BloodHound CE runtime
First identify which BloodHound tools the active agent actually has. The hosted engagement uses its own isolated bloodhound_mcp_* companion and live GUI; standalone OSS may expose bhce_* tools for a different local instance. Never mix their data or credentials. Load /skills/standard/ad/bloodhound-query/SKILL.md only when its detailed procedure is relevant; the path is not an automatic prerequisite.
In hosted runs, bloodhound_mcp_upload_collection ingests an authorized ZIP under /workspace. Verify terminal Complete, then use bloodhound_mcp_data_quality, bloodhound_mcp_domain_overview, bloodhound_mcp_find_objects, bloodhound_mcp_object_profile, bloodhound_mcp_exposure_finder, and bloodhound_mcp_shortest_path. Use bloodhound_mcp_cypher_query only for bounded read-only questions. The resulting GUI follow status is evidence that the remote viewer moved; opening the viewer by itself does not collect or analyze AD data.
In standalone OSS, when bhce_* tools are actually registered, use bhce_status to check the local instance, bhce_ingest_zip to import a compatible collection, and bhce_cypher to inspect its graph. The retired in-house bh_ingest_zip and related wrappers are deprecated. Do not assume a fixed BloodHound version or schema; use the actual instance's reported capabilities.
