Hive One-Person Company Builder

Problem

Hive addresses a single-operator workflow where one person has to handle scheduling, execution, review, and archival at once. It is not a one-shot prompt tool; it scaffolds a multi-agent structure on a fresh OpenClaw instance. The CEO receives requests, breaks down tasks, and replies uniformly, while business subagents produce concrete artifacts and core state is written to files for inspection.

How It Works

The skill initializes in a fixed order:
- creates {WORKSPACE}/hive/ and a shared artifact path {WORKSPACE}/hive/artifacts/
- initializes #1 HRM, #2 OPS, and business specialists #3 through #8
- writes ROLE.md files for Collector, Analyst, Writer, QA, PM, and Doc, defining models, duties, and workflow placement
- updates proactivity/session-state.md, generates SETUP.md, and runs a test task to verify coordination

CEO is the main session and does not execute tasks directly; it dispatches subagents via sessions_spawn. HRM remains active for adding specialists and designing workflows, while OPS only performs version checks and compatibility evaluation when initiated by the CEO. Subagents do not answer the user directly; the CEO aggregates the final response.

Boundaries

Use this skill to build Hive from scratch. It is not the trigger for “Hive supports X” expansion requests, which route to HRM. It does not make OPS a system administrator: it will not change configs or run system-level actions without approval. The setup assumes an existing OpenClaw workspace, a user documents directory, and placeholders such as {DEFAULT_MODEL}, {FAST_MODEL}, and {STRONG_MODEL}.