---
name: automation-architecture
description: "Design and diagnose reliable workflows with explicit data ownership and failure handling. Use when the user requests automation architecture. Select relevant methods for the requested outcome; unrelated tasks do not require this skill."
metadata:
  version: "1.0.0"
---

# Automation Architecture

Design and diagnose reliable workflows with explicit data ownership and failure handling.

## OLLIE Runtime integration

Use this capability under the active OLLIE Operator. Read live Workspace Context and tenant bindings for client facts and required inputs. This generic package contains no client configuration. Consume ask-ollie-authority-core for authority, outloop-access-fallback for access and recovery, email-lifecycle-core for communication, and the bootstrapper Runtime Contract for ownership, scheduling and completion. Preserve domain constraints; this Skill creates no competing execution path. Use one method, combine skills, or fill missing steps with authorized tools as the task requires.

External access uses the active Workspace’s approved Outloop capabilities. Never request, read, embed or expose credentials. A service named in a method is a possible dependency, not an access grant. Local file and application work uses the host’s authorized tools.

## Required inputs

- Workflow export
- Input and output contracts
- Field ownership
- Retry and volume constraints

## Methods

### Trace an item

Trace one representative item end to end. Identify transformations, branching, joins, external writes and places where item context is lost.

### Define identity and ownership

Define stable event identifiers and field ownership. Separate a retryable read from an uncertain write and record reconciliation needs.

### Stage schema changes

For schema changes, stage migrations and test representative old and new records. Avoid replacing whole workflow structures when a narrow fix is sufficient.

### Test failure paths

Test duplicate events, empty pages, partial batches, timeouts and unexpected provider responses using fixtures. Runtime owns scheduling and execution authority; this skill supplies workflow-specific design.

## Completion

Verify findings against source evidence and read back the resulting artifact or authorized change. Report inputs used, methods selected, output location, assumptions and unresolved dependencies. Do not report a plan, queue entry, draft, test fixture or accepted request as a completed external action. Use the task’s requested format; if a tool is unavailable, finish independent work and identify the exact unfulfilled deliverable.

## Test before relying on it

Read [the synthetic evaluation cases](references/evaluation.md) for activation, output and isolation checks. These cases are test instructions, not evidence that a host or provider test has passed.
