mcp_contract.yaml
Tools, resources, auth scope, and transport boundaries.
Connected systems.
I help CTOs and engineering leaders orchestrate the tools their teams already rely on, add a Control Layer around AI execution, and ship cross-system workflows that stay reliable when the stakes become operational.
How I Work
Three escalation lanes: Workflow Infrastructure, Reliability and Control Layer, and Enterprise Extension.
1
Lane 1
Workflow Infrastructure
I build the integration and workflow logic for the operations that matter most, with clear trust boundaries and production-ready behavior.
2
Lane 2
Reliability and Control Layer
I add evals, release gates, policy rules, and incident reviews so automation stays reliable as workflow scope expands.
3
Lane 3
Enterprise Extension
For high-stakes workflows, I build orchestration layers across systems, with governance and traceability that native automation cannot yet provide.
What's Included
Start with workflow infrastructure, add reliability controls as risk grows, and extend for enterprise constraints.
Workflow Infrastructure
Workflow infrastructure designed specifically for your business rules and trust boundaries.
Includes
- Workflow endpoints
- Business-rule mapping
- Secure auth boundaries
- Handoff artifacts
The real product
Reliability and Control Layer
The reliability layer for production automation. Evals, release checks, and policy controls.
Includes
- Pre-release quality gates
- Policy and approval boundaries
- Postmortem loops
- Monthly tuning
Enterprise Extension
Beyond native limits. Architecture and orchestration for strict governance constraints.
Includes
- Cross-system orchestration
- Auditability controls
- Deterministic retries
- Architecture support
The Model
Native agent features are getting easier to adopt. That is good for teams and good for velocity.
The hard gap remains workflow-specific logic, approval boundaries, and live blocked states. That is where `.agency` compounds value.
- Safe actions run automatically once the workflow and tenant are in good standing.
- Risky actions route to approval instead of silently becoming manual cleanup.
- Disallowed actions stop with a reason buyers and operators can both understand.
When a client primarily needs full system development and team onboarding, I refer that engagement to Half Dozen directly.
Reliable workflows are the product. Operational leverage is the outcome.
Governed Execution
Control Layer
Hub MCP routes the request, and the Control Layer decides what can run automatically, what waits for approval, and what stops with a reason.
Client LLM
Ops Inbox
Background Agent
Routes
Hub MCP Tenant, host, session
Decides
Control Layer Reason-coded governance
Auto-allowApprovalBlock
CRM
ERP
Workflow System
Safe actions run fast. Risky actions route to approval. Disallowed actions stop with a reason.
Operating Artifacts
What makes the Control Layer durable
The buyer promise is backed by explicit contracts, runbooks, and release evidence, not by founder memory or hidden prompts.
agent_contract.yaml
Allowed actions, approvals, escalation triggers, and operating limits.
outcome_contract.md
Success metrics, manual fallback, and ownership boundaries.
runbook.md
Recovery steps, operator lanes, and rollback expectations.
golden-task checks
Regression evidence that keeps releases tied to real workflow behavior.
Need trust you can inspect?
Book a Policy Mapping Session and I will map your policy boundary and show what the artifact bundle looks like for your workflows.