hive-essentials

Hive Essentials

Read this before the other Hive skills. It gives you the mental model; the other skills give you the workflows.

What Hive Is

Hive is an agentic workflow orchestrator. It manages parallel task execution across multiple agents (human or LLM) with governance, evaluation, and observability.

The console is the primary interface for humans. The CLI is the primary interface for agents. Both read and write the same underlying state.

Entity Hierarchy

Workspace
├── Projects          (goals with policy)
│   ├── AGENCY.md     (narrative context for humans)
│   └── PROGRAM.md    (execution policy: budgets, allowed paths, evaluators)
├── Tasks             (units of work, canonical in .hive/tasks/*.md)
│   ├── status        (proposed → ready → claimed → in_progress → review → done)
│   ├── claims        (time-limited work leases)
│   └── links         (blocks, parent_of, relates_to, duplicates, supersedes)
├── Runs              (governed execution attempts on a task)
│   ├── worktree      (isolated git branch)
│   ├── transcript    (event stream)
│   ├── approvals     (driver ↔ human round-trip)
│   └── eval_results  (evaluator decision)
├── Campaigns         (multi-task orchestration with lanes and cadence)
│   └── Lanes         (exploit, explore, review, maintenance)
└── Memory            (observations and reflections, review-gated)

The Observe-and-Steer Pattern

Hive separates execution from judgment:

1. Agents execute. They claim tasks, launch runs, produce artifacts.
2. Evaluators gate. PROGRAM.md defines what passes. Runs are accepted, rejected, or escalated.
3. Humans steer. Via console or CLI, they approve, pause, reroute, or cancel runs. They set policy in PROGRAM.md.

As an agent, your job is to execute within policy and surface decisions that need human judgment. You do not self-approve governed runs unless the evaluator policy allows it.

Drivers

Drivers are the execution backends that actually run agent work. Each driver has different capabilities.