feat(wiki): add multimodal knowledge pipeline script

codenib/wiki/__init__.py

@@ -24,13 +24,15 @@
     get_visual_evidence,
     search_visual_context,
 )
+from .media_pipeline import build_multimodal_repository_knowledge
 from .media_vlm import OpenAICompatibleVisualFactExtractor
 
 __all__ = [
     "WikiBuilder",
     "OpenAICompatibleVisualFactExtractor",
     "build_media_evidence_pack",
     "build_multimodal_knowledge_view",
+    "build_multimodal_repository_knowledge",
     "build_visual_facts_manifest",
     "deterministic_visual_facts",
     "discover_media_manifest",

codenib/wiki/media_pipeline.py

@@ -0,0 +1,72 @@
+# SPDX-FileCopyrightText: 2025-2026 CodeNib Contributors
+#
+# SPDX-License-Identifier: Apache-2.0
+
+"""End-to-end construction for multimodal repository knowledge."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any, Mapping
+
+from ..repository_source_selection import (
+    DEFAULT_REPOSITORY_SOURCE_SELECTION,
+    RepositorySourceSelection,
+)
+from .media_artifacts import discover_media_manifest
+from .media_facts import VisualFactExtractor, build_visual_facts_manifest
+from .media_grounding import (
+    discover_source_symbol_candidates,
+    ground_visual_facts_to_sources,
+)
+from .media_knowledge import build_multimodal_knowledge_view
+
+
+def build_multimodal_repository_knowledge(
+    repo_path: str | Path,
+    *,
+    commit: str | None = None,
+    exclude_roots: tuple[str | Path, ...] = (),
+    selection: RepositorySourceSelection = DEFAULT_REPOSITORY_SOURCE_SELECTION,
+    extractor: VisualFactExtractor | None = None,
+    max_artifacts: int = 4096,
+    max_source_candidates: int = 8192,
+) -> dict[str, Any]:
+    """Build the deterministic multimodal repository knowledge bundle."""
+
+    media_manifest = discover_media_manifest(
+        repo_path,
+        commit=commit,
+        exclude_roots=exclude_roots,
+        selection=selection,
+        max_artifacts=max_artifacts,
+    )
+    facts_kwargs: dict[str, Any] = {}
+    if extractor is not None:
+        facts_kwargs["extractor"] = extractor
+    visual_facts_manifest = build_visual_facts_manifest(media_manifest, **facts_kwargs)
+    source_candidates = discover_source_symbol_candidates(
+        repo_path,
+        exclude_roots=exclude_roots,
+        selection=selection,
+        max_candidates=max_source_candidates,
+    )
+    grounding_manifest = ground_visual_facts_to_sources(
+        visual_facts_manifest,
+        source_candidates,
+    )
+    knowledge_view = build_multimodal_knowledge_view(
+        media_manifest,
+        visual_facts_manifest,
+        grounding_manifest,
+    )
+    return {
+        "media_manifest": media_manifest,
+        "visual_facts_manifest": visual_facts_manifest,
+        "source_candidate_count": len(source_candidates),
+        "grounding_manifest": grounding_manifest,
+        "knowledge_view": knowledge_view,
+    }
+
+
+__all__ = ["build_multimodal_repository_knowledge"]

docs/experiments/multimodal_repository_knowledge.md

@@ -105,6 +105,22 @@ print(view["entry_count"])
 This creates a deterministic local view. A VLM extractor can be added later by
 passing a custom extractor into `build_visual_facts_manifest()`.
 
+For callers that want the full deterministic pipeline in one step:
+
+```python
+from codenib.wiki import build_multimodal_repository_knowledge
+
+bundle = build_multimodal_repository_knowledge(repo)
+view = bundle["knowledge_view"]
+```
+
+The same deterministic bundle can be written from the command line:
+
+```text
+python scripts/build_multimodal_knowledge.py /path/to/repository \
+  --output /tmp/multimodal-knowledge.json
+```
+
 ```python
 from codenib.wiki import OpenAICompatibleVisualFactExtractor
 

scripts/build_multimodal_knowledge.py

@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+# SPDX-FileCopyrightText: 2025-2026 CodeNib Contributors
+#
+# SPDX-License-Identifier: Apache-2.0
+
+"""Build a deterministic multimodal repository knowledge bundle."""
+
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+
+from codenib.wiki import build_multimodal_repository_knowledge
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("repo", help="Repository root to scan")
+    parser.add_argument(
+        "--output",
+        "-o",
+        required=True,
+        help="Path to write the multimodal knowledge bundle JSON",
+    )
+    parser.add_argument(
+        "--commit",
+        default=None,
+        help="Optional commit identity to record in the media manifest",
+    )
+    parser.add_argument(
+        "--max-artifacts",
+        type=int,
+        default=4096,
+        help="Maximum media artifacts to include",
+    )
+    parser.add_argument(
+        "--max-source-candidates",
+        type=int,
+        default=8192,
+        help="Maximum source-symbol candidates to consider for grounding",
+    )
+    return parser
+
+
+def main(argv: list[str] | None = None) -> int:
+    args = build_parser().parse_args(argv)
+    bundle = build_multimodal_repository_knowledge(
+        args.repo,
+        commit=args.commit,
+        max_artifacts=args.max_artifacts,
+        max_source_candidates=args.max_source_candidates,
+    )
+    output = Path(args.output)
+    output.parent.mkdir(parents=True, exist_ok=True)
+    output.write_text(
+        json.dumps(bundle, ensure_ascii=False, indent=2, sort_keys=True) + "\n",
+        encoding="utf-8",
+    )
+    counts = {
+        "media_artifacts": bundle["media_manifest"]["artifact_count"],
+        "visual_fact_packs": bundle["visual_facts_manifest"]["fact_count"],
+        "source_candidates": bundle["source_candidate_count"],
+        "visual_code_bindings": bundle["grounding_manifest"]["binding_count"],
+        "knowledge_entries": bundle["knowledge_view"]["entry_count"],
+    }
+    print(json.dumps(counts, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

test/scripts/test_build_multimodal_knowledge.py

@@ -0,0 +1,51 @@
+# SPDX-FileCopyrightText: 2025-2026 CodeNib Contributors
+#
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+
+
+def test_build_multimodal_knowledge_script_writes_bundle(tmp_path):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    (repo / "docs").mkdir()
+    (repo / "src").mkdir()
+    (repo / "docs" / "architecture.svg").write_text(
+        "<svg>WikiService</svg>",
+        encoding="utf-8",
+    )
+    (repo / "README.md").write_text(
+        "![WikiService architecture](docs/architecture.svg)",
+        encoding="utf-8",
+    )
+    (repo / "src" / "wiki.py").write_text(
+        "class WikiService: pass",
+        encoding="utf-8",
+    )
+    output = tmp_path / "bundle.json"
+
+    completed = subprocess.run(
+        [
+            sys.executable,
+            "scripts/build_multimodal_knowledge.py",
+            str(repo),
+            "--output",
+            str(output),
+            "--commit",
+            "abc123",
+        ],
+        check=True,
+        stdout=subprocess.PIPE,
+        text=True,
+    )
+
+    counts = json.loads(completed.stdout)
+    bundle = json.loads(output.read_text(encoding="utf-8"))
+    assert counts["media_artifacts"] == 1
+    assert counts["knowledge_entries"] == 1
+    assert bundle["media_manifest"]["commit"] == "abc123"
+    assert bundle["knowledge_view"]["entry_count"] == 1

test/wiki/test_media_pipeline.py

@@ -6,6 +6,7 @@
 
 from codenib.wiki import (
     build_multimodal_knowledge_view,
+    build_multimodal_repository_knowledge,
     build_visual_facts_manifest,
     discover_media_manifest,
     discover_source_symbol_candidates,
@@ -57,3 +58,24 @@ def test_multimodal_repository_knowledge_pipeline(tmp_path):
     assert view["entry_count"] == 1
     assert search_visual_context(view, "IndexCompiler architecture")
     assert find_visual_code_links(view, "src/compiler.py", symbol="IndexCompiler")
+
+
+def test_build_multimodal_repository_knowledge_wraps_pipeline(tmp_path):
+    (tmp_path / "docs").mkdir()
+    (tmp_path / "src").mkdir()
+    (tmp_path / "docs" / "diagram.png").write_bytes(b"png")
+    (tmp_path / "README.md").write_text(
+        "![WikiService diagram](docs/diagram.png)",
+        encoding="utf-8",
+    )
+    (tmp_path / "src" / "wiki.py").write_text(
+        "class WikiService: pass",
+        encoding="utf-8",
+    )
+
+    bundle = build_multimodal_repository_knowledge(tmp_path, commit="abc123")
+
+    assert bundle["media_manifest"]["artifact_count"] == 1
+    assert bundle["visual_facts_manifest"]["fact_count"] == 1
+    assert bundle["source_candidate_count"] >= 1
+    assert bundle["knowledge_view"]["entry_count"] == 1