myaidev-method 0.3.3 → 0.3.5

package/skills/myai-skill-builder/scripts/aggregate_benchmark.py

@@ -0,0 +1,190 @@
+#!/usr/bin/env python3
+"""Aggregate grading.json and timing.json files into a benchmark.json summary.
+
+Usage:
+    python3 aggregate_benchmark.py <iteration-dir>
+    python3 aggregate_benchmark.py <workspace-dir> --multi-run
+
+Single iteration mode:
+    Reads all grading.json and timing.json from an iteration directory,
+    produces benchmark.json with pass rates, tokens, and timing per config.
+
+Multi-run mode (--multi-run):
+    Reads benchmark.json from multiple iteration directories,
+    computes mean ± stddev across runs.
+"""
+
+import argparse
+import sys
+from pathlib import Path
+
+# Add script directory to path for utils import
+sys.path.insert(0, str(Path(__file__).parent))
+from utils import (
+    load_json, save_json, collect_grading_files, collect_timing_files,
+    mean, stddev, now_iso
+)
+
+
+def aggregate_iteration(iteration_dir):
+    """Aggregate results from a single iteration into benchmark.json."""
+    iteration_dir = Path(iteration_dir)
+    gradings = collect_grading_files(iteration_dir)
+    timings = collect_timing_files(iteration_dir)
+
+    if not gradings:
+        print(f"No grading.json files found in {iteration_dir}", file=sys.stderr)
+        sys.exit(1)
+
+    # Group by config
+    configs = {}
+    for g in gradings:
+        config = g.get("config", "unknown")
+        if config not in configs:
+            configs[config] = {"gradings": [], "timings": []}
+        configs[config]["gradings"].append(g)
+
+    for t in timings:
+        config = t.get("config", "unknown")
+        if config in configs:
+            configs[config]["timings"].append(t)
+
+    # Build benchmark
+    benchmark = {
+        "iteration": int(iteration_dir.name.split("-")[1]) if "-" in iteration_dir.name else 1,
+        "timestamp": now_iso(),
+        "configs": {},
+        "comparison": {}
+    }
+
+    for config_name, data in configs.items():
+        pass_rates = [g["pass_rate"] for g in data["gradings"]]
+        tokens = [t["total_tokens"] for t in data["timings"]]
+        durations = [t["duration_ms"] for t in data["timings"]]
+
+        evals = []
+        for g in data["gradings"]:
+            eval_entry = {
+                "eval_id": g["eval_id"],
+                "pass_rate": g["pass_rate"],
+                "pass_count": g["pass_count"],
+                "fail_count": g["fail_count"],
+            }
+            # Find matching timing
+            matching_timing = [t for t in data["timings"] if t["eval_id"] == g["eval_id"]]
+            if matching_timing:
+                eval_entry["tokens"] = matching_timing[0]["total_tokens"]
+                eval_entry["duration_ms"] = matching_timing[0]["duration_ms"]
+            evals.append(eval_entry)
+
+        benchmark["configs"][config_name] = {
+            "overall_pass_rate": mean(pass_rates),
+            "total_tokens_mean": mean(tokens) if tokens else 0,
+            "total_tokens_stddev": stddev(tokens) if tokens else 0,
+            "duration_ms_mean": mean(durations) if durations else 0,
+            "duration_ms_stddev": stddev(durations) if durations else 0,
+            "evals": evals
+        }
+
+    # Compute comparison if both configs exist
+    if "with_skill" in benchmark["configs"] and "without_skill" in benchmark["configs"]:
+        ws = benchmark["configs"]["with_skill"]
+        wo = benchmark["configs"]["without_skill"]
+        benchmark["comparison"] = {
+            "pass_rate_delta": ws["overall_pass_rate"] - wo["overall_pass_rate"],
+            "token_overhead_percent": (
+                ((ws["total_tokens_mean"] - wo["total_tokens_mean"]) / wo["total_tokens_mean"] * 100)
+                if wo["total_tokens_mean"] > 0 else 0
+            ),
+            "time_overhead_percent": (
+                ((ws["duration_ms_mean"] - wo["duration_ms_mean"]) / wo["duration_ms_mean"] * 100)
+                if wo["duration_ms_mean"] > 0 else 0
+            ),
+        }
+
+        # Find non-discriminating assertions
+        non_disc = []
+        ws_evals = {e["eval_id"]: e for e in ws["evals"]}
+        wo_evals = {e["eval_id"]: e for e in wo["evals"]}
+        for eval_id in ws_evals:
+            if eval_id in wo_evals:
+                if ws_evals[eval_id]["pass_rate"] == 1.0 and wo_evals[eval_id]["pass_rate"] == 1.0:
+                    non_disc.append(eval_id)
+        benchmark["comparison"]["non_discriminating_evals"] = non_disc
+
+    output_path = iteration_dir / "benchmark.json"
+    save_json(output_path, benchmark)
+    return benchmark
+
+
+def aggregate_multi_run(workspace_dir):
+    """Aggregate benchmark.json files from multiple iterations for statistical analysis."""
+    workspace_dir = Path(workspace_dir)
+    benchmarks = []
+
+    for d in sorted(workspace_dir.iterdir()):
+        if d.is_dir() and d.name.startswith("iteration-"):
+            bm_path = d / "benchmark.json"
+            if bm_path.exists():
+                benchmarks.append(load_json(bm_path))
+
+    if len(benchmarks) < 2:
+        print(f"Need at least 2 iteration benchmarks for multi-run analysis, found {len(benchmarks)}", file=sys.stderr)
+        sys.exit(1)
+
+    result = {
+        "timestamp": now_iso(),
+        "num_runs": len(benchmarks),
+        "configs": {}
+    }
+
+    # Aggregate per config
+    for config_name in ["with_skill", "without_skill"]:
+        pass_rates = [b["configs"][config_name]["overall_pass_rate"]
+                      for b in benchmarks if config_name in b.get("configs", {})]
+        tokens = [b["configs"][config_name]["total_tokens_mean"]
+                  for b in benchmarks if config_name in b.get("configs", {})]
+        durations = [b["configs"][config_name]["duration_ms_mean"]
+                     for b in benchmarks if config_name in b.get("configs", {})]
+
+        if pass_rates:
+            result["configs"][config_name] = {
+                "pass_rate": {"mean": mean(pass_rates), "stddev": stddev(pass_rates)},
+                "tokens": {"mean": mean(tokens), "stddev": stddev(tokens)},
+                "duration_ms": {"mean": mean(durations), "stddev": stddev(durations)},
+            }
+
+    output_path = workspace_dir / "benchmark.json"
+    save_json(output_path, result)
+    return result
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Aggregate eval results into benchmark summary")
+    parser.add_argument("path", help="Iteration directory or workspace directory (with --multi-run)")
+    parser.add_argument("--multi-run", action="store_true", help="Aggregate across multiple iterations")
+    args = parser.parse_args()
+
+    if args.multi_run:
+        result = aggregate_multi_run(args.path)
+        print(f"\nMulti-run benchmark ({result['num_runs']} runs):")
+        for config, stats in result["configs"].items():
+            pr = stats["pass_rate"]
+            print(f"  {config}: pass_rate={pr['mean']:.2f} ± {pr['stddev']:.2f}")
+    else:
+        result = aggregate_iteration(args.path)
+        print(f"\nIteration {result['iteration']} benchmark:")
+        for config, stats in result["configs"].items():
+            print(f"  {config}: pass_rate={stats['overall_pass_rate']:.2f}, "
+                  f"tokens={stats['total_tokens_mean']:.0f}, "
+                  f"time={stats['duration_ms_mean']:.0f}ms")
+        if result.get("comparison"):
+            c = result["comparison"]
+            print(f"\n  Comparison:")
+            print(f"    Pass rate delta: +{c['pass_rate_delta']:.2f}")
+            print(f"    Token overhead: {c['token_overhead_percent']:.1f}%")
+            print(f"    Time overhead: {c['time_overhead_percent']:.1f}%")
+
+
+if __name__ == "__main__":
+    main()

package/skills/myai-skill-builder/scripts/generate_review.py

@@ -0,0 +1,381 @@
+#!/usr/bin/env python3
+"""Generate an HTML review page for eval results.
+
+Creates an interactive HTML page showing eval outputs, grading results,
+and benchmark data for human review.
+
+Usage:
+    python3 generate_review.py <workspace-dir>
+    python3 generate_review.py <workspace-dir> --static output.html
+    python3 generate_review.py <workspace-dir> --open
+
+Options:
+    --static <path>   Generate a standalone HTML file instead of serving
+    --open            Open the generated HTML in the default browser
+    --iteration <n>   Show specific iteration (default: latest)
+"""
+
+import argparse
+import json
+import os
+import sys
+import webbrowser
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).parent))
+from utils import load_json, find_latest_iteration
+
+
+def collect_eval_data(iteration_dir):
+    """Collect all eval data from an iteration directory."""
+    iteration_dir = Path(iteration_dir)
+    evals = []
+
+    for eval_dir in sorted(iteration_dir.iterdir()):
+        if not eval_dir.is_dir() or not eval_dir.name.startswith("eval-"):
+            continue
+
+        eval_id = eval_dir.name.replace("eval-", "")
+        eval_data = {"id": eval_id, "configs": {}}
+
+        for config in ["with_skill", "without_skill"]:
+            config_dir = eval_dir / config
+            if not config_dir.exists():
+                continue
+
+            config_data = {"outputs": {}, "grading": None, "timing": None}
+
+            # Read outputs
+            outputs_dir = config_dir / "outputs"
+            if outputs_dir.exists():
+                for f in sorted(outputs_dir.iterdir()):
+                    if f.is_file():
+                        try:
+                            config_data["outputs"][f.name] = f.read_text()[:5000]
+                        except (UnicodeDecodeError, PermissionError):
+                            config_data["outputs"][f.name] = "(binary file)"
+
+            # Read grading
+            grading_path = config_dir / "grading.json"
+            if grading_path.exists():
+                config_data["grading"] = load_json(grading_path)
+
+            # Read timing
+            timing_path = config_dir / "timing.json"
+            if timing_path.exists():
+                config_data["timing"] = load_json(timing_path)
+
+            eval_data["configs"][config] = config_data
+
+        evals.append(eval_data)
+
+    # Read benchmark
+    benchmark = None
+    benchmark_path = iteration_dir / "benchmark.json"
+    if benchmark_path.exists():
+        benchmark = load_json(benchmark_path)
+
+    return evals, benchmark
+
+
+def generate_html(evals, benchmark, iteration_num, workspace_dir):
+    """Generate the HTML review page."""
+    # Escape JSON for embedding in HTML
+    evals_json = json.dumps(evals, indent=2)
+    benchmark_json = json.dumps(benchmark, indent=2) if benchmark else "null"
+
+    html = f"""<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Skill Eval Review - Iteration {iteration_num}</title>
+    <style>
+        * {{ margin: 0; padding: 0; box-sizing: border-box; }}
+        body {{ font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, monospace; background: #0d1117; color: #c9d1d9; }}
+        .header {{ background: #161b22; border-bottom: 1px solid #30363d; padding: 16px 24px; display: flex; justify-content: space-between; align-items: center; }}
+        .header h1 {{ font-size: 20px; color: #f0f6fc; }}
+        .tabs {{ display: flex; gap: 0; border-bottom: 1px solid #30363d; background: #161b22; padding: 0 24px; }}
+        .tab {{ padding: 12px 20px; cursor: pointer; border-bottom: 2px solid transparent; color: #8b949e; font-size: 14px; }}
+        .tab:hover {{ color: #c9d1d9; }}
+        .tab.active {{ color: #f0f6fc; border-bottom-color: #f78166; }}
+        .content {{ padding: 24px; max-width: 1400px; margin: 0 auto; }}
+        .eval-card {{ background: #161b22; border: 1px solid #30363d; border-radius: 6px; margin-bottom: 16px; overflow: hidden; }}
+        .eval-header {{ padding: 12px 16px; background: #1c2128; border-bottom: 1px solid #30363d; display: flex; justify-content: space-between; align-items: center; }}
+        .eval-header h3 {{ font-size: 16px; color: #f0f6fc; }}
+        .badge {{ padding: 2px 8px; border-radius: 12px; font-size: 12px; font-weight: 600; }}
+        .badge.pass {{ background: #1a3a2a; color: #3fb950; }}
+        .badge.fail {{ background: #3a1a1a; color: #f85149; }}
+        .badge.mixed {{ background: #3a2a1a; color: #d29922; }}
+        .eval-body {{ padding: 16px; }}
+        .config-columns {{ display: grid; grid-template-columns: 1fr 1fr; gap: 16px; }}
+        .config-col {{ background: #0d1117; border: 1px solid #30363d; border-radius: 6px; padding: 12px; }}
+        .config-col h4 {{ font-size: 14px; margin-bottom: 8px; color: #8b949e; }}
+        .assertion {{ padding: 8px; margin: 4px 0; border-radius: 4px; font-size: 13px; }}
+        .assertion.pass {{ background: #1a3a2a; border-left: 3px solid #3fb950; }}
+        .assertion.fail {{ background: #3a1a1a; border-left: 3px solid #f85149; }}
+        .evidence {{ color: #8b949e; font-size: 12px; margin-top: 4px; }}
+        .output-block {{ background: #0d1117; border: 1px solid #30363d; border-radius: 4px; padding: 12px; margin: 8px 0; max-height: 300px; overflow-y: auto; font-size: 13px; white-space: pre-wrap; }}
+        .benchmark-grid {{ display: grid; grid-template-columns: repeat(auto-fit, minmax(250px, 1fr)); gap: 16px; }}
+        .metric-card {{ background: #161b22; border: 1px solid #30363d; border-radius: 6px; padding: 16px; text-align: center; }}
+        .metric-value {{ font-size: 32px; font-weight: 700; color: #f0f6fc; margin: 8px 0; }}
+        .metric-label {{ font-size: 13px; color: #8b949e; }}
+        .metric-sub {{ font-size: 12px; color: #8b949e; margin-top: 4px; }}
+        .feedback-section {{ margin-top: 16px; }}
+        .feedback-section textarea {{ width: 100%; height: 80px; background: #0d1117; border: 1px solid #30363d; border-radius: 6px; padding: 8px; color: #c9d1d9; font-family: inherit; font-size: 13px; resize: vertical; }}
+        .feedback-section select {{ background: #0d1117; border: 1px solid #30363d; border-radius: 6px; padding: 6px 12px; color: #c9d1d9; font-size: 13px; margin-right: 8px; }}
+        .btn {{ padding: 8px 16px; border-radius: 6px; border: none; cursor: pointer; font-size: 14px; font-weight: 600; }}
+        .btn-primary {{ background: #238636; color: #fff; }}
+        .btn-primary:hover {{ background: #2ea043; }}
+        .submit-bar {{ position: fixed; bottom: 0; left: 0; right: 0; background: #161b22; border-top: 1px solid #30363d; padding: 12px 24px; display: flex; justify-content: flex-end; gap: 12px; }}
+        .hidden {{ display: none; }}
+        .bar-chart {{ display: flex; align-items: flex-end; gap: 8px; height: 120px; margin: 16px 0; }}
+        .bar {{ flex: 1; border-radius: 4px 4px 0 0; min-width: 30px; position: relative; }}
+        .bar.with {{ background: #238636; }}
+        .bar.without {{ background: #6e7681; }}
+        .bar-label {{ position: absolute; bottom: -20px; left: 50%; transform: translateX(-50%); font-size: 10px; white-space: nowrap; }}
+    </style>
+</head>
+<body>
+    <div class="header">
+        <h1>Skill Eval Review &mdash; Iteration {iteration_num}</h1>
+        <span style="color: #8b949e; font-size: 13px;">{workspace_dir}</span>
+    </div>
+    <div class="tabs">
+        <div class="tab active" onclick="showTab('outputs')">Outputs</div>
+        <div class="tab" onclick="showTab('benchmark')">Benchmark</div>
+    </div>
+    <div id="outputs-tab" class="content"></div>
+    <div id="benchmark-tab" class="content hidden"></div>
+    <div class="submit-bar">
+        <select id="action-select">
+            <option value="iterate">Iterate (improve &amp; rerun)</option>
+            <option value="publish">Publish (submit to marketplace)</option>
+            <option value="stop">Stop (done for now)</option>
+        </select>
+        <button class="btn btn-primary" onclick="submitFeedback()">Submit All Reviews</button>
+    </div>
+
+    <script>
+    const evals = {evals_json};
+    const benchmark = {benchmark_json};
+
+    function showTab(name) {{
+        document.querySelectorAll('.tab').forEach(t => t.classList.remove('active'));
+        document.querySelectorAll('.content').forEach(c => c.classList.add('hidden'));
+        event.target.classList.add('active');
+        document.getElementById(name + '-tab').classList.remove('hidden');
+    }}
+
+    function renderOutputs() {{
+        const container = document.getElementById('outputs-tab');
+        let html = '';
+
+        evals.forEach(ev => {{
+            const wsGrading = ev.configs.with_skill?.grading;
+            const woGrading = ev.configs.without_skill?.grading;
+            const wsRate = wsGrading ? wsGrading.pass_rate : null;
+            const woRate = woGrading ? woGrading.pass_rate : null;
+
+            let badgeClass = 'mixed';
+            let badgeText = 'N/A';
+            if (wsRate !== null) {{
+                if (wsRate === 1.0) {{ badgeClass = 'pass'; badgeText = 'ALL PASS'; }}
+                else if (wsRate === 0) {{ badgeClass = 'fail'; badgeText = 'ALL FAIL'; }}
+                else {{ badgeText = Math.round(wsRate * 100) + '% pass'; }}
+            }}
+
+            html += '<div class="eval-card">';
+            html += '<div class="eval-header">';
+            html += '<h3>eval-' + ev.id + '</h3>';
+            html += '<span class="badge ' + badgeClass + '">' + badgeText + '</span>';
+            html += '</div>';
+            html += '<div class="eval-body">';
+            html += '<div class="config-columns">';
+
+            ['with_skill', 'without_skill'].forEach(config => {{
+                const data = ev.configs[config];
+                html += '<div class="config-col">';
+                html += '<h4>' + config.replace('_', ' ') + '</h4>';
+
+                if (data && data.grading) {{
+                    data.grading.expectations.forEach(exp => {{
+                        html += '<div class="assertion ' + exp.verdict.toLowerCase() + '">';
+                        html += '<strong>' + exp.verdict + '</strong>: ' + exp.assertion;
+                        html += '<div class="evidence">' + (exp.evidence || '') + '</div>';
+                        html += '</div>';
+                    }});
+                }}
+
+                if (data && Object.keys(data.outputs).length > 0) {{
+                    html += '<details><summary style="cursor:pointer;margin-top:8px;color:#8b949e">Show outputs</summary>';
+                    Object.entries(data.outputs).forEach(([name, content]) => {{
+                        html += '<div style="margin-top:4px;font-size:12px;color:#8b949e">' + name + '</div>';
+                        html += '<div class="output-block">' + escapeHtml(content) + '</div>';
+                    }});
+                    html += '</details>';
+                }}
+
+                if (data && data.timing) {{
+                    html += '<div style="margin-top:8px;font-size:12px;color:#8b949e">';
+                    html += 'Tokens: ' + data.timing.total_tokens + ' | Time: ' + data.timing.duration_ms + 'ms';
+                    html += '</div>';
+                }}
+
+                html += '</div>';
+            }});
+
+            html += '</div>';
+            html += '<div class="feedback-section">';
+            html += '<select data-eval="' + ev.id + '" class="eval-rating">';
+            html += '<option value="good">Good</option>';
+            html += '<option value="needs-work">Needs Work</option>';
+            html += '<option value="bad">Bad</option>';
+            html += '</select>';
+            html += '<textarea data-eval="' + ev.id + '" class="eval-comment" placeholder="Feedback for this eval..."></textarea>';
+            html += '</div>';
+            html += '</div></div>';
+        }});
+
+        container.innerHTML = html;
+    }}
+
+    function renderBenchmark() {{
+        const container = document.getElementById('benchmark-tab');
+        if (!benchmark) {{
+            container.innerHTML = '<p style="color:#8b949e">No benchmark data available. Run evals first.</p>';
+            return;
+        }}
+
+        let html = '<div class="benchmark-grid">';
+
+        const configs = benchmark.configs || {{}};
+        ['with_skill', 'without_skill'].forEach(config => {{
+            const data = configs[config];
+            if (!data) return;
+
+            html += '<div class="metric-card">';
+            html += '<div class="metric-label">' + config.replace('_', ' ') + ' — Pass Rate</div>';
+            html += '<div class="metric-value">' + Math.round(data.overall_pass_rate * 100) + '%</div>';
+            html += '</div>';
+
+            html += '<div class="metric-card">';
+            html += '<div class="metric-label">' + config.replace('_', ' ') + ' — Avg Tokens</div>';
+            html += '<div class="metric-value">' + Math.round(data.total_tokens_mean) + '</div>';
+            if (data.total_tokens_stddev) html += '<div class="metric-sub">&plusmn; ' + Math.round(data.total_tokens_stddev) + '</div>';
+            html += '</div>';
+
+            html += '<div class="metric-card">';
+            html += '<div class="metric-label">' + config.replace('_', ' ') + ' — Avg Time</div>';
+            html += '<div class="metric-value">' + Math.round(data.duration_ms_mean / 1000 * 10) / 10 + 's</div>';
+            if (data.duration_ms_stddev) html += '<div class="metric-sub">&plusmn; ' + Math.round(data.duration_ms_stddev) + 'ms</div>';
+            html += '</div>';
+        }});
+
+        if (benchmark.comparison) {{
+            const c = benchmark.comparison;
+            html += '<div class="metric-card">';
+            html += '<div class="metric-label">Pass Rate Delta</div>';
+            html += '<div class="metric-value" style="color:' + (c.pass_rate_delta >= 0 ? '#3fb950' : '#f85149') + '">';
+            html += (c.pass_rate_delta >= 0 ? '+' : '') + Math.round(c.pass_rate_delta * 100) + '%';
+            html += '</div></div>';
+
+            html += '<div class="metric-card">';
+            html += '<div class="metric-label">Token Overhead</div>';
+            html += '<div class="metric-value">' + Math.round(c.token_overhead_percent) + '%</div>';
+            html += '</div>';
+        }}
+
+        html += '</div>';
+        container.innerHTML = html;
+    }}
+
+    function escapeHtml(text) {{
+        const div = document.createElement('div');
+        div.textContent = text;
+        return div.innerHTML;
+    }}
+
+    function submitFeedback() {{
+        const feedback = {{
+            timestamp: new Date().toISOString(),
+            iteration: {iteration_num},
+            eval_feedback: [],
+            action: document.getElementById('action-select').value
+        }};
+
+        document.querySelectorAll('.eval-rating').forEach(select => {{
+            const evalId = select.dataset.eval;
+            const comment = document.querySelector('.eval-comment[data-eval="' + evalId + '"]').value;
+            feedback.eval_feedback.push({{
+                eval_id: evalId,
+                rating: select.value,
+                comment: comment
+            }});
+        }});
+
+        // Save as downloadable file
+        const blob = new Blob([JSON.stringify(feedback, null, 2)], {{ type: 'application/json' }});
+        const a = document.createElement('a');
+        a.href = URL.createObjectURL(blob);
+        a.download = 'feedback.json';
+        a.click();
+        alert('Feedback saved! Place feedback.json in the workspace directory.');
+    }}
+
+    renderOutputs();
+    renderBenchmark();
+    </script>
+</body>
+</html>"""
+    return html
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Generate HTML eval review page")
+    parser.add_argument("workspace", help="Path to skill workspace directory")
+    parser.add_argument("--static", help="Output path for standalone HTML file")
+    parser.add_argument("--open", action="store_true", help="Open in browser")
+    parser.add_argument("--iteration", type=int, help="Specific iteration to show")
+    args = parser.parse_args()
+
+    workspace = Path(args.workspace)
+    if not workspace.exists():
+        print(f"Error: {workspace} not found", file=sys.stderr)
+        sys.exit(1)
+
+    # Find iteration
+    if args.iteration:
+        iteration_num = args.iteration
+    else:
+        iteration_num = find_latest_iteration(workspace)
+
+    if iteration_num == 0:
+        print("Error: No iterations found in workspace", file=sys.stderr)
+        sys.exit(1)
+
+    iteration_dir = workspace / f"iteration-{iteration_num}"
+    if not iteration_dir.exists():
+        print(f"Error: {iteration_dir} not found", file=sys.stderr)
+        sys.exit(1)
+
+    # Collect data
+    evals, benchmark = collect_eval_data(iteration_dir)
+    if not evals:
+        print("Error: No eval data found", file=sys.stderr)
+        sys.exit(1)
+
+    # Generate HTML
+    html = generate_html(evals, benchmark, iteration_num, str(workspace))
+
+    # Output
+    output_path = args.static or str(workspace / "review.html")
+    Path(output_path).write_text(html)
+    print(f"Generated: {output_path}")
+
+    if args.open:
+        webbrowser.open(f"file://{os.path.abspath(output_path)}")
+
+
+if __name__ == "__main__":
+    main()

package/skills/myai-skill-builder/scripts/package_skill.py

@@ -0,0 +1,91 @@
+#!/usr/bin/env python3
+"""Package a skill directory into a distributable archive.
+
+Creates a .tar.gz archive of a skill directory, excluding workspace
+and temporary files.
+
+Usage:
+    python3 package_skill.py .claude/skills/my-skill
+    python3 package_skill.py .claude/skills/my-skill --output my-skill-v1.tar.gz
+"""
+
+import argparse
+import os
+import sys
+import tarfile
+from pathlib import Path
+
+# Directories and patterns to exclude from the archive
+EXCLUDE_PATTERNS = {
+    "*-workspace",
+    "__pycache__",
+    "*.pyc",
+    ".DS_Store",
+    "description_eval.json",
+    "feedback.json",
+    "review.html",
+}
+
+
+def should_exclude(path):
+    """Check if a path should be excluded from the archive."""
+    name = Path(path).name
+    for pattern in EXCLUDE_PATTERNS:
+        if pattern.startswith("*") and name.endswith(pattern[1:]):
+            return True
+        if pattern.endswith("*") and name.startswith(pattern[:-1]):
+            return True
+        if name == pattern:
+            return True
+    return False
+
+
+def package_skill(skill_dir, output_path=None):
+    """Create a .tar.gz archive of the skill directory."""
+    skill_dir = Path(skill_dir).resolve()
+    if not skill_dir.exists():
+        print(f"Error: {skill_dir} not found", file=sys.stderr)
+        sys.exit(1)
+
+    skill_md = skill_dir / "SKILL.md"
+    if not skill_md.exists():
+        print(f"Error: No SKILL.md found in {skill_dir}", file=sys.stderr)
+        sys.exit(1)
+
+    slug = skill_dir.name
+    if output_path is None:
+        output_path = skill_dir.parent / f"{slug}.tar.gz"
+    else:
+        output_path = Path(output_path)
+
+    file_count = 0
+    with tarfile.open(output_path, "w:gz") as tar:
+        for root, dirs, files in os.walk(skill_dir):
+            # Filter out excluded directories
+            dirs[:] = [d for d in dirs if not should_exclude(d)]
+
+            for f in files:
+                filepath = Path(root) / f
+                if should_exclude(filepath):
+                    continue
+
+                arcname = str(filepath.relative_to(skill_dir.parent))
+                tar.add(filepath, arcname=arcname)
+                file_count += 1
+
+    size_kb = output_path.stat().st_size / 1024
+    print(f"Packaged {file_count} files into {output_path} ({size_kb:.1f} KB)")
+    return str(output_path)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Package a skill for distribution")
+    parser.add_argument("skill_dir", help="Path to skill directory")
+    parser.add_argument("--output", "-o", help="Output archive path")
+    args = parser.parse_args()
+
+    package_skill(args.skill_dir, args.output)
+
+
+if __name__ == "__main__":
+    main()