hive-project-setup - SKILL.md Agent Skill

name: hive-project-setup description: Set up and configure Hive workspaces, projects, tasks, and evaluator policy. Use this skill when bootstrapping, creating projects, managing tasks, or configuring PROGRAM.md.

Hive Project Setup

This skill covers workspace bootstrap, project creation, task management, and evaluator configuration.

Bootstrap a New Workspace

Full onboarding (recommended)

hive onboard <project-slug> --json

This creates:

.hive/ substrate directory with cache and task storage
projects/<slug>/AGENCY.md — narrative project document
projects/<slug>/PROGRAM.md — execution policy with forgiving defaults
3–5 starter tasks with dependency chains
local-smoke evaluator template
Workspace checkpoint commit

Default PROGRAM.md uses forgiving settings so the first hive finish succeeds cleanly:

auto_close_task: true
allow_accept_without_changes: true

Tighten these for production use.

Minimal init

hive init --json

Creates the .hive/ layout only. No starter project or tasks.

Adopt an existing repo

hive adopt <project-slug> --json

Wraps existing v1 state and enables v2 features incrementally.

Migrate from v1

hive migrate v1-to-v2 --dry-run --json     # preview
hive migrate v1-to-v2 --project <id> --json # execute

Create and Manage Projects

hive project create <slug> --title "Project title" --json
hive project list --json
hive project show <project-id> --json
hive project sync --json

Each project gets:

projects/<slug>/AGENCY.md — write narrative context here (objective, notes, recommended flow)
projects/<slug>/PROGRAM.md — define execution policy here

Create and Manage Tasks

Task CRUD

hive task create --project-id <id> --title "Task title" --json
hive task create --project-id <id> --title "..." --status ready --json
hive task list --project-id <id> --json
hive task list --status ready --status review --json
hive task show <task-id> --json
hive task update <task-id> --status in_progress --priority 1 --json
hive task update <task-id> --title "New title" --label "bug" --json

Task status flow

proposed → ready → claimed → in_progress → review → done

Side states: blocked (manual), archived (deprecated).

Priority levels

0 = critical, 1 = high, 2 = medium, 3 = low. hive next returns highest priority first.

Task dependencies

hive task link <blocked-id> blocked_by <blocker-id> --json
hive task link <parent-id> parent_of <child-id> --json
hive task link <id-a> relates_to <id-b> --json
hive deps --json     # project-level blocker summary

Edge types: blocks, parent_of, relates_to, duplicates, supersedes.

Claims

hive task claim <task-id> --owner <name> --ttl-minutes 30 --json
hive task release <task-id> --json

Claims are time-limited work leases. Use them, not hand-edited owner fields in AGENCY.md.

Configure Evaluator Policy

Inspect current policy

hive program doctor <project-id> --json

This shows the current PROGRAM.md state, available evaluator templates, and any policy issues.

Add an evaluator

hive program add-evaluator <project-id> local-smoke --json

Available templates:

local-smoke — minimal validation stub, proves the evaluator loop works

Custom evaluators can be defined directly in PROGRAM.md.

PROGRAM.md structure

program_version: 1
mode: workflow
default_executor: local
budgets:
  max_wall_clock_minutes: 30
  max_steps: 25
  max_tokens: 20000
  max_cost_usd: 2.0
paths:
  allow:
    - README.md
    - docs/**
    - projects/<project>/**
  deny:
    - secrets/**
    - infra/prod/**
commands:
  allow: []
  deny:
    - rm -rf /
evaluators: []
promotion:
  allow_unsafe_without_evaluators: false
  auto_close_task: true
  requires_all: []
escalation:
  when_paths_match: []
  when_commands_match: []

Read PROGRAM.md before:

Running evaluators
Broad file edits
Autonomous command execution
Accepting or escalating runs

If work needs new guardrails, update PROGRAM.md rather than inventing ad hoc rules.

Workspace Maintenance

Checkpoint

hive workspace checkpoint --message "before refactor" --json

Creates a git commit to freeze workspace state. hive work does this automatically before launching runs.

Refresh projections

hive cache rebuild --json
hive sync projections --json

Run this after task, run, or memory changes so GLOBAL.md, AGENCY.md task rollups, and AGENTS.md stay current.

Health check

hive doctor --json                          # full workspace health
hive doctor workspace --json                # workspace structure only
hive doctor program <project-id> --json     # PROGRAM.md policy check

Rules of Thumb

Canonical task state lives in .hive/tasks/*.md. AGENCY.md is narrative, not machine state.
Keep AGENCY.md readable and human-oriented. Keep operational truth in .hive/.
Use task claims for work leases and task links for blockers.
Start with forgiving defaults (auto_close_task, allow_accept_without_changes) and tighten for production.
Do not use the project owner field as the primary locking mechanism — use task claims.