Run production multi-agent workflows with Hive
Use Hive as a model-agnostic harness for long-running multi-agent workflows that need state, recovery, observability, human oversight, and business-system tools.
npx skills add agentskillexchange/skills --skill run-production-multi-agent-workflows-with-hive
https://github.com/aden-hive/hive.git, enter the repository, then run ./quickstart.sh on macOS/Linux or ./quickstart.ps1 on Windows. Do not use pip install -e . from the repository root; the upstream README states that Hive uses a uv workspace layout.Hive is an open-source multi-agent harness for moving agent workflows beyond one-off scripts. It generates graph-based execution plans, coordinates specialized agents, persists role-based memory, connects tools through MCP, and adds the operational controls needed for long-running work: state management, failure recovery, observability, cost limits, and human-in-the-loop review.
What this skill actually does
Use this skill when a team wants an agent to execute recurring business processes or complex multi-step workloads where a single prompt, chain, or chat session is too brittle. Good fits include workflows that need parallel task execution, crash recovery, model-provider flexibility, auditability, and access to CRM, support, messaging, data, file, or internal API tools.
Scope boundary: this is not a generic agent framework card. The workflow is specifically to stand up Hive/OpenHive as the runtime harness, define or select agent workflows, connect the needed tools and model providers, and run those workflows with explicit state, recovery, and oversight controls.
Inputs and prerequisites: Hive/OpenHive, Python 3.11+, uv workspace setup, LLM provider credentials, optional MCP business-system tools, optional ripgrep.
Setup notes: Clone `https://github.com/aden-hive/hive.git`, enter the repository, then run `./quickstart.sh` on macOS/Linux or `./quickstart.ps1` on Windows. Do not use `pip install -e .` from the repository root; the upstream README states that Hive uses a uv workspace layout.
Source and verification boundary: use https://docs.adenhq.com/ as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Multi-Framework workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.