2 skills · sorted by relevance
Execute Indirect Prompt Injection attacks against Large Language Models (LLMs) by subtly embedding malicious instructions within external data sources (e.g., websites, documents, databases) that the LLM autonomously ingests. This forces the model to execute attacker-controlled commands under the guise of processing legitimate user requests.
The English summary is being prepared. Showing the original description.
Open-format data ingestion for NV oOS Pro toolkits — the reusable pattern behind the F&B data-ingestion track (Proposal 078). Covers structural (header) matching over filename mapping, the configurable table_map + header-matcher + name-guess resolution order and the slug-first regression, the multi-format loader (CSV/TSV/JSON/JSONL/XLSX/ODS/PDF/photo), the Observation + terminology-map layer (vitals/FHIR analogue) with raw preservation, dedupe/retention/out-of-band banding, complete-by-default/resumable (offset) imports, and the read-only ingest-review dashboard anatomy (edit_posts review vs manage_options write, dry-run-first). Use when adding or extending data ingestion to a toolkit, ingesting heterogeneous owner files, wiring XLSX/PDF/photo imports, building an ingest/review dashboard, or debugging partial/truncated imports.
The English summary is being prepared. Showing the original description.