chroma-hybrid-search

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In scripts/search.py the required workflow auto-calls indexer.ensure_fresh() (which runs indexer.sync() and kb_reader.read_knowledge_base() / read_cold_notes()), ingesting outsider-authored free text from user-writable cold store cold-notes/raw.jsonl and hot markdown files under knowledge-base/*.md and experience/*.md at runtime (e.g., those produced via scripts/write_cold.py).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The code calls huggingface_hub.hf_hub_download at runtime to fetch ONNX model weights/tokenizers from the Hugging Face repos "intfloat/multilingual-e5-small" and "BAAI/bge-reranker-base", and those downloaded artifacts are loaded/executed by onnxruntime/tokenizer (i.e., runtime-fetched model files control inference), so they are runtime external dependencies.