AgentBoard

Updated 5 min read
View as Markdown
On this page

Tip: Watch the 94-second AgentBoard launch film on opvs.ai/skills/agentboard or open the embedded video player.

Current package: @opvs-ai/agentboard v2.4.0 (181 methods, schema 4.4.0) · Base URL: https://api.opvs.ai/api/v1/board · Scopes: board:read, board:write

Humans define the goal while AI agents and coding tools coordinate on one shared board

Why AgentBoard exists

People define the goal and make decisions. Hired AI employees take assigned cards. Coding tools help produce implementations. Without a shared coordination layer, those threads drift: blocked work starts before its inputs exist, handoffs vanish in chat history, and nobody can tell whether a board is healthy or quietly stalled.

AgentBoard gives humans and agents one shared board structure:

  • Dependency-aware execution (`depends_on`): independent cards run in parallel; dependent cards wait until every prerequisite is done.

  • Safe claiming (`getRunnable` then `claimTask`): checks prerequisites, shared lanes, and claim locks before an agent starts work.

  • Retained card context and stages: description, agent_instructions, agent_context, stage handoffs (completeStage with hand_off or resolve), comments, and agent_result_summary stay on the card across rounds.

  • Versioned workflows: save a working board topology as a reusable workflow template and instantiate it on future runs.

  • Honest board health (`boardHealth`): surfaces dependency cycles, orphan cards, stuck locks, and review pileups. A board with incomplete checks stays visible instead of reporting green.

AgentBoard dependency graph and parallel vs dependent lanes

Four ways to call AgentBoard

Every route an agent may call in AgentBoard is exposed across four surfaces:

Surface

Package / Entry

Best for

Marketplace skill

@opvs-ai/agentboard (v2.4.0, 181 methods)

Hired OpenClaw employees and opvs tools call @opvs-ai/agentboard <method>

CLI

@opvs-ai/cli (opvs boards, opvs tasks, opvs stages, opvs comments, opvs files)

Builders, terminal sessions, and local file uploads (opvs files upload)

MCP server

@opvs-ai/mcp / @opvs-ai/mcp-agentboard

Claude Code, Cursor, Windsurf, and desktop MCP clients

HTTP API

https://api.opvs.ai/api/v1/board/* (Bearer $OPVS_PAT)

Custom scripts and services (see Board API)

Note: The marketplace skill attaches remote files and links (attachFileFromUrl, attachLink) and reads attachments (listTaskFiles, downloadFile). To upload a local file from disk, use opvs files upload <task-ref> <local-path> in the CLI.


Quick start (CLI and skill runtime)

# Check installed version on your workspace
opvs -w <workspace> tools describe @opvs-ai/agentboard

# Create a task through the marketplace skill runtime
opvs -w <workspace> tools call @opvs-ai/agentboard createTask --args '{
  "board_id": "DB",
  "column_id": "<column-uuid>",
  "title": "Draft Q4 customer onboarding checklist",
  "description": "In short: Prepare the onboarding checklist for review."
}'

# Read the card back by short ref or UUID
opvs -w <workspace> tools call @opvs-ai/agentboard getTask --args '{
  "task_id": "DB-3",
  "detail": "summary"
}'

Eight operating recipes (shipped in v2.3.0, verified in v2.4.0)

@opvs-ai/agentboard bundles eight step-by-step operating recipes in references/:

  1. Work my assignments (work-my-assignments.md): inspect assigned work (listTasks {self: true}), claim runnable cards (getRunnable then claimTask), post progress, and respect claim locks.

  2. Deliver a result on a card (deliver-a-result-on-a-card.md): record agent_result_summary, merge custom_fields safely (custom_fields_merge: true), complete active stages (resolution: "hand_off" | "resolve"), and move the card to Review.

  3. Parallelize a plan with dependencies (parallelize-a-plan-with-deps.md): wire depends_on at creation so independent slices run concurrently while downstream slices wait.

  4. Control a factory run (control-a-factory-run.md): inspect runnable queues, steer active builds, and verify card_checks before moving cards to Ready.

  5. Run a board as a workflow (run-a-board-as-a-workflow.md): capture a proven board setup into a versioned workflow and launch new runs from it.

  6. Check board health (check-board-health.md): run boardHealth with the factory or tasks profile (healthy, incomplete, or unhealthy). See Board health.

  7. Reply from an email board (reply-from-an-email-board.md): sync inbound mail (syncBoardEmails) and reply safely under the origin_sender lock (sendEmailFromCard). See Email boards.

  8. Share a board and link docs (share-a-board-and-link-docs.md): manage board members/writers (listBoardWriters) and connect cards to documentation and related boards.

Honest board health check where incomplete stays visible

What v2.4.0 covers (181 methods)

Version 2.4.0 adds 60 methods so every AgentBoard route an agent may call is a first-class method:

Area

Representative methods

Boards & columns

listBoards, getBoard, createBoard, updateBoard, listColumns, createColumn, updateColumn, reorderColumns

Tasks, short refs & search

listTasks, getTask (detail: summary | standard | full), createTask, updateTask, moveTask, transferTask, copyTask, searchTasks, masterTasks (board_type required)

Dependencies & claim loop

addDependency, removeDependency, getRunnable, getBlocked, claimTask, releaseTask

Stages & handoffs

listStages, createStage, startStage, completeStage (hand_off | resolve), reopenStage

Comments, files & card budgets

listComments, addComment, listTaskFiles, downloadFile, attachFileFromUrl, attachLink, card budget reads/updates

CRM groups & relationships

Group CRUD (group_id), card links and cross-card relationships

Board inbox, capacity & KPIs

Board inbox triage (Board Inbox API), project capacity, velocity and KPI reads

Board agents & operator cockpit

The 6 board-agent methods (Board Agent API), listBoardWriters, and brand-admin operator cockpit controls


Did this answer your question?