chore: vgi-lint 0.37.0 clean

Make vgi-nlp pass vgi-lint-check 0.37.0 at fail-on=info:

- VGI138: emit vgi.keywords as a JSON array of strings everywhere (catalog
  and schema tags in nlp_worker.py; object_tags now serializes its
  comma-separated keywords via json.dumps in vgi_nlp/meta.py).
- VGI139: drop the redundant per-object vgi.source_url (removed from the
  schema tags and from object_tags); source_url stays only on the Catalog.
- VGI312: every function/macro argument already carries an accurate doc
  (Param/ConstParam/Arg doc=...); verified clean.
- Bump the VGI SDK floor to vgi-python>=0.8.5 (pyproject extras + the PEP 723
  header in nlp_worker.py) so per-argument docs serialize; uv.lock updated.
- Pin the vgi-lint-check action to version "0.37.0" in CI.

Verified: vgi-lint --execute --fail-on info reports score 100, 0 findings;
ruff/mypy/pydoclint and pytest all green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: rustyconover

.github/workflows/ci.yml

@@ -214,5 +214,6 @@ jobs:
       - name: vgi-lint
         uses: Query-farm/vgi-lint-check@v1
         with:
+          version: "0.37.0"
           location: "uv run --python 3.13 nlp_worker.py"
           fail-on: info

nlp_worker.py

@@ -1,7 +1,7 @@
 # /// script
 # requires-python = ">=3.13"
 # dependencies = [
-#     "vgi-python[http]>=0.8.4",
+#     "vgi-python[http]>=0.8.5",
 #     "spacy>=3.7",
 #     "fasttext-wheel",
 #     "vaderSentiment",
@@ -35,6 +35,7 @@
 
 from __future__ import annotations
 
+import json
 from typing import Any
 
 from vgi import Worker
@@ -97,10 +98,23 @@
     source_url="https://github.com/Query-farm/vgi-nlp",
     tags={
         "vgi.title": "Classical NLP for SQL",
-        "vgi.keywords": (
-            "nlp, natural language processing, language detection, sentiment analysis, "
-            "named entity recognition, ner, tokenization, lemmatize, stop words, "
-            "noun chunks, spacy, fasttext, vader, text enrichment"
+        "vgi.keywords": json.dumps(
+            [
+                "nlp",
+                "natural language processing",
+                "language detection",
+                "sentiment analysis",
+                "named entity recognition",
+                "ner",
+                "tokenization",
+                "lemmatize",
+                "stop words",
+                "noun chunks",
+                "spacy",
+                "fasttext",
+                "vader",
+                "text enrichment",
+            ]
         ),
         "vgi.doc_llm": _CATALOG_DESCRIPTION_LLM,
         "vgi.doc_md": _CATALOG_DESCRIPTION_MD,
@@ -116,16 +130,29 @@
             comment="Classical NLP: language ID, sentiment, NER, tokenization for SQL",
             tags={
                 "vgi.title": "NLP Functions (main)",
-                "vgi.keywords": (
-                    "nlp, language detection, sentiment, ner, entities, tokens, "
-                    "sentences, noun chunks, lemmatize, strip stopwords, normalize, "
-                    "spacy, fasttext, vader"
+                "vgi.keywords": json.dumps(
+                    [
+                        "nlp",
+                        "language detection",
+                        "sentiment",
+                        "ner",
+                        "entities",
+                        "tokens",
+                        "sentences",
+                        "noun chunks",
+                        "lemmatize",
+                        "strip stopwords",
+                        "normalize",
+                        "spacy",
+                        "fasttext",
+                        "vader",
+                    ]
                 ),
                 # VGI123 classifying tags use BARE keys (not vgi.-namespaced).
                 "domain": "text-analytics",
                 "category": "natural-language-processing",
                 "topic": "language-detection-sentiment-ner",
-                "vgi.source_url": "https://github.com/Query-farm/vgi-nlp/blob/main/nlp_worker.py",
+                # VGI139: source_url lives only on the catalog object, not here.
                 "vgi.doc_llm": _SCHEMA_DESCRIPTION_LLM,
                 "vgi.doc_md": _SCHEMA_DESCRIPTION_MD,
                 "vgi.example_queries": _SCHEMA_EXAMPLE_QUERIES,

pyproject.toml

@@ -8,7 +8,7 @@ license = { file = "LICENSE" }
 authors = [{ name = "Query Farm", email = "hello@query.farm" }]
 keywords = ["vgi", "duckdb", "nlp", "spacy", "fasttext", "sentiment", "ner"]
 dependencies = [
-    "vgi-python>=0.8.4",
+    "vgi-python>=0.8.5",
     "spacy>=3.7",
     "fasttext-wheel",
     "vaderSentiment",
@@ -19,7 +19,7 @@ dependencies = [
 # addition to stdio. That path needs vgi-python's `http` extra (waitress).
 # CI's http transport leg installs this via `uv sync --extra http`.
 http = [
-    "vgi-python[http]>=0.8.4",
+    "vgi-python[http]>=0.8.5",
 ]
 dev = [
     "pytest>=8",

uv.lock

@@ -8,25 +8,26 @@
 - ``vgi.doc_llm`` (VGI112)      -- Markdown narrative aimed at LLMs/agents
 - ``vgi.doc_md`` (VGI113)       -- Markdown narrative for human docs (distinct
   content from ``doc_llm``)
-- ``vgi.keywords`` (VGI126)     -- comma-separated search terms/synonyms
-- ``vgi.source_url`` (VGI128)   -- link to the implementing source file
+- ``vgi.keywords`` (VGI126/VGI138) -- a JSON array of search-term/synonym strings
 
-``source_url(file)`` builds the canonical GitHub blob URL so every object points
-at exactly where it is implemented.
+Per-object ``vgi.source_url`` is intentionally NOT emitted here: VGI139 requires
+``source_url`` to live only on the catalog object (set on ``Catalog(...)`` in
+``nlp_worker.py``), not on every function/schema.
 """  # noqa: D205
 
 from __future__ import annotations
 
-#: Base GitHub blob URL for source files in this repo (pinned to ``main``).
-SOURCE_BASE = "https://github.com/Query-farm/vgi-nlp/blob/main"
+import json
 
 
-def source_url(relative_path: str) -> str:
-    """Build the implementation ``vgi.source_url`` for a repo-relative file.
+def keywords_array(keywords: str) -> str:
+    """Serialize comma-separated keywords as a JSON array of strings (VGI138).
 
-    e.g. ``source_url("vgi_nlp/scalars.py")``.
+    ``keywords`` is a comma-separated list (e.g. ``"ner, entities, spacy"``);
+    each term is trimmed and emitted as one element of a JSON string array.
     """
-    return f"{SOURCE_BASE}/{relative_path}"
+    terms = [k.strip() for k in keywords.split(",") if k.strip()]
+    return json.dumps(terms)
 
 
 def object_tags(
@@ -36,14 +37,15 @@ def object_tags(
     keywords: str,
     relative_path: str,
 ) -> dict[str, str]:
-    """Build the five standard per-object discovery/description tags.
+    """Build the standard per-object discovery/description tags.
 
-    ``relative_path`` is the implementing file relative to the repo root.
+    ``relative_path`` is the implementing file relative to the repo root; it is
+    retained for call-site documentation but no longer emitted as a per-object
+    ``vgi.source_url`` (VGI139 keeps ``source_url`` on the catalog only).
     """
     return {
         "vgi.title": title,
         "vgi.doc_llm": doc_llm,
         "vgi.doc_md": doc_md,
-        "vgi.keywords": keywords,
-        "vgi.source_url": source_url(relative_path),
+        "vgi.keywords": keywords_array(keywords),
     }