# Taskboard Implementation Status

## Implemented

- Next.js App Router migration with React 19, Tailwind CSS, and shadcn-style UI primitives
- Typed fleet model loaded from `config/fleet.json`
- Typed task templates loaded from `config/task-templates.json`
- Unified task intake for OpenClaw, ZeroClaw, and direct SSH targets
- Dispatch lifecycle states:
  - `planned`
  - `assigned`
  - `dispatched`
  - `acknowledged`
  - `completed`
  - `failed`
- SQLite-backed audit log in `task_events`
- Dispatch history API and UI
- Failure queue UI
- Family-specific pages:
  - `/openclaw`
  - `/zeroclaw`
- Architecture page rendered from tracked fleet config
- Agent cards with:
  - heartbeat age
  - workload
  - last dispatch event
  - failure counts
- OpenClaw swarm dispatch:
  - repo map lookup
  - safe-directory git handling for mounted repos
  - worktree creation
  - queue insertion into `~/.clawdbot/active-tasks.json`
- ZeroClaw webhook dispatch:
  - bearer-token support for paired gateways
  - direct gateway mode for testing
- Direct SSH dispatch:
  - typed direct target definitions in `config/fleet.json`
  - safe built-in host actions for `pve`, `truenas`, and `panda`
  - completion written through the callback pipeline
  - dedicated `truenas-admin` audit action that correlates live ZFS datasets with repo dependency signals

## Verified Live

- `grizzley` ZeroClaw webhook dispatch from taskboard
- `ice` ZeroClaw webhook dispatch from taskboard
- OpenClaw swarm queue creation and host worktree creation on `ubuntu`
- direct SSH host actions can now be dispatched for `pve`, `truenas-admin`, and `panda`

## Current Limits

- Taskboard can dispatch OpenClaw swarm tasks, but it does not yet monitor tmux session progress automatically.
- ZeroClaw acknowledgements and completions are still operator-driven; remote runtimes do not push completion state back yet.
- The board records remote webhook responses, but not structured per-step execution output from the agents.
- Direct targets are intentionally restricted to configured safe actions and do not expose arbitrary shell execution in the UI.
- `truenas-admin` remains review-gated for cleanup decisions; the taskboard produces audit results and candidate datasets, not direct deletion.