Claude Skills by Hivemind-OSS

Snapshot, reset, or restore the Hivemind data store via the `hive` CLI. Use when asked to back up / snapshot the store, wipe and recreate it empty, recover from a post-upgrade boot crash-loop, or roll the store back to a prior snapshot. `reset` and `restore` are destructive and require a typed confirmation.

Stand up, restart, stop, or health-check the Hivemind MCP server on its host via the `hive` CLI. Use when asked to start / boot / bring up / restart / stop the hive server, check whether it is healthy, follow its logs, or diagnose a boot crash-loop. Covers the zero-config first boot, the bounded health-wait, and the schema-refusal failure mode.

Install the optional agent-loop harness — the hive-loop Claude Code plugin — on a workstation and point it at a running Hivemind server: on your own device, or as a copy-pasteable block an admin forwards so a teammate's agent installs it at a pinned commit. Use when asked to install / enable / roll out the harness or the hive-loop plugin, make the memory loop mechanical or enforced client-side, onboard a teammate to the harness, or diagnose an installed harness that appears to do nothing.

Register a repo with the server-side census sync: `hive repo add <url>` writes the durable repo registry (the sync daemon picks it up on its next tick — no restart), private remotes authenticate via a token env var NAME (never a stored secret), and `hive_health(include_census_health=true)` verifies both that the sync daemon itself is alive (the `fleet` block) and that the per-repo change-outcome feed is live (the `repos` blocks). Use when asked to connect / register / wire a repo to the hive,...

Connect agents and teammates to a running Hivemind server: pick the transport (local loopback / ngrok tunnel / SSH), mint and hand off per-seat tokens, register the MCP client, and revoke seats. Use when asked to connect / add / provision an agent or teammate, expose the server to remote machines, share the hive, or offboard a seat.

Operate and tune a running Hivemind server: read the convergence KPIs over MCP, interpret them, turn the recall / safety knobs, manage the synced-repo registry (hive repo add/remove, hive repos), watch the per-repo automatic census feed, and feed manual census receipts into the evidence ledger (hive ingest). Use when asked to check hive health or KPIs, run the weekly maintenance sweep, tune recall or promotion, fix coverage starvation or silent rot, decide demand_m / tau_serve / conflict.tau,...

Move a running Hivemind server to a different release ref safely via `hive upgrade` — snapshot-gated, health-verified, auto-rolling-back — after a schema pre-flight that proves the target ref will accept the store you already have. Use when asked to upgrade / update / move / bump the hive server to a new version, tag, or release; to roll back to an older ref; or to check whether a new version is safe to adopt before committing to it. This build has NO in-place store migration, so a schema-bre...