chore: auto-commit (286 archivos)

- .claude/agents/fn-orquestador/SKILL.md
- .claude/commands/fn_claude.md
- .claude/rules/INDEX.md
- .claude/rules/cpp_apps.md
- .claude/rules/ids_naming.md
- CHANGELOG.md
- apps/dag_engine/README.md
- apps/dag_engine/api.go
- apps/dag_engine/dags_migrated/example.yaml
- apps/dag_engine/dags_migrated/example_lineage_tracking.yaml
- ...

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docs/capabilities/nlp.md

@@ -1,6 +1,6 @@
 # Capability: nlp
 
-_(Descripcion del grupo — editar a mano)_
+Pipeline NLP para extraccion de entities/relations sobre documentos en castellano (proyecto OSINT principalmente). Cubre: lectura de PDFs (`extract_pdf_text`, `clean_pdf_text`), OCR fallback, chunking (`chunk_with_overlap`), inferencia GLiNER (NER) + GLiREL (relation extraction), dedup (`dedup_entities`, `dedup_relations`), agregacion (`aggregate_extraction_results`), extraccion de elementos especificos (URLs, crypto wallets, IPs, dominios).
 
 ## Funciones
 
@@ -43,8 +43,37 @@ _(Descripcion del grupo — editar a mano)_
 
 ## Ejemplo canonico
 
-_(Anadir 1-2 bloques de codigo end-to-end)_
+### Pipeline completo PDF -> entities + relations
+
+```python
+import os, sys
+sys.path.insert(0, os.path.join(os.environ["FN_REGISTRY_ROOT"], "python", "functions"))
+
+from core import extract_pdf_text, clean_pdf_text, chunk_with_overlap
+from datascience import (
+    gliner_extract_entities, glirel_extract_relations,
+    dedup_entities, dedup_relations, aggregate_extraction_results,
+)
+
+raw = extract_pdf_text("/path/to/doc.pdf")
+text = clean_pdf_text(raw)
+chunks = chunk_with_overlap(text, size=512, overlap=64)
+
+entities = []
+relations = []
+for ch in chunks:
+    entities.extend(gliner_extract_entities(ch, labels=["PERSON","ORG","ACCOUNT"]))
+    relations.extend(glirel_extract_relations(ch, entity_pairs=entities))
+
+result = aggregate_extraction_results(
+    entities=dedup_entities(entities),
+    relations=dedup_relations(relations),
+)
+```
 
 ## Fronteras
 
-_(Que NO cubre este grupo)_
+- **NO entrena modelos**. Solo inferencia con GLiNER/GLiREL pre-entrenados (HuggingFace).
+- **NO maneja embeddings densos** (sentence-transformers / e5). Para vectores, usa funciones del grupo `ml`.
+- **NO hace traduccion ni summarization LLM**. Solo NER + RE. Para LLM, ver tag `llm`.
+- **NO escribe a BD** automaticamente. La persistencia (vault, sqlite, parquet) la maneja el caller via funciones de `infra`/`datascience`.