Must be named by the user (Patent Map / Case Map / /专利地图 / /patent-map）。
Do notenter automatically due to "reading patents / writing disclosure documents". Interpretation continues to only ingest into the vault; this package only reads the vault and starts a local page.
Readprompts/guardrails.md→intake.md- Confirm the vault path (the same one as for interpretation:
PATENT_READER_OBSIDIAN_VAULTor~/.patent-disclosure-skill/obsidian_vault.txt） - Startup:
python skills/patent-map/tools/serve_map.py
stdout machine-readable prefix:MAP_URL: / MAP_PORT: / MAP_VAULT: / MAP_SOURCE: / MAP_CACHE: / MAP_MODEL:。
The port is randomly assigned by the system (127.0.0.1:0), to avoid conflicts. Send the URL to the user in the conversation;Do notweld the service into interpretation ingestion.
The terrain map is spread out according to the vectors of the interpretation short texts. The loading name is BAAI/bge-small-zh-v1.5, and only Qdrant ONNXis installed on disk (not PyTorch). Default directories for data and models:{Documents}/patent-disclosure-skill/patent-map/(at the same level as oa). When fastembed is not installed, the map can still open, and the terrain falls back to IPC clusters.
Optional:
pip install -r skills/patent-map/tools/requirements.txt
python skills/patent-map/tools/ensure_model.py
When copying this package separately:python tools/serve_map.py。
Fetch data: preferentially scan the vault directory Research/Patents/**/*_解读_*.md. If the local obsidian CLI is available and the vault is already open, use obsidian vaults / base:query to verify, without replacing disk scanning.