Skip to content

Comergence — Skills Inventory

Every capability the agent fleet needs to bring Comergence to life. Compiled 2026-02-23

How to Read This

Skills are organized by layer — from foundational infrastructure up through domain-specific capabilities to human-facing and physical-world integration. Each skill includes what it does, which agents need it, and dependencies.

A "skill" in OpenClaw terms is a capability an agent can invoke — a script, an API connector, a protocol, a reasoning pattern. Some are universal (every agent needs them). Some are specialized (only the production agent needs PLC query capability).

Layer 1: Signal Infrastructure

The nervous system. Without these, nothing else works.

1.1 Signal Publishing

  • What: Compose and write signal objects to the signal bus (file-based → Redis/NATS)
  • Who: Every agent
  • Details: Validate against signal schema, set TTL, handle status transitions, auto-fire actions marked auto: true
  • Dependencies: Signal schema, bus write access

1.2 Signal Subscription

  • What: Watch the signal bus for signals matching path patterns, invoke callbacks
  • Who: Every agent, dashboard, notification router
  • Details: Wildcard path matching (production.*, *.work-centers.*), debounce rapid state changes
  • Dependencies: Bus read access, chokidar (file) or pub/sub client (Redis/NATS)

1.3 Domain Registration

  • What: Write/update comergence.json manifest to the registry directory
  • Who: Every agent at startup
  • Details: Copy workspace manifest to ~/.comergence/registry/, handle updates, deregister on shutdown
  • Dependencies: Registry directory, manifest schema

1.4 Health Evaluation

  • What: Periodically evaluate health rules from manifest against collected metrics, publish status signal
  • Who: Every agent
  • Details: Rule expression evaluation, worst-status-wins logic, configurable publish interval
  • Dependencies: Signal publishing, data source access

1.5 Status Propagation

  • What: Parent agents compute composite status from children's signals using intelligent reasoning (not just worst-of)
  • Who: Agents with children in the domain hierarchy
  • Details: Context-aware roll-up (scheduled downtime ≠ failure), unknown doesn't override known status, offline child → parent amber
  • Dependencies: Signal subscription, domain hierarchy awareness

1.6 Signal Archival

  • What: Move resolved/expired signals to history directory with metadata for later analysis
  • Who: Signal bus daemon or dedicated archival agent
  • Details: Retain full signal object, index by domain/time/code for querying
  • Dependencies: History directory, archive format spec

Layer 2: Data Connectors

How agents read the world.

2.1 MS SQL Connector (PLC/Machine Data)

  • What: Query MS SQL on automation-svr for real-time PLC data — cycle counts, temperatures, pressures, fault codes, operating hours
  • Who: Production agents, work center agents
  • Details: Connection pooling, parameterized queries, timeout handling, retry logic
  • Dependencies: SQL driver (mssql/tedious), network access to automation-svr, credentials in Doppler

2.2 NetSuite Connector

  • What: Query NetSuite via SuiteQL and REST for orders, inventory, financials, work orders, item records
  • Who: Production, fulfillment, finance, inventory agents
  • Details: OAuth token management, rate limit awareness, SuiteQL for complex queries, REST for CRUD
  • Dependencies: NetSuite API credentials, existing connection (already live)

2.3 Shopify Connector

  • What: Read orders, products, inventory levels, fulfillment status from Shopify
  • Who: Fulfillment agent, sales agent
  • Details: GraphQL preferred (quota-efficient), webhook subscription for real-time events
  • Dependencies: Shopify API credentials (already live)

2.4 HubSpot Connector

  • What: Read deals, contacts, pipeline stages, activities from HubSpot CRM
  • Who: Sales agent, marketing agent
  • Details: REST API, deal stage tracking, activity timestamps for stall detection
  • Dependencies: HubSpot API credentials (already live)

2.5 ShipStation Connector

  • What: Read shipment status, carrier rates, tracking, label generation status
  • Who: Fulfillment agent
  • Details: REST API, webhook for shipment events, carrier delay detection
  • Dependencies: ShipStation API credentials

2.6 Klaviyo / Meta / GA4 Connectors

  • What: Campaign performance metrics, email delivery/open/click, ad performance, web analytics
  • Who: Marketing agent
  • Details: Batch reads on schedule (not real-time), anomaly detection on trends
  • Dependencies: Respective API credentials

2.7 QuickBooks Online Connector

  • What: AR aging, cash flow data, budget actuals, expense tracking
  • Who: Finance agent
  • Details: OAuth refresh, report endpoints, transaction queries
  • Dependencies: QBO API credentials

2.8 Home Assistant Connector

  • What: Read sensor states, control entities (lights, switches, speakers), subscribe to state changes
  • Who: Home agents, physical bridge (andon)
  • Details: WebSocket API for real-time, REST for commands, entity ID mapping
  • Dependencies: HA instance, long-lived access token

2.9 Knowledge Base / Document Retrieval

  • What: Search and retrieve from structured documentation — maintenance manuals, SOPs, work instructions, troubleshooting guides
  • Who: Any agent that needs reference material (maintenance, production, quality)
  • Details: File-based initially (markdown/PDF in known directories), upgrade path to vector search. Key: agent must be able to find the right document for the current situation without human direction
  • Dependencies: Organized document repository, search/index capability

2.10 OpenClaw Internal Connector

  • What: Read agent session files, cron status, config, gateway health from OpenClaw's own data
  • Who: AI System agent (Maxrow)
  • Details: JSONL parsing, session age calculation, cron success/failure tracking, cost aggregation
  • Dependencies: Filesystem access to ~/.openclaw/

Layer 3: Agent-to-Agent Communication

How agents coordinate without a central brain.

3.1 Agent Messaging (Signal-Based)

  • What: One agent publishes a signal that another agent is subscribed to, triggering action
  • Who: Every agent
  • Details: This is the primary coordination mechanism. The maintenance cascade (machine → maintenance → MRO → scheduling) happens entirely through signal publish/subscribe
  • Dependencies: Signal bus, subscription patterns

3.2 Agent Task Spawning

  • What: One agent spawns a task on another agent with context (the agent-task action type)
  • Who: Any agent that needs another agent to do focused work
  • Details: OpenClaw sub-agent spawning with injected prompt, result auto-announcement back to requester
  • Dependencies: OpenClaw agent fleet, task prompt templates

3.3 Escalation Chain Management

  • What: If an agent can't resolve a situation or a human doesn't acknowledge, escalate to the next level
  • Who: Every agent with paired humans or critical signals
  • Details: Configurable timeout windows, escalation chain from manifest, tracking of acknowledgments
  • Dependencies: Notification routing, acknowledgment tracking

3.4 Cross-Domain Correlation

  • What: Detect that signals from different domains are related (machine down → schedule slip → shipment delay → customer impact)
  • Who: Parent-level agents, or a dedicated correlation agent
  • Details: Temporal correlation (signals within time window), causal chain reasoning, composite impact assessment
  • Dependencies: Multi-domain signal subscription, reasoning capability

Layer 4: Notification & Alert Routing

Getting the right signal to the right human at the right time.

4.1 Human Notification Router

  • What: Route signals to humans via their preferred channel (Telegram, SMS, email, Slack)
  • Who: Centralized notification service or each agent independently
  • Details: Channel preference lookup, severity-based routing (amber → Telegram, red → Telegram + SMS), quiet hours respect, dedup within window
  • Dependencies: Channel connectors (Telegram already live), human preference registry

4.2 Aperture Management

  • What: Filter and contextualize signals based on the observer's role, current task, and information needs
  • Who: Every agent paired with a human
  • Details: This is the core "minimum sufficient signal" capability. A floor supervisor sees work center roll-ups. An operator sees their machine's detail. A CEO sees the single light. Same data, different aperture
  • Dependencies: Role definitions, current task context, signal subscription with filtering

4.3 Guidance Generation

  • What: Translate a technical signal into actionable plain-language guidance for the specific human
  • Who: Every agent paired with an operator
  • Details: "X-axis feed is drifting +2.3%. Check the collet before the next cut." — not raw data, not alarm codes, but guidance the human can act on immediately
  • Dependencies: Domain knowledge (machine manuals, failure mode history), human skill level awareness

4.4 Acknowledgment Tracking

  • What: Track whether a human has seen and acknowledged a signal, trigger escalation if not
  • Who: Notification router
  • Details: Inline buttons in Telegram ("Acknowledged" / "Need Help"), timeout → escalation chain
  • Dependencies: Channel interaction support (Telegram inline buttons already available)

Layer 5: Domain-Specific Intelligence

The business knowledge that makes agents useful, not just connected.

5.1 Production Monitoring

  • What: Watch work orders, machine status, cycle times, throughput, quality metrics in real time
  • Who: Production agent (Sawyer)
  • Details: Order aging against SLA, machine utilization, WIP tracking, bottleneck detection
  • Dependencies: MS SQL connector, NetSuite connector

5.2 Machine Health / Equipment Monitoring

  • What: Track operating hours, cycle counts, vibration/temperature trends, fault history per machine
  • Who: Work center agents
  • Details: Predictive patterns (this motor's vibration signature looks like it did 2 weeks before the last failure), maintenance due calculations from simple JSON configs
  • Dependencies: MS SQL connector (PLC data), maintenance config files

5.3 Maintenance Cascade Orchestration

  • What: When maintenance is due: pull documentation, check MRO inventory, generate pick list, schedule task, notify technician — all automatically
  • Who: Machine agent → maintenance agent → MRO agent → scheduling agent
  • Details: The canonical Comergence example. No CMMS needed. Agent reads plain-language maintenance specs and reasons about timing, parts, and scheduling
  • Dependencies: Knowledge base, MRO inventory skill, scheduling skill, agent messaging

5.4 MRO Inventory Management

  • What: Track maintenance/repair/operations parts inventory, flag low stock, generate purchase requests
  • Who: MRO inventory agent (could be a capability of the production agent initially)
  • Details: Par levels per SKU, lead time awareness, auto-reorder thresholds, shortage flagging before maintenance is due
  • Dependencies: NetSuite inventory connector or dedicated MRO tracking

5.5 Quality Monitoring

  • What: Track defect rates, SPC data, hold/quarantine status, customer complaints correlation
  • Who: Quality agent or production agent with quality capability
  • Details: Statistical process control, trend detection, quality-gate enforcement, root cause correlation
  • Dependencies: Quality data source (MS SQL, manual entry, or NetSuite)

5.6 Scheduling Intelligence

  • What: Manage production schedule, detect conflicts, suggest rescheduling when disruptions occur
  • Who: Scheduling agent or production agent
  • Details: When a machine goes down, immediately assess impact on downstream jobs and suggest reallocation. Capacity-aware, priority-aware
  • Dependencies: Work order data, machine status, capacity model

5.7 Fulfillment Tracking

  • What: Monitor open orders from sale through shipment, flag SLA risks, track carrier performance
  • Who: Fulfillment agent (Sawyer)
  • Details: Order-level health (green/amber/red by ship date proximity), backlog monitoring, carrier delay detection
  • Dependencies: NetSuite connector, ShipStation connector, Shopify connector

5.8 Sales Pipeline Intelligence

  • What: Monitor deal velocity, stall detection, conversion rates, spec-to-bid pipeline
  • Who: Sales agent (Maven)
  • Details: Deal aging alerts, pipeline coverage analysis, activity gap detection
  • Dependencies: HubSpot connector

5.9 Financial Health Monitoring

  • What: AR aging, cash flow projection, budget variance, cost anomaly detection
  • Who: Finance agent (TBD)
  • Details: Weekly/monthly cadence sufficient (not real-time), threshold alerts on aging buckets
  • Dependencies: QBO connector, NetSuite financials

5.10 Integration Health Monitoring

  • What: Track API connectivity, auth token expiry, sync success/failure, rate limit usage across all connected systems
  • Who: AI System agent (Maxrow) or dedicated integration monitor
  • Details: Proactive token refresh alerts, sync failure detection, degraded performance flagging
  • Dependencies: API health check endpoints, token expiry tracking

Layer 6: Visual Layer & Interface

How humans see the system.

6.1 Tile Rendering

  • What: Render domain tiles with health color borders, names, icons, subtitles from registry + live signals
  • Who: Dashboard frontend
  • Details: Real-time update on signal change, smooth color transitions, responsive layout
  • Dependencies: Domain registry, signal subscription

6.2 Fractal Drill-Down Navigation

  • What: Click a tile → see its children. Same component at every level. Breadcrumb trail back up
  • Who: Dashboard frontend
  • Details: Domain hierarchy from registry, animated transitions, breadcrumb with click-to-navigate
  • Dependencies: Domain registry hierarchy

6.3 The Single Light

  • What: Top-level composite — one color gradient representing the entire operation's health
  • Who: Dashboard root view
  • Details: RGB-based composite from all top-level domains, numeric health score, gradient rendering
  • Dependencies: All top-level agent signals, composite calculation

6.4 Signal Detail View

  • What: Click a tile's status → see the signal history, current alert details, available actions
  • Who: Dashboard detail panel
  • Details: Signal timeline, action buttons that trigger agent tasks or notifications, resolution tracking
  • Dependencies: Signal archive, action execution

6.5 Mobile Interface

  • What: Operator-facing mobile view — their work center, their guidance, their acknowledgment buttons
  • Who: Mobile web app (PWA)
  • Details: Minimal, focused — the aperture for floor operators. Not a dashboard, a partner interface
  • Dependencies: Agent pairing, Telegram or PWA

Layer 7: Physical World Integration

Where digital signals become physical reality.

7.1 Andon Light Control

  • What: Drive physical RGB lights (Hue, WLED, or industrial) to match domain health status
  • Who: Physical bridge via Home Assistant
  • Details: Map signal status → light color, priority logic when multiple signals compete, flash pattern for new alerts
  • Dependencies: Home Assistant connector, physical lights installed

7.2 Audio Alerts (Sonos / Speakers)

  • What: Play distinct tones or TTS announcements on floor speakers for critical signals
  • Who: Physical bridge via Home Assistant
  • Details: Severity-mapped sounds, volume control by time/shift, TTS for specific guidance
  • Dependencies: Home Assistant media player entities, Sonos/speaker setup

7.3 Display Boards

  • What: Render the dashboard on wall-mounted displays in the shop, office, and shipping areas
  • Who: Dashboard in kiosk mode
  • Details: Auto-rotate through relevant domains, enlarged for visibility, touch-enabled for drill-down
  • Dependencies: Dashboard frontend, display hardware

7.4 Sensor Ingestion

  • What: Read physical sensors (temperature, humidity, vibration, pressure) via Home Assistant or direct GPIO
  • Who: Work center agents, facility agent
  • Details: Threshold monitoring, trend analysis, correlation with machine health
  • Dependencies: Sensor hardware, Home Assistant or direct sensor protocol

Layer 8: System Administration & Governance

Keeping the system itself healthy and trustworthy.

8.1 Configuration Management

  • What: Manage agent configs, manifests, health rules, notification preferences as version-controlled files
  • Who: AI System agent (Maxrow), David
  • Details: Git-backed, change tracking, rollback capability, validation before deploy
  • Dependencies: Git, workspace conventions

8.2 Audit Trail

  • What: Immutable log of every signal published, every action taken, every status change
  • Who: Signal bus / archival service
  • Details: Who published what, when, what actions fired, what humans were notified, what was acknowledged. Critical for compliance and post-incident review
  • Dependencies: Signal archival, append-only log

8.3 Access Control / Role-Based Aperture

  • What: Define what each human role can see and do — operators see their machine, supervisors see the floor, Jerry sees everything
  • Who: Dashboard auth layer, notification router
  • Details: Role → domain path visibility mapping, action permission levels
  • Dependencies: Role definitions, auth system

8.4 System Self-Monitoring

  • What: Comergence monitors itself — bus health, agent liveness, manifest staleness, signal freshness
  • Who: AI System domain (Maxrow)
  • Details: The first domain deployed. If the monitoring system can't monitor itself, nothing else is trustworthy
  • Dependencies: OpenClaw internal connector

8.5 Backup & Recovery

  • What: Signal history, manifests, configs backed up. Recovery procedures documented and tested
  • Who: Maxrow or infrastructure automation
  • Details: Daily archives, recovery runbook, tested restore from backup
  • Dependencies: Backup storage, automation scripts

8.6 Cost Monitoring

  • What: Track API costs across all agents, flag spikes, project monthly spend
  • Who: AI System agent
  • Details: Per-agent, per-connector cost tracking. Alert on unusual patterns
  • Dependencies: OpenClaw cost data, API usage metrics

Skill Priority Matrix

Skill Phase 0 (Foundation) Phase 1 (First Domains) Phase 2 (Expand) Phase 3 (Physical)
Signal Publishing ✅
Signal Subscription ✅
Domain Registration ✅
Health Evaluation ✅
MS SQL Connector ✅
NetSuite Connector ✅
Agent Messaging ✅
Human Notification ✅
Tile Rendering ✅
Drill-Down Navigation ✅
Status Propagation ✅
Aperture Management ✅
Maintenance Cascade ✅
Knowledge Base ✅
Andon Light Control ✅
Audio Alerts ✅
Mobile Interface ✅
Sensor Ingestion ✅

Every skill serves the same principle: minimum sufficient signal, delivered to the right observer, at the right time, through the right channel.