
-                affected_repos=repos,
-                example_sessions=sessions_list,
-            ))
-
-    # Slow tools (avg latency > 5000ms across 5+ calls)
-    for tool, latencies in tool_latencies.items():
-        if len(latencies) >= 5:
-            avg_ms = sum(latencies) / len(latencies)
-            if avg_ms > 5000:
-                hours = sum(latencies) / 1000 / 3600
-                patterns.append(WastePattern(
-                    pattern_type="slow_tool",
-                    description=f"Tool '{tool}' averages {avg_ms:.0f}ms latency",
-                    occurrences=len(latencies),
-                    total_time_hours=round(hours, 1),
-                    affected_repos=["global"],
-                    example_sessions=[],
-                ))
-
-    # Failed retries
-    for tool, count in retry_counts.items():
-        if count >= 2:
-            hours = count * 10 / 60  # ~10 min per failed retry sequence
-            patterns.append(WastePattern(
-                pattern_type="failed_retry",
-                description=f"Tool '{tool}' had {count} retry sequences (3+ consecutive calls)",
-                occurrences=count,
-                total_time_hours=round(hours, 1),
-                affected_repos=["global"],
-                example_sessions=[],
-            ))
-
-    # Manual processes
-    for seq in manual_sequences:
-        hours = seq["duration_min"] / 60
-        patterns.append(WastePattern(
-            pattern_type="manual_process",
-            description=f"Session {seq['session_id'][:12]}: {seq['tool_count']} tool calls with only {len(seq['unique_tools'])} unique tools",
-            occurrences=1,
-            total_time_hours=round(hours, 1),
-            affected_repos=[seq["repo"]],
-            example_sessions=[seq["session_id"]],
-        ))
-
-    return sorted(patterns, key=lambda p: p.total_time_hours, reverse=True)
-
-
-def _normalize_error(message: str) -> str:
-    """Normalize error message to a common key."""
-    if not message:
-        return ""
-    msg = message.lower().strip()
-    # Remove variable parts (paths, IDs, timestamps)
-    import re
-    msg = re.sub(r'/\S+', '/PATH', msg)
-    msg = re.sub(r'\b[0-9a-f]{8,}\b', 'HASH', msg)
-    msg = re.sub(r'\d{4}-\d{2}-\d{2}[tT]\d{2}:\d{2}', 'TIME', msg)
-    return msg[:150]
-
-
-# ── Proposal generation ───────────────────────────────────────
-
-def generate_proposals(
-    patterns: List[WastePattern],
-    hourly_rate: float = 50.0,
-    implementation_overhead: float = 1.5,
-) -> List[ImprovementProposal]:
-    """
-    Generate improvement proposals from waste patterns.
-
-    Args:
-        patterns: Detected waste patterns
-        hourly_rate: Developer hourly rate for ROI calculation
-        implementation_overhead: Multiplier for implementation time estimate
-    """
-    proposals = []
-
-    # Group patterns by type
-    by_type: Dict[str, List[WastePattern]] = defaultdict(list)
-    for p in patterns:
-        by_type[p.pattern_type].append(p)
-
-    # 1. Repeated errors → Create fix/skill
-    for p in by_type.get("repeated_error", []):
-        weekly_hours = p.total_time_hours / 4  # monthly → weekly
-        impl_hours = max(1.0, p.occurrences * 0.25) * implementation_overhead
-        roi_weeks = impl_hours / weekly_hours if weekly_hours > 0 else float('inf')
-
-        proposals.append(ImprovementProposal(
-            title=f"Fix repeated error: {p.description[:60]}",
-            problem=f"This error occurred {p.occurrences} times across {len(p.affected_repos)} repos, wasting ~{p.total_time_hours:.1f} hours.",
-            proposed_solution="Root-cause the error and create a permanent fix or mitigation skill.",
-            estimated_weekly_hours_saved=round(weekly_hours, 1),
-            estimated_monthly_hours_saved=round(p.total_time_hours, 1),
-            implementation_hours=round(impl_hours, 1),
-            roi_weeks=round(roi_weeks, 1),
-            priority=_priority_from_roi(roi_weeks),
-            affected_area="reliability",
-            supporting_evidence=[f"{p.occurrences} occurrences in sessions: {', '.join(p.example_sessions[:3])}"],
-        ))
-
-    # 2. Slow tools → Optimize or replace
-    for p in by_type.get("slow_tool", []):
-        weekly_hours = p.total_time_hours / 4
-        impl_hours = 3.0 * implementation_overhead  # optimization task
-        roi_weeks = impl_hours / weekly_hours if weekly_hours > 0 else float('inf')
-
-        proposals.append(ImprovementProposal(
-            title=f"Optimize slow tool: {p.description[:60]}",
-            problem=f"Tool has {p.occurrences} calls with high latency, wasting ~{p.total_time_hours:.1f} hours total.",
-            proposed_solution="Profile the tool, add caching, or replace with a faster alternative.",
-            estimated_weekly_hours_saved=round(weekly_hours, 1),
-            estimated_monthly_hours_saved=round(p.total_time_hours, 1),
-            implementation_hours=round(impl_hours, 1),
-            roi_weeks=round(roi_weeks, 1),
-            priority=_priority_from_roi(roi_weeks),
-            affected_area="performance",
-            supporting_evidence=[f"{p.occurrences} slow calls detected"],
-        ))
-
-    # 3. Failed retries → Add retry logic or validation
-    for p in by_type.get("failed_retry", []):
-        weekly_hours = p.total_time_hours / 4
-        impl_hours = 2.0 * implementation_overhead
-        roi_weeks = impl_hours / weekly_hours if weekly_hours > 0 else float('inf')
-
-        proposals.append(ImprovementProposal(
-            title=f"Reduce retries for tool '{p.description[:50]}'",
-            problem=f"Tool had {p.occurrences} retry sequences, wasting ~{p.total_time_hours:.1f} hours.",
-            proposed_solution="Add input validation, pre-flight checks, or automatic retry with backoff.",
-            estimated_weekly_hours_saved=round(weekly_hours, 1),
-            estimated_monthly_hours_saved=round(p.total_time_hours, 1),
-            implementation_hours=round(impl_hours, 1),
-            roi_weeks=round(roi_weeks, 1),
-            priority=_priority_from_roi(roi_weeks),
-            affected_area="reliability",
-            supporting_evidence=[f"{p.occurrences} retry sequences detected"],
-        ))
-
-    # 4. Manual processes → Automate
-    total_manual_hours = sum(p.total_time_hours for p in by_type.get("manual_process", []))
-    manual_patterns = by_type.get("manual_process", [])
-    if manual_patterns:
-        weekly_hours = total_manual_hours / 4
-        impl_hours = len(manual_patterns) * 2.0 * implementation_overhead
-        roi_weeks = impl_hours / weekly_hours if weekly_hours > 0 else float('inf')
-
-        proposals.append(ImprovementProposal(
-            title=f"Automate {len(manual_patterns)} manual processes",
-            problem=f"{len(manual_patterns)} sessions had long manual tool sequences, wasting ~{total_manual_hours:.1f} hours.",
-            proposed_solution="Create composite skills or scripts that combine the repeated tool sequences into single operations.",
-            estimated_weekly_hours_saved=round(weekly_hours, 1),
-            estimated_monthly_hours_saved=round(total_manual_hours, 1),
-            implementation_hours=round(impl_hours, 1),
-            roi_weeks=round(roi_weeks, 1),
-            priority=_priority_from_roi(roi_weeks),
-            affected_area="automation",
-            supporting_evidence=[f"{len(manual_patterns)} manual sessions detected"],
-        ))
-
-    return sorted(proposals, key=lambda p: p.estimated_monthly_hours_saved, reverse=True)
-
-
-def _priority_from_roi(roi_weeks: float) -> str:
-    """Determine priority from ROI break-even time."""
-    if roi_weeks <= 1:
-        return "critical"
-    elif roi_weeks <= 4:
-        return "high"
-    elif roi_weeks <= 12:
-        return "medium"
-    return "low"
-
-
-# ── Output formatting ─────────────────────────────────────────
-
-def format_proposals_markdown(
-    proposals: List[ImprovementProposal],
-    patterns: List[WastePattern],
-    generated_at: str,
-) -> str:
-    """Format proposals as a markdown document."""
-    lines = [
-        "# Improvement Proposals",
-        "",
-        f"Generated: {generated_at}",
-        f"Based on analysis of {sum(p.occurrences for p in patterns)} waste events across {len(set(r for p in patterns for r in p.affected_repos))} repos.",
-        "",
-        "---",
-        "",
-        "## Summary",
-        "",
-        f"| Metric | Value |",
-        f"|--------|-------|",
-        f"| Total proposals | {len(proposals)} |",
-        f"| Critical priority | {sum(1 for p in proposals if p.priority == 'critical')} |",
-        f"| Total monthly hours wasted | {sum(p.estimated_monthly_hours_saved for p in proposals):.1f}h |",
-        f"| Total weekly hours recoverable | {sum(p.estimated_weekly_hours_saved for p in proposals):.1f}h |",
-        f"| Implementation investment | {sum(p.implementation_hours for p in proposals):.1f}h |",
-        "",
-        "---",
-        "",
-    ]
-
-    for i, prop in enumerate(proposals, 1):
-        lines.extend([
-            f"## {i}. {prop.title}",
-            "",
-            f"**Priority:** {prop.priority.upper()}  ",
-            f"**Area:** {prop.affected_area}  ",
-            f"**ROI break-even:** {prop.roi_weeks:.1f} weeks",
-            "",
-            "### Problem",
-            "",
-            prop.problem,
-            "",
-            "### Proposed Solution",
-            "",
-            prop.proposed_solution,
-            "",
-            "### ROI Estimate",
-            "",
-            f"- Weekly hours saved: **{prop.estimated_weekly_hours_saved}h**",
-            f"- Monthly hours saved: **{prop.estimated_monthly_hours_saved}h**",
-            f"- Implementation time: **{prop.implementation_hours}h**",
-            f"- Break-even: **{prop.roi_weeks:.1f} weeks**",
-            "",
-        ])
-
-        if prop.supporting_evidence:
-            lines.extend([
-                "### Evidence",
-                "",
-            ])
-            for ev in prop.supporting_evidence:
-                lines.append(f"- {ev}")
-            lines.append("")
-
-        lines.extend(["---", ""])
-
-    # Waste pattern appendix
-    lines.extend([
-        "## Appendix: Detected Waste Patterns",
-        "",
-        "| Type | Description | Occurrences | Hours |",
-        "|------|-------------|-------------|-------|",
-    ])
-    for p in patterns[:20]:
-        lines.append(
-            f"| {p.pattern_type} | {p.description[:60]} | {p.occurrences} | {p.total_time_hours}h |"
-        )
-    lines.append("")
-
-    return "\n".join(lines)
-
-
-def format_proposals_json(proposals: List[ImprovementProposal]) -> str:
-    """Format proposals as JSON."""
-    return json.dumps(
-        [asdict(p) for p in proposals],
-        indent=2,
-        default=str,
-    )
-
-
-# ── Main ──────────────────────────────────────────────────────
-
-def main():
-    parser = argparse.ArgumentParser(description="Generate improvement proposals from session analytics")
-    parser.add_argument("--input", required=True, help="Path to session analytics JSON file")
-    parser.add_argument("--output", help="Output file path (default: stdout)")
-    parser.add_argument("--format", choices=["markdown", "json"], default="markdown", help="Output format")
-    parser.add_argument("--hourly-rate", type=float, default=50.0, help="Developer hourly rate for ROI calc")
-    parser.add_argument("--threshold", type=float, default=2.0, help="Min monthly hours to include proposal")
-
-    args = parser.parse_args()
-
-    with open(args.input) as f:
-        data = json.load(f)
-
-    sessions = data if isinstance(data, list) else data.get("sessions", [])
-
-    if not sessions:
-        print("No sessions found in input file.", file=sys.stderr)
-        sys.exit(1)
-
-    # Analyze
-    patterns = analyze_sessions(sessions)
-    proposals = generate_proposals(patterns, hourly_rate=args.hourly_rate)
-
-    # Filter by threshold
-    proposals = [p for p in proposals if p.estimated_monthly_hours_saved >= args.threshold]
-
-    generated_at = datetime.now(timezone.utc).isoformat()
-
-    if args.format == "markdown":
-        output = format_proposals_markdown(proposals, patterns, generated_at)
-    else:
-        output = format_proposals_json(proposals)
-
-    if args.output:
-        with open(args.output, "w") as f:
-            f.write(output)
-        print(f"Wrote {len(proposals)} proposals to {args.output}", file=sys.stderr)
-    else:
-        print(output)
-
-
-if __name__ == "__main__":
-    main()

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/session_pair_harvester.py

@@ -0,0 +1,234 @@
+#!/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_improvement_proposals.py

@@ -1,282 +0,0 @@
-#!/usr/bin/env python3
-"""Tests for scripts/improvement_proposals.py — 15 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(
-    "ip", os.path.join(os.path.dirname(__file__) or ".", "improvement_proposals.py")
-)
-mod = importlib.util.module_from_spec(spec)
-spec.loader.exec_module(mod)
-
-analyze_sessions = mod.analyze_sessions
-generate_proposals = mod.generate_proposals
-format_proposals_markdown = mod.format_proposals_markdown
-format_proposals_json = mod.format_proposals_json
-_normalize_error = mod._normalize_error
-
-
-# ── Helper to build test sessions ─────────────────────────────
-
-def _make_session(session_id, repo="test-repo", errors=None, tool_calls=None, duration=30):
-    return {
-        "session_id": session_id,
-        "repo": repo,
-        "errors": [{"message": e} for e in (errors or [])],
-        "tool_calls": tool_calls or [],
-        "duration_minutes": duration,
-        "messages": [],
-    }
-
-
-def _make_tool_calls(repeats):
-    """Create tool call list with repeated tools."""
-    calls = []
-    for tool, count in repeats:
-        for _ in range(count):
-            calls.append({"tool": tool, "latency_ms": 100})
-    return calls
-
-
-# ── Tests ─────────────────────────────────────────────────────
-
-def test_empty_sessions():
-    patterns = analyze_sessions([])
-    assert patterns == []
-    print("PASS: test_empty_sessions")
-
-
-def test_no_patterns_on_clean_sessions():
-    sessions = [
-        _make_session("s1", tool_calls=[{"tool": "read_file", "latency_ms": 50}]),
-        _make_session("s2", tool_calls=[{"tool": "write_file", "latency_ms": 80}]),
-    ]
-    patterns = analyze_sessions(sessions)
-    # No repeated errors, no slow tools, no retries
-    assert len(patterns) == 0
-    print("PASS: test_no_patterns_on_clean_sessions")
-
-
-def test_repeated_error_detection():
-    """Same error across 3+ sessions triggers pattern."""
-    sessions = [
-        _make_session(f"s{i}", errors=["ModuleNotFoundError: No module named bannerlord_trace"])
-        for i in range(4)
-    ]
-    patterns = analyze_sessions(sessions)
-    repeated = [p for p in patterns if p.pattern_type == "repeated_error"]
-    assert len(repeated) == 1
-    assert repeated[0].occurrences == 4
-    print("PASS: test_repeated_error_detection")
-
-
-def test_repeated_error_threshold():
-    """2 occurrences should NOT trigger (threshold is 3)."""
-    sessions = [
-        _make_session("s1", errors=["TimeoutError: connection timed out"]),
-        _make_session("s2", errors=["TimeoutError: connection timed out"]),
-    ]
-    patterns = analyze_sessions(sessions)
-    repeated = [p for p in patterns if p.pattern_type == "repeated_error"]
-    assert len(repeated) == 0
-    print("PASS: test_repeated_error_threshold")
-
-
-def test_slow_tool_detection():
-    """Tool with avg latency > 5000ms across 5+ calls."""
-    calls = [{"tool": "git_push", "latency_ms": 8000} for _ in range(10)]
-    sessions = [_make_session("s1", tool_calls=calls)]
-    patterns = analyze_sessions(sessions)
-    slow = [p for p in patterns if p.pattern_type == "slow_tool"]
-    assert len(slow) == 1
-    assert "git_push" in slow[0].description
-    print("PASS: test_slow_tool_detection")
-
-
-def test_fast_tool_not_flagged():
-    """Tool under 5000ms avg should not trigger."""
-    calls = [{"tool": "read_file", "latency_ms": 50} for _ in range(10)]
-    sessions = [_make_session("s1", tool_calls=calls)]
-    patterns = analyze_sessions(sessions)
-    slow = [p for p in patterns if p.pattern_type == "slow_tool"]
-    assert len(slow) == 0
-    print("PASS: test_fast_tool_not_flagged")
-
-
-def test_failed_retry_detection():
-    """3+ consecutive calls to same tool triggers retry pattern."""
-    calls = _make_tool_calls([("execute_code", 5)])
-    sessions = [_make_session("s1", tool_calls=calls)]
-    sessions.extend([
-        _make_session(f"s{i}", tool_calls=_make_tool_calls([("execute_code", 4)]))
-        for i in range(2, 5)
-    ])
-    patterns = analyze_sessions(sessions)
-    retries = [p for p in patterns if p.pattern_type == "failed_retry"]
-    assert len(retries) >= 1
-    print("PASS: test_failed_retry_detection")
-
-
-def test_manual_process_detection():
-    """10+ tool calls with <= 3 unique tools."""
-    calls = _make_tool_calls([("terminal", 8), ("read_file", 5)])
-    sessions = [_make_session("s1", tool_calls=calls, duration=25)]
-    patterns = analyze_sessions(sessions)
-    manual = [p for p in patterns if p.pattern_type == "manual_process"]
-    assert len(manual) == 1
-    print("PASS: test_manual_process_detection")
-
-
-def test_generate_proposals_from_patterns():
-    """Proposals generated from waste patterns."""
-    sessions = [
-        _make_session(f"s{i}", errors=["Error: push timeout"])
-        for i in range(5)
-    ]
-    patterns = analyze_sessions(sessions)
-    proposals = generate_proposals(patterns)
-    assert len(proposals) >= 1
-    assert proposals[0].estimated_monthly_hours_saved > 0
-    assert proposals[0].priority in ("critical", "high", "medium", "low")
-    print("PASS: test_generate_proposals_from_patterns")
-
-
-def test_proposal_roi_positive():
-    """ROI weeks should be a positive number for recoverable time."""
-    patterns = [mod.WastePattern(
-        pattern_type="repeated_error",
-        description="Test error",
-        occurrences=10,
-        total_time_hours=5.0,
-        affected_repos=["test"],
-    )]
-    proposals = generate_proposals(patterns)
-    assert len(proposals) == 1
-    assert proposals[0].roi_weeks > 0
-    assert proposals[0].roi_weeks < 100
-    print("PASS: test_proposal_roi_positive")
-
-
-def test_proposals_sorted_by_impact():
-    """Proposals should be sorted by monthly hours saved (descending)."""
-    sessions = [
-        _make_session("s1", errors=["Minor warning"] * 3, duration=5),
-        _make_session("s2", errors=["Critical failure: deploy crashed"] * 5, duration=60),
-    ]
-    # Add more sessions to cross threshold
-    for i in range(3, 7):
-        sessions.append(_make_session(f"s{i}", errors=["Critical failure: deploy crashed"]))
-
-    patterns = analyze_sessions(sessions)
-    proposals = generate_proposals(patterns)
-    if len(proposals) >= 2:
-        for i in range(len(proposals) - 1):
-            assert proposals[i].estimated_monthly_hours_saved >= proposals[i + 1].estimated_monthly_hours_saved
-    print("PASS: test_proposals_sorted_by_impact")
-
-
-def test_format_markdown():
-    """Markdown output should contain expected sections."""
-    patterns = [mod.WastePattern(
-        pattern_type="repeated_error", description="Test", occurrences=5,
-        total_time_hours=2.5, affected_repos=["repo"],
-    )]
-    proposals = generate_proposals(patterns)
-    md = format_proposals_markdown(proposals, patterns, "2026-04-15T00:00:00Z")
-    assert "# Improvement Proposals" in md
-    assert "## Summary" in md
-    assert "### Problem" in md
-    assert "### ROI Estimate" in md
-    assert "## Appendix" in md
-    print("PASS: test_format_markdown")
-
-
-def test_format_json():
-    """JSON output should be valid and parseable."""
-    patterns = [mod.WastePattern(
-        pattern_type="slow_tool", description="Slow", occurrences=10,
-        total_time_hours=3.0, affected_repos=["global"],
-    )]
-    proposals = generate_proposals(patterns)
-    output = format_proposals_json(proposals)
-    parsed = json.loads(output)
-    assert isinstance(parsed, list)
-    assert len(parsed) == len(proposals)
-    assert "title" in parsed[0]
-    assert "roi_weeks" in parsed[0]
-    print("PASS: test_format_json")
-
-
-def test_normalize_error():
-    """Error normalization should remove paths and hashes."""
-    err1 = _normalize_error("Failed to clone /Users/apayne/repo with token abc123def456")
-    assert "/PATH" in err1
-    assert "HASH" in err1
-    assert "/Users/apayne" not in err1
-
-    err2 = _normalize_error("")
-    assert err2 == ""
-
-    err3 = _normalize_error("Simple error message")
-    assert "simple error" in err3
-    print("PASS: test_normalize_error")
-
-
-def test_cli_integration():
-    """End-to-end test: write input JSON, run script, check output."""
-    import subprocess
-
-    sessions = [
-        _make_session(f"s{i}", errors=["Connection refused: port 8080"])
-        for i in range(5)
-    ]
-
-    with tempfile.TemporaryDirectory() as tmpdir:
-        input_path = os.path.join(tmpdir, "analytics.json")
-        output_path = os.path.join(tmpdir, "proposals.md")
-
-        with open(input_path, "w") as f:
-            json.dump({"sessions": sessions}, f)
-
-        script = os.path.join(os.path.dirname(__file__) or ".", "improvement_proposals.py")
-        result = subprocess.run(
-            [sys.executable, script, "--input", input_path, "--output", output_path],
-            capture_output=True, text=True, timeout=10,
-        )
-
-        assert result.returncode == 0, f"CLI failed: {result.stderr}"
-        assert os.path.exists(output_path)
-
-        with open(output_path) as f:
-            content = f.read()
-        assert "# Improvement Proposals" in content
-        print("PASS: test_cli_integration")
-
-
-def run_all():
-    test_empty_sessions()
-    test_no_patterns_on_clean_sessions()
-    test_repeated_error_detection()
-    test_repeated_error_threshold()
-    test_slow_tool_detection()
-    test_fast_tool_not_flagged()
-    test_failed_retry_detection()
-    test_manual_process_detection()
-    test_generate_proposals_from_patterns()
-    test_proposal_roi_positive()
-    test_proposals_sorted_by_impact()
-    test_format_markdown()
-    test_format_json()
-    test_normalize_error()
-    test_cli_integration()
-    print("\nAll 15 tests passed!")
-
-
-if __name__ == "__main__":
-    run_all()

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()

scripts/test_session_pair_harvester.py

@@ -0,0 +1,90 @@
+#!/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.")