Dashboard TUI Reference
Architectural Lineage (Credits):
SwarmCraft is an architectural fork and deep rewrite of the multi-agent swarm engine created by Mojomast in mojomast/swarmussy .
The dashboard pattern is also downstream of Mojomast’s original “mission control” approach for observing swarm activity.
SwarmCraft’s deterministic layering is derived from the meta-structure of Abstract Wiki Architect (AWA) .
Full details: Credits & Lineage
POWERED BY GROK
The SwarmCraft Dashboard is a Terminal UI (TUI) that observes and steers the engine while keeping UI and execution decoupled.
Key principle:
Engine runs the AI work
Dashboard renders state and sends control signals
This prevents UI freezes and supports restart-safe operation.
1) What the Dashboard Watches
The dashboard should primarily read:
data/matrix.json (runtime state)
data/story_bible/outline.json (structure reference, optional)
logs (engine logs, tool logs)
token usage metrics (if tracked)
The dashboard must never be the source of truth; it is a view/controller.
Matrix reference: Central Matrix
2) Recommended Layout (3-Column Mission Control)
A recommended high-level layout:
2.1 Context Panel (Left)
Shows “what the engine is currently focused on”:
active project (project_id)
active chapter + part (from matrix.active_task)
cast in scope (optional, derived from outline or tags)
location in scope (optional)
active tool (what the engine is doing right now)
2.2 Action Panel (Center)
Shows “what is happening”:
prose stream (latest generated text excerpt)
system log tail (planner decisions, tool calls, warnings/errors)
current step indicator (SCAN / PLAN / EXECUTE)
2.3 Progress Panel (Right)
Shows “how the project is doing”:
parts table (status, locks, word counts)
chapter rollups (optional)
integrity signals (missing files, schema drift)
cost/tokens counters
3.1 Active Task Card
Displays:
target_type (should be part)
target_id (Part ID)
action (DRAFT, REVISE, REVIEW)
reason (planner rationale)
elapsed time
3.2 Matrix Table
Backed by matrix.parts:
Part ID
Chapter ID
Status (EMPTY, DRAFTING, REVIEW_READY, REVISION_NEEDED, LOCKED)
Locked indicator
Word count
Last modified
3.3 Integrity / Alerts
Derived from scan validation:
missing manuscripts
outline/template mismatch (thread keys)
orphan manuscripts
locked conflicts
3.4 Token / Cost Tracker (Optional)
If token tracking is enabled:
tokens in / out per step
per-session totals
optional cost estimate
4) Control Surface (Director Override)
The dashboard may write control signals to a project-scoped control file, e.g.:
projects/<project_id>/data/control.json
Typical fields (recommended):
{
"run_state": "RUNNING",
"architect_override": {
"force_target_part_id": null,
"force_action": null,
"note": null
}
}
4.1 Run states
4.2 Override behaviors (Recommended)
Overrides should be applied at PLAN-time:
force focus on a specific part
force a specific action (review vs revise)
inject a short director note that becomes part of the plan rationale
The deterministic loop remains intact:
overrides influence selection, not execution ordering
Pipeline: Deterministic Pipeline
5) Multi-Project UX
If multiple projects exist, dashboard SHOULD display:
active project id
quick switch mechanism (implementation dependent)
project health summary
Multi-project: Multi-Project Management
6) Failure and Recovery UX
The dashboard should handle:
engine crash (dashboard stays up)
engine restart (dashboard reconnects to state files)
partial writes (prefer atomic writes in engine)
Recommended indicators:
“Engine heartbeat” or “last update timestamp” from Matrix
last SCAN time
last EXECUTE time
Matrix includes timestamps: Central Matrix
7) What the Dashboard Should Not Do
It should not run LLM calls.
It should not mutate manuscripts directly.
It should not edit Story Bible files directly unless explicitly a dedicated editor UI (separate feature).
All modifications should go through the engine tools layer.
8) Related Pages
© 2026 The kOA Initiative. Architected by Réjean McCormick .
AI access