oa-web-dashboard - SKILL.md Agent Skill

name: oa-web-dashboard description: "Access and interpret the oa web UI and Textual TUI. Use when opening the dashboard, checking agent status visually, reading logs, or using the bridge API. Activates for: oa dashboard, oa web, web UI, localhost:5174, bridge API." user-invocable: false allowed-tools: Bash(oa *)

Critical Rules

ALWAYS use oa status for quick CLI checks — because the dashboard is for visual monitoring, not scripted automation.
NEVER rely on the bridge API from inside agent prompts — because agents do not have network access to localhost; use oa CLI commands instead.

Decision Tree

Need to check agent state?
├── Quick CLI check → oa status
├── Detailed logs for one agent → oa logs <name> or oa collect <name>
├── Visual overview of all agents → oa dashboard (Textual TUI)
└── React web UI with canvas → oa web → open http://localhost:5174

Instructions

Open the Textual TUI dashboard:
```
oa dashboard
```
Start the React web UI (bridge server on port 5174):
```
oa web
```
Then open: http://localhost:5174
Check bridge health:
```
curl http://localhost:5174/api/health
```
Query agent list via API:
```
curl http://localhost:5174/api/agents
```

Get live output for a specific agent:

curl http://localhost:5174/api/agents/<name>/output?lines=50

Bridge API Endpoints

Method	Endpoint	Purpose
GET	`/api/health`	Health check
GET	`/api/agents`	List all agents with status
GET	`/api/agents/<name>`	Get single agent detail
GET	`/api/agents/<name>/output`	Live terminal output (`?lines=N`)
POST	`/api/agents`	Spawn new agent (`task`, `name`, `model`)
POST	`/api/agents/<name>/kill`	Kill agent
POST	`/api/agents/<name>/pause`	Pause agent (suspend tmux pane)
POST	`/api/agents/<name>/resume`	Resume paused agent
POST	`/api/clean`	Clean finished workspaces
GET	`/api/session/status`	Check tmux session exists
POST	`/api/session/start`	Start tmux session
GET	`/api/pipeline`	List pipeline agents (pipe-*)
GET	`/api/guardians`	List configured guardians
POST	`/api/guardians/trigger`	Manually trigger a guardian
GET	`/api/messages/<name>`	Get agent inbox (`?unread=true`)
POST	`/api/messages`	Send message (`from`, `to`, `content`)
POST	`/api/broadcast`	Broadcast to all agents
GET	`/api/teams`	List teams
POST	`/api/teams`	Create team (`name`, `members`)
GET	`/api/teams/<name>`	Get team detail
GET	`/api/tasks/<team>`	List team tasks
POST	`/api/tasks/<team>`	Create task (`title`, `description`)
GET	`/api/templates`	List library templates
GET	`/api/templates/<id>`	Load specific template
GET	`/api/checkpoints`	List incomplete checkpoints
POST	`/api/resume/<agent>`	Resume from checkpoint

Patterns

Pattern 1: Spawn agent via bridge API

curl -X POST http://localhost:5174/api/agents \
  -H "Content-Type: application/json" \
  -d '{"task": "research topic X", "name": "researcher-1", "model": "claude/sonnet"}'

Pattern 2: Check unread messages

curl "http://localhost:5174/api/messages/my-agent?unread=true"

Anti-Patterns

Bad: curl localhost:5174/api/agents from inside an agent prompt — agents cannot reach localhost bridge.
Good: Use oa status and oa collect <name> for agent-to-system communication.
Bad: Opening oa web for every status check — TUI or oa status is faster for scripted use.
Good: Reserve oa web for visual canvas work or when monitoring many agents simultaneously.

References

Related: oa-orchestration-spawn, oa-teams-coordination, oa-library-discovery