Automation Playbooks

Bringing automation together. Playbooks for your organisation, skills for your agents, and blueprints for your systems.

Most company automations fail not because the idea was bad, but because they exist as isolated, brittle scripts. The moment a schema changes, a model updates, or a non-standard document arrives, the automation quietly dies.

Square Agent connects every layer of the automation stack: proven playbooks that mirror your operational procedures, modular skills that give your agents real competencies, and blueprints that run reliably on Cloudflare OS.

The Architecture

Three layers.
One reliable operation.

01 For your Organisation

Playbooks

The end-to-end standard operating procedure. Defines the business trigger, the handoffs between systems and teams, the approval gates, and the final deliverable.

Key components
  • Business triggers
  • Team handoffs
  • Approval gates
  • Target deliverables
02 For your Agents

Skills

The specific competencies agents draw upon to perform the work. Modular and testable units of intelligence and document understanding.

Key components
  • Document extraction
  • Policy & clause retrieval
  • Calculation & modeling
  • Drafting & summarisation
03 Works with AI OS

Blueprints

The battle-tested runtime scaffolding that executes the work safely. Handles sandboxed execution, tool connections, state, and audit logs.

Key components
  • Stateful Durable Objects
  • Secure tool sandboxes
  • MCP connectors
  • 24/7 run logging

Execution Flow

How a playbook actually runs

Every step is traceable, gated by policy, and verified before state changes on your production tools.

Step 01

Business Trigger

An inbound RFP arrives in Outlook, a new customer submits an onboarding form, or a Monday 06:00 cron triggers.

event.received → queue
Step 02

Agent Skills Execute

The agent parses attachments, cross-references historical data in the vector store, and calculates options.

tools.invoke → verify
Step 03 • Gate

Human Sign-off

High-stakes decisions (pricing, exceptions, deletions) pause for human review via interactive Slack or email cards.

policy.check → approved
Step 04

Deliverable & Log

Actions commit to target tools. A complete audit receipt with duration, tokens, and cost is permanently archived.

commit → receipt.save

Consultative Engineering

Build your own playbook.
Your process. Your rules.

When off-the-shelf automations don't match how your company actually operates, we build around your real handoffs, messy files, and specific human checks.

01

Map the real handoffs

We sit with the people who do the work today. We map the messy spreadsheets, unwritten checks, tool transitions, and real-world edge cases.

Deliverable Playbook Specification & Boundary Map
02

Assemble skills and blueprints

We equip agents with the domain-specific skills required and wire them into reliable Cloudflare OS blueprints using approved connectors.

Deliverable Working Staging Playbook in Cloudflare OS
03

Calibrate human gates

You define the boundaries of autonomy: what proceeds automatically, what requires human confirmation, and what triggers an immediate escalation.

Deliverable Custom Approval Policy & Permissions Matrix
04

Host, monitor & maintain

We deploy the playbook to production, log every run with exact token costs, and maintain the integration when upstream APIs or schemas evolve.

Deliverable 24/7 Run Monitoring & Ongoing Maintenance SLA

Reliability & Governance

The week two standard

Anyone can build an automation demo. We host, monitor, and maintain every playbook so it keeps running as models evolve and APIs change.

01

Every run is transparent

No mystery execution.

Every execution logs the initiating trigger, tools called, documents referenced, decisions made, and exact token costs. You can audit any run down to the second.

02

Approvals where you already work

Control without bottlenecks.

When an action requires sign-off, the playbook sends a structured review card into Slack, Teams, or email with the diff and supporting evidence.

03

Week two is our problem

Built for the real world.

Most automations quietly fail when a website updates, an API schema shifts, or credentials rotate. We monitor playbooks actively and fix breakage before you notice.

Talk through the work

Bring a technical problem or an opportunity you think AI could help with. We will use 20 minutes to work out whether there is a useful next step.

Book a conversation