adversarial-workflow 0.7.0__py3-none-any.whl → 0.9.0__py3-none-any.whl

adversarial_workflow/library/commands.py

@@ -0,0 +1,849 @@
+"""CLI commands for the evaluator library."""
+
+import difflib
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import List, Optional
+
+import yaml
+
+from .client import LibraryClient, LibraryClientError, NetworkError, ParseError
+from .models import IndexData, InstalledEvaluatorMeta, UpdateInfo
+
+# ANSI color codes (matching cli.py)
+RESET = "\033[0m"
+BOLD = "\033[1m"
+GREEN = "\033[92m"
+YELLOW = "\033[93m"
+RED = "\033[91m"
+CYAN = "\033[96m"
+GRAY = "\033[90m"
+
+
+def get_evaluators_dir() -> Path:
+    """Get the evaluators directory for the current project."""
+    return Path.cwd() / ".adversarial" / "evaluators"
+
+
+def format_table(
+    headers: List[str], rows: List[List[str]], widths: Optional[List[int]] = None
+) -> str:
+    """
+    Format data as a simple table.
+
+    Args:
+        headers: Column headers.
+        rows: List of row data (each row is a list of strings).
+        widths: Optional column widths. If None, auto-calculated.
+
+    Returns:
+        Formatted table string.
+    """
+    if not widths:
+        # Calculate widths from data
+        widths = [len(h) for h in headers]
+        for row in rows:
+            for i, cell in enumerate(row):
+                if i < len(widths):
+                    widths[i] = max(widths[i], len(str(cell)))
+
+    # Format header
+    header_line = "  ".join(h.ljust(w) for h, w in zip(headers, widths))
+    lines = [header_line]
+
+    # Format rows
+    for row in rows:
+        row_line = "  ".join(str(c).ljust(w) for c, w in zip(row, widths))
+        lines.append(row_line)
+
+    return "\n".join(lines)
+
+
+def generate_provenance_header(provider: str, name: str, version: str) -> str:
+    """Generate the provenance header for installed evaluators."""
+    timestamp = datetime.now(timezone.utc).isoformat(timespec="seconds").replace("+00:00", "Z")
+    return f"""# Installed from adversarial-evaluator-library
+# Source: {provider}/{name}
+# Version: {version}
+# Installed: {timestamp}
+#
+# To check for updates: adversarial library check-updates
+# To update: adversarial library update {name}
+#
+# Feel free to edit this file - it's yours now!
+
+_meta:
+  source: adversarial-evaluator-library
+  source_path: {provider}/{name}
+  version: "{version}"
+  installed: "{timestamp}"
+
+"""
+
+
+def scan_installed_evaluators() -> List[InstalledEvaluatorMeta]:
+    """
+    Scan the evaluators directory for installed library evaluators.
+
+    Returns:
+        List of metadata for installed evaluators with _meta blocks.
+    """
+    evaluators_dir = get_evaluators_dir()
+    if not evaluators_dir.exists():
+        return []
+
+    installed = []
+    for yaml_file in evaluators_dir.glob("*.yml"):
+        try:
+            with open(yaml_file, "r", encoding="utf-8") as f:
+                data = yaml.safe_load(f)
+
+            if data and "_meta" in data:
+                meta = InstalledEvaluatorMeta.from_dict(data["_meta"])
+                if meta and meta.source == "adversarial-evaluator-library":
+                    meta.file_path = str(yaml_file)  # Track file path for updates
+                    installed.append(meta)
+        except (yaml.YAMLError, OSError):
+            # Skip files that can't be parsed
+            continue
+
+    return installed
+
+
+def library_list(
+    provider: Optional[str] = None,
+    category: Optional[str] = None,
+    verbose: bool = False,
+    no_cache: bool = False,
+) -> int:
+    """
+    List available evaluators from the library.
+
+    Args:
+        provider: Filter by provider name.
+        category: Filter by category name.
+        verbose: Show detailed information.
+        no_cache: Bypass cache and fetch fresh data.
+
+    Returns:
+        Exit code (0 for success, 1 for error).
+    """
+    client = LibraryClient()
+
+    try:
+        index, from_cache = client.fetch_index(no_cache=no_cache)
+    except NetworkError as e:
+        print(f"{RED}Error: Network unavailable{RESET}")
+        print(f"  {e}")
+        print()
+        print("Check your internet connection and try again.")
+        print(f"Or use {CYAN}--no-cache{RESET} to force a fresh fetch.")
+        return 1
+    except ParseError as e:
+        print(f"{RED}Error: Could not parse library index{RESET}")
+        print(f"  {e}")
+        return 1
+
+    # Filter evaluators
+    evaluators = index.evaluators
+
+    if provider:
+        evaluators = [e for e in evaluators if e.provider == provider]
+        if not evaluators:
+            print(f"{YELLOW}No evaluators found for provider: {provider}{RESET}")
+            print()
+            print("Available providers:")
+            providers = sorted(set(e.provider for e in index.evaluators))
+            for p in providers:
+                print(f"  - {p}")
+            return 1
+
+    if category:
+        evaluators = [e for e in evaluators if e.category == category]
+        if not evaluators:
+            print(f"{YELLOW}No evaluators found for category: {category}{RESET}")
+            print()
+            print("Available categories:")
+            for cat_name, cat_desc in sorted(index.categories.items()):
+                print(f"  - {cat_name}: {cat_desc}")
+            return 1
+
+    # Print header
+    cache_note = f" {GRAY}(cached){RESET}" if from_cache else ""
+    print()
+    print(
+        f"{BOLD}Available evaluators from adversarial-evaluator-library (v{index.version}){RESET}{cache_note}"
+    )
+    print()
+
+    if verbose:
+        # Detailed view
+        for e in evaluators:
+            print(f"{CYAN}{e.provider}/{e.name}{RESET}")
+            print(f"  Model: {e.model}")
+            print(f"  Category: {e.category}")
+            print(f"  Description: {e.description}")
+            print()
+    else:
+        # Table view
+        headers = ["PROVIDER", "NAME", "CATEGORY", "DESCRIPTION"]
+        rows = []
+        for e in evaluators:
+            # Truncate description if too long
+            desc = e.description
+            if len(desc) > 40:
+                desc = desc[:37] + "..."
+            rows.append([e.provider, e.name, e.category, desc])
+
+        print(format_table(headers, rows))
+        print()
+
+    # Summary
+    count = len(evaluators)
+    total = len(index.evaluators)
+    if provider or category:
+        print(f"{count} evaluators shown (of {total} total).")
+    else:
+        print(f"{count} evaluators available.")
+    print()
+    print(f"Use '{CYAN}adversarial library install <provider>/<name>{RESET}' to install.")
+
+    return 0
+
+
+def library_info(evaluator_spec: str) -> int:
+    """
+    Show detailed information about a library evaluator.
+
+    Args:
+        evaluator_spec: Evaluator in 'provider/name' format.
+
+    Returns:
+        Exit code (0 for success, 1 for error).
+    """
+    client = LibraryClient()
+
+    # Parse spec
+    parts = evaluator_spec.split("/")
+    if len(parts) != 2:
+        print(f"{RED}Error: Invalid format. Use provider/name (e.g., google/gemini-flash){RESET}")
+        return 1
+
+    provider, name = parts
+
+    # Fetch index
+    try:
+        index, _ = client.fetch_index()
+    except NetworkError as e:
+        print(f"{RED}Error: Network unavailable{RESET}")
+        print(f"  {e}")
+        return 1
+    except ParseError as e:
+        print(f"{RED}Error: Could not parse library index{RESET}")
+        print(f"  {e}")
+        return 1
+
+    # Get evaluator entry
+    entry = index.get_evaluator(provider, name)
+    if not entry:
+        print(f"{RED}Error: Evaluator not found: {evaluator_spec}{RESET}")
+        print()
+        print("Use 'adversarial library list' to see available evaluators.")
+        return 1
+
+    # Display basic info from index
+    print()
+    print(f"{BOLD}{provider}/{name}{RESET}")
+    print()
+    print(f"Version:     {index.version}")
+    print(f"Provider:    {provider}")
+    print(f"Model:       {entry.model}")
+    print(f"Category:    {entry.category}")
+    print()
+    print("Description:")
+    print(f"  {entry.description}")
+    print()
+
+    # Try to fetch extended info from README
+    readme = client.fetch_readme(provider, name)
+    if readme:
+        _display_extended_info(readme)
+    else:
+        print(f"{GRAY}Extended info unavailable (README.md not found).{RESET}")
+        print()
+
+    # Installation hint
+    print(f"Install: {CYAN}adversarial library install {evaluator_spec}{RESET}")
+    print()
+
+    return 0
+
+
+def _display_extended_info(readme: str) -> None:
+    """
+    Parse and display extended info from README.md.
+
+    Extracts changelog, cost estimates, and API key info from README.
+    """
+    lines = readme.split("\n")
+
+    # Look for API Key section
+    api_key = None
+    for line in lines:
+        if "api_key" in line.lower() or "API key" in line:
+            # Try to extract env var name
+            import re
+
+            match = re.search(r"([A-Z_]+_API_KEY)", line)
+            if match:
+                api_key = match.group(1)
+                break
+
+    if api_key:
+        print(f"API Key:     {api_key}")
+        print()
+
+    # Look for Changelog section
+    in_changelog = False
+    changelog_lines = []
+    for line in lines:
+        if line.strip().lower().startswith("## changelog") or line.strip().lower().startswith(
+            "# changelog"
+        ):
+            in_changelog = True
+            continue
+        if in_changelog:
+            if line.strip().startswith("##") or line.strip().startswith("# "):
+                break
+            if line.strip():
+                changelog_lines.append(line.strip())
+            if len(changelog_lines) >= 5:  # Limit changelog entries
+                break
+
+    if changelog_lines:
+        print("Changelog:")
+        for cl in changelog_lines[:5]:
+            print(f"  {cl}")
+        print()
+
+    # Look for Cost section
+    in_cost = False
+    cost_lines = []
+    for line in lines:
+        if "cost" in line.lower() and line.strip().startswith("#"):
+            in_cost = True
+            continue
+        if in_cost:
+            if line.strip().startswith("#"):
+                break
+            if line.strip():
+                cost_lines.append(line.strip())
+            if len(cost_lines) >= 3:
+                break
+
+    if cost_lines:
+        print("Estimated Cost:")
+        for cl in cost_lines[:3]:
+            print(f"  {cl}")
+        print()
+
+
+def library_install(
+    evaluator_specs: List[str],
+    force: bool = False,
+    skip_validation: bool = False,
+    dry_run: bool = False,
+    category: Optional[str] = None,
+    yes: bool = False,
+) -> int:
+    """
+    Install evaluators from the library.
+
+    Args:
+        evaluator_specs: List of evaluator specs in 'provider/name' format.
+        force: Overwrite existing files.
+        skip_validation: Skip schema validation.
+        dry_run: Preview without making changes.
+        category: Install all evaluators in this category.
+        yes: Skip confirmation prompts (required for non-TTY).
+
+    Returns:
+        Exit code (0 for success, 1 for error).
+    """
+    client = LibraryClient()
+
+    # Non-TTY detection: require --yes for non-interactive mode (unless dry-run)
+    if not yes and not dry_run and not sys.stdin.isatty():
+        print(f"{RED}Error: Use --yes for non-interactive mode{RESET}")
+        return 1
+
+    # Fetch index first
+    try:
+        index, _ = client.fetch_index()
+    except NetworkError as e:
+        print(f"{RED}Error: Network unavailable{RESET}")
+        print(f"  {e}")
+        return 1
+    except ParseError as e:
+        print(f"{RED}Error: Could not parse library index{RESET}")
+        print(f"  {e}")
+        return 1
+
+    # Handle --category flag: get all evaluators in that category
+    if category:
+        matching = index.filter_by_category(category)
+        if not matching:
+            print(f"{RED}Error: No evaluators found in category '{category}'{RESET}")
+            print()
+            print("Available categories:")
+            for cat_name, cat_desc in sorted(index.categories.items()):
+                print(f"  - {cat_name}: {cat_desc}")
+            return 1
+
+        print(f"Installing all evaluators in category '{category}':")
+        print()
+        for e in matching:
+            print(f"  - {e.provider}/{e.name} (v{index.version})")
+        print()
+
+        # Skip confirmation for --yes or --dry-run (dry-run makes no changes)
+        if not yes and not dry_run:
+            response = input("Proceed? [y/N]: ").strip().lower()
+            if response not in ("y", "yes"):
+                print("Cancelled.")
+                return 0
+
+        evaluator_specs = [f"{e.provider}/{e.name}" for e in matching]
+
+    # Require at least one evaluator spec
+    if not evaluator_specs:
+        print(f"{RED}Error: No evaluators specified{RESET}")
+        print()
+        print("Usage:")
+        print("  adversarial library install <provider>/<name> [<provider>/<name> ...]")
+        print("  adversarial library install --category <category-name>")
+        return 1
+
+    evaluators_dir = get_evaluators_dir()
+    if not dry_run:
+        evaluators_dir.mkdir(parents=True, exist_ok=True)
+
+    success_count = 0
+    for spec in evaluator_specs:
+        # Parse spec (provider/name or provider/name@version - version ignored for now)
+        if "@" in spec:
+            spec = spec.split("@")[0]  # Strip version for now
+
+        parts = spec.split("/")
+        if len(parts) != 2:
+            print(f"{RED}Error: Invalid evaluator spec: {spec}{RESET}")
+            print("  Expected format: provider/name (e.g., google/gemini-flash)")
+            continue
+
+        provider, name = parts
+
+        # Check if evaluator exists in index
+        entry = index.get_evaluator(provider, name)
+        if not entry:
+            print(f"{RED}Error: Evaluator not found: {spec}{RESET}")
+            print("  Use 'adversarial library list' to see available evaluators.")
+            continue
+
+        # Check if file already exists (use provider-name format to avoid collisions)
+        dest_path = evaluators_dir / f"{provider}-{name}.yml"
+
+        if dry_run:
+            # Dry-run mode: show preview without making changes
+            print(f"Dry run: Would install {CYAN}{spec}{RESET} (v{index.version})")
+            print()
+            print(f"  Target: {dest_path}")
+            if dest_path.exists():
+                print(f"  Status: {YELLOW}File exists (would overwrite with --force){RESET}")
+            else:
+                print(f"  Status: {GREEN}New file (clean install){RESET}")
+            print()
+
+            # Fetch and preview evaluator config
+            preview_success = False
+            try:
+                yaml_content = client.fetch_evaluator(provider, name)
+                yaml_content_clean = yaml_content.lstrip()
+                if yaml_content_clean.startswith("---"):
+                    yaml_content_clean = yaml_content_clean[3:].lstrip("\n")
+
+                print("  Evaluator config preview:")
+                print("  " + "─" * 25)
+                preview_lines = yaml_content_clean.split("\n")[:10]
+                for line in preview_lines:
+                    print(f"  {line}")
+                if len(yaml_content_clean.split("\n")) > 10:
+                    print("  ...")
+                print()
+                preview_success = True
+            except NetworkError as e:
+                print(f"  {RED}Error: Could not fetch preview{RESET}")
+                print(f"    {e}")
+                print()
+
+            print(f"{YELLOW}No changes made (dry run).{RESET}")
+            print()
+            if preview_success:
+                success_count += 1
+            continue
+
+        if dest_path.exists() and not force:
+            print(f"{YELLOW}Skipping: {provider}-{name}.yml already exists{RESET}")
+            print(f"  Use {CYAN}--force{RESET} to overwrite.")
+            continue
+
+        # Fetch evaluator config
+        print(f"Installing {CYAN}{spec}{RESET}...")
+        try:
+            yaml_content = client.fetch_evaluator(provider, name)
+        except NetworkError as e:
+            print(f"  {RED}Error: Failed to fetch evaluator{RESET}")
+            print(f"    {e}")
+            continue
+
+        # Strip leading YAML document separator to prevent multi-document issues
+        yaml_content_clean = yaml_content.lstrip()
+        if yaml_content_clean.startswith("---"):
+            # Remove the document separator and any following newline
+            yaml_content_clean = yaml_content_clean[3:].lstrip("\n")
+
+        # Validate YAML
+        try:
+            parsed = yaml.safe_load(yaml_content_clean)
+            if not parsed:
+                raise ValueError("Empty YAML content")
+        except (yaml.YAMLError, ValueError) as e:
+            print(f"  {RED}Error: Invalid YAML in evaluator{RESET}")
+            print(f"    {e}")
+            if skip_validation:
+                print(f"  {YELLOW}Warning: --skip-validation is set, continuing anyway{RESET}")
+            else:
+                continue
+
+        # Add provenance header
+        full_content = (
+            generate_provenance_header(provider, name, index.version) + yaml_content_clean
+        )
+
+        # Write file
+        try:
+            with open(dest_path, "w", encoding="utf-8") as f:
+                f.write(full_content)
+            print(f"  {GREEN}Installed: {dest_path}{RESET}")
+            success_count += 1
+        except OSError as e:
+            print(f"  {RED}Error: Could not write file{RESET}")
+            print(f"    {e}")
+            continue
+
+    # Summary
+    print()
+    if dry_run:
+        if success_count == 0 and len(evaluator_specs) > 0:
+            print(f"{RED}Dry run failed: No evaluators could be previewed.{RESET}")
+            return 1
+        print(
+            f"{CYAN}Dry run complete. {success_count} evaluator(s) previewed successfully.{RESET}"
+        )
+        return 0
+    elif success_count == len(evaluator_specs):
+        print(f"{GREEN}All {success_count} evaluator(s) installed successfully.{RESET}")
+    elif success_count > 0:
+        print(f"{YELLOW}{success_count} of {len(evaluator_specs)} evaluator(s) installed.{RESET}")
+    else:
+        print(f"{RED}No evaluators installed.{RESET}")
+        return 1
+
+    if not dry_run:
+        print()
+        print("Next steps:")
+        print(f"  1. Configure API keys if needed (check {CYAN}.env{RESET})")
+        print(f"  2. Run: {CYAN}adversarial <evaluator-name> <task-file>{RESET}")
+
+    return 0
+
+
+def library_check_updates(name: Optional[str] = None, no_cache: bool = False) -> int:
+    """
+    Check for available updates to installed evaluators.
+
+    Args:
+        name: Specific evaluator name to check (optional).
+        no_cache: Bypass cache and fetch fresh data.
+
+    Returns:
+        Exit code (0 for success, 1 for error).
+    """
+    client = LibraryClient()
+
+    # Scan installed evaluators
+    installed = scan_installed_evaluators()
+    if not installed:
+        print(f"{YELLOW}No library-installed evaluators found.{RESET}")
+        print()
+        print("Install evaluators with: adversarial library install <provider>/<name>")
+        return 0
+
+    # Filter by name if specified
+    if name:
+        installed = [m for m in installed if m.name == name]
+        if not installed:
+            print(f"{YELLOW}Evaluator '{name}' not found or not from library.{RESET}")
+            return 1
+
+    # Fetch index
+    print("Checking for evaluator updates...")
+    print()
+
+    try:
+        index, from_cache = client.fetch_index(no_cache=no_cache)
+    except NetworkError as e:
+        print(f"{RED}Error: Network unavailable{RESET}")
+        print(f"  {e}")
+        return 1
+    except ParseError as e:
+        print(f"{RED}Error: Could not parse library index{RESET}")
+        print(f"  {e}")
+        return 1
+
+    # Compare versions
+    updates: List[UpdateInfo] = []
+    for meta in installed:
+        entry = index.get_evaluator(meta.provider, meta.name)
+        if entry:
+            is_outdated = meta.version != index.version
+            updates.append(
+                UpdateInfo(
+                    name=meta.name,
+                    installed_version=meta.version,
+                    available_version=index.version,
+                    is_outdated=is_outdated,
+                )
+            )
+        else:
+            # Evaluator no longer in index
+            updates.append(
+                UpdateInfo(
+                    name=meta.name,
+                    installed_version=meta.version,
+                    available_version="-",
+                    is_outdated=False,
+                    is_local_only=True,
+                )
+            )
+
+    # Print table
+    headers = ["EVALUATOR", "INSTALLED", "AVAILABLE", "STATUS"]
+    rows = []
+    for u in updates:
+        status = u.status
+        if u.is_outdated:
+            status = f"{YELLOW}{status}{RESET}"
+        elif u.is_local_only:
+            status = f"{GRAY}{status}{RESET}"
+        else:
+            status = f"{GREEN}{status}{RESET}"
+        rows.append([u.name, u.installed_version, u.available_version, status])
+
+    print(format_table(headers, rows))
+    print()
+
+    # Summary
+    outdated_count = sum(1 for u in updates if u.is_outdated)
+    if outdated_count > 0:
+        print(f"{YELLOW}{outdated_count} update(s) available.{RESET}")
+        print()
+        print(f"Run '{CYAN}adversarial library update <name>{RESET}' to update.")
+        print(f"Or '{CYAN}adversarial library update --all{RESET}' to update all.")
+    else:
+        print(f"{GREEN}All evaluators are up to date.{RESET}")
+
+    return 0
+
+
+def library_update(
+    name: Optional[str] = None,
+    all_evaluators: bool = False,
+    yes: bool = False,
+    diff_only: bool = False,
+    no_cache: bool = False,
+    dry_run: bool = False,
+) -> int:
+    """
+    Update installed evaluators to newer versions.
+
+    Args:
+        name: Specific evaluator name to update.
+        all_evaluators: Update all outdated evaluators.
+        yes: Skip confirmation prompts.
+        diff_only: Show diff without applying changes (same as dry_run).
+        no_cache: Bypass cache.
+        dry_run: Preview without making changes (same as diff_only).
+
+    Returns:
+        Exit code (0 for success, 1 for error).
+    """
+    # Combine dry_run and diff_only (they do the same thing)
+    preview_only = dry_run or diff_only
+
+    client = LibraryClient()
+
+    # Non-TTY detection: require --yes for non-interactive mode (unless preview mode)
+    if not yes and not preview_only and not sys.stdin.isatty():
+        print(f"{RED}Error: Use --yes for non-interactive mode{RESET}")
+        return 1
+
+    if not name and not all_evaluators:
+        print(f"{RED}Error: Specify an evaluator name or use --all{RESET}")
+        print()
+        print("Usage:")
+        print("  adversarial library update <name>")
+        print("  adversarial library update --all")
+        return 1
+
+    # Scan installed evaluators
+    installed = scan_installed_evaluators()
+    if not installed:
+        print(f"{YELLOW}No library-installed evaluators found.{RESET}")
+        return 0
+
+    # Fetch index
+    try:
+        index, _ = client.fetch_index(no_cache=no_cache)
+    except NetworkError as e:
+        print(f"{RED}Error: Network unavailable{RESET}")
+        print(f"  {e}")
+        return 1
+    except ParseError as e:
+        print(f"{RED}Error: Could not parse library index{RESET}")
+        print(f"  {e}")
+        return 1
+
+    # Find evaluators to update
+    to_update = []
+    for meta in installed:
+        if name and meta.name != name:
+            continue
+
+        entry = index.get_evaluator(meta.provider, meta.name)
+        if not entry:
+            if name:
+                print(f"{YELLOW}Evaluator '{name}' not found in library.{RESET}")
+            continue
+
+        if meta.version != index.version:
+            to_update.append((meta, entry))
+        elif name:
+            print(f"{GREEN}Evaluator '{name}' is already up to date (v{meta.version}).{RESET}")
+            return 0
+
+    if not to_update:
+        if all_evaluators:
+            print(f"{GREEN}All evaluators are up to date.{RESET}")
+        return 0
+
+    evaluators_dir = get_evaluators_dir()
+    updated_count = 0
+
+    for meta, entry in to_update:
+        print()
+        print(f"Updating {CYAN}{meta.name}{RESET} ({meta.version} → {index.version})...")
+
+        # Fetch new content
+        try:
+            new_yaml = client.fetch_evaluator(entry.provider, entry.name)
+        except NetworkError as e:
+            print(f"  {RED}Error: Failed to fetch evaluator{RESET}")
+            print(f"    {e}")
+            continue
+
+        # Read current content using tracked file path
+        if meta.file_path:
+            current_path = Path(meta.file_path)
+        else:
+            # Fallback: check both old and new naming conventions
+            new_path = evaluators_dir / f"{meta.provider}-{meta.name}.yml"
+            old_path = evaluators_dir / f"{meta.name}.yml"
+            current_path = new_path if new_path.exists() else old_path
+        try:
+            with open(current_path, "r", encoding="utf-8") as f:
+                current_content = f.read()
+        except OSError as e:
+            print(f"  {RED}Error: Could not read current file{RESET}")
+            print(f"    {e}")
+            continue
+
+        # Generate new content with updated provenance
+        # Strip leading YAML document separator to prevent multi-document issues
+        new_yaml_clean = new_yaml.lstrip()
+        if new_yaml_clean.startswith("---"):
+            new_yaml_clean = new_yaml_clean[3:].lstrip("\n")
+        new_content = (
+            generate_provenance_header(entry.provider, entry.name, index.version) + new_yaml_clean
+        )
+
+        # Show diff
+        print()
+        diff = list(
+            difflib.unified_diff(
+                current_content.splitlines(keepends=True),
+                new_content.splitlines(keepends=True),
+                fromfile=f"{meta.name}.yml (current)",
+                tofile=f"{meta.name}.yml (new)",
+            )
+        )
+
+        if not diff:
+            print(f"  {GRAY}No changes in evaluator content.{RESET}")
+            continue
+
+        print("  Changes:")
+        for line in diff[:50]:  # Limit diff output
+            if line.startswith("+") and not line.startswith("+++"):
+                print(f"  {GREEN}{line.rstrip()}{RESET}")
+            elif line.startswith("-") and not line.startswith("---"):
+                print(f"  {RED}{line.rstrip()}{RESET}")
+            else:
+                print(f"  {line.rstrip()}")
+
+        if len(diff) > 50:
+            print(f"  {GRAY}... ({len(diff) - 50} more lines){RESET}")
+
+        if preview_only:
+            print()
+            print(f"  {GRAY}(dry run mode, no changes applied){RESET}")
+            continue
+
+        # Confirm update
+        if not yes:
+            print()
+            response = input(f"  Apply update? [y/N]: ").strip().lower()
+            if response not in ("y", "yes"):
+                print(f"  {GRAY}Skipped.{RESET}")
+                continue
+
+        # Apply update
+        try:
+            with open(current_path, "w", encoding="utf-8") as f:
+                f.write(new_content)
+            print(f"  {GREEN}Updated!{RESET}")
+            updated_count += 1
+        except OSError as e:
+            print(f"  {RED}Error: Could not write file{RESET}")
+            print(f"    {e}")
+
+    # Summary
+    print()
+    if preview_only:
+        print(f"Dry run complete. Use without {CYAN}--dry-run{RESET} to apply changes.")
+    elif updated_count > 0:
+        print(f"{GREEN}{updated_count} evaluator(s) updated.{RESET}")
+    else:
+        print(f"{YELLOW}No evaluators were updated.{RESET}")
+
+    return 0