PATRICK RIBBSAETER
CH•
Open Source & Case StudiesAGENT ARCHITECTURE
PRODUCTION ENGINEERING

Designing a Two-Authority AI Production System

Designed and validated a routing architecture that separates end-to-end commercial production from individual-shot prompting while preserving one canonical authority per workflow.

Role

  • Agent architecture
  • Skill routing
  • Prompt systems
  • AI governance
  • Authority design
  • Validation
  • Production workflows

Category

Agents

Engineering discipline

PRODUCTION ENGINEERING

Non-Competing Authorities:2 Active
Routing Scenarios:20 Validated
Structural Validations:4 Passes ("Skill is valid!")
Two-authority AI production system decoupling campaign planning from shot generation

Patrick Ribbsaeter · Independent open-source contribution

Engineering objective

System challenge

A single oversized AI skill would have mixed two different responsibilities: end-to-end commercial, campaign, branded-film, and multi-shot production versus individual-shot prompting, amendments, continuity, camera direction, typography, style, and audio. This creates ambiguous routing, competing instructions, and weak ownership.

Constraints

Working within real limits

  • ✕Preserve two explicit non-competing authorities without flattening instructions into one oversized file
  • ✕Commercial production authority must own the complete campaign or production workflow
  • ✕Shot-level prompting authority must own single shots, amendments, camera instructions, style transfers, typography, and audio tasks
  • ✕Routing discovery descriptions must be routing-sensitive without over-specifying
  • ✕Canonical source and installed skill trees must maintain 100% byte-identical parity (zero file-set and zero hash differences)

Architecture

Two-Authority Routing & Authority Separation Diagram

Bespoke Animated Routing ArchitectureCentral Router → Two Non-Competing Authorities

1. Creative Brief & Intent Router

Ambiguity filter resolves prompt ownership before execution

Path A: Commercial Production Authority

Campaigns, advertisements, multi-shot film strategy (10 scenarios)

Path B: Shot-Level Prompting Authority

Single shot, camera direction, style transfer, typography, audio (10 scenarios)

Parity & Installation Verification

Canonical Source to Installed Codex Skill Parity

Canonical Source & Installation Parity EvidenceByte-Identical Verification Across Skill Trees
Source of TruthCanonical Skill Tree2 Authorities Active
File-Set ComparisonZero Differences100% Tree Match
Hash VerificationZero Hash DiffByte-Identical
Codex InstallationInstalled Authority4x “Skill is valid!”
Structural Validations4 Successful Passes
Scenario Coverage20 Scenarios (10 Commercial + 10 Shot)
Relative References100% Resolved

Approach

How it was built

  1. 01

    Authority decision: Preserved two explicit authorities—commercial production and shot-level prompting—with a dedicated routing authority and source-of-truth document rather than flattening all instructions into one file.

  2. 02

    Audit and correction: Read the complete authority documents and both skill trees before making changes. Identified two shot-level test prompts written like full campaign requests and corrected them into unambiguous single-shot tasks. Strengthened discovery wording for individual shots, continuation, camera direction, style transfer, typography, audio sync, and targeted amendments.

  3. 03

    Validation and installation: Validated 20 routing scenarios—10 commercial-production and 10 shot-level—mirrored canonical skill trees into Codex discovery, ran structural validation returning "Skill is valid!" four times across canonical and installed copies, confirmed zero file-set differences and zero hash differences, resolved all relative references, and confirmed exactly two installed authorities.

  4. 04

    Governance model: Defined measurable validation criteria for authority ownership, routing separation, scenario coverage, reference integrity, and canonical-to-installed parity.

Technical validation

Implementation evidence

Observable engineering evidence. Discipline: PRODUCTION ENGINEERING

Non-Competing Authorities

2 Active

Routing Scenarios

20 Validated

Structural Validations

4 Passes ("Skill is valid!")

File-Set Differences

Zero (100% Parity)

Hash Differences

Zero (Byte-Identical)

Relative References

100% Resolved

Commercial Scenarios

10 Validated

Shot-Level Scenarios

10 Validated

Authority Ownership

Exactly 2 active authorities

Outcome

What was delivered

  • Separated commercial campaign orchestration from shot-level prompting into two non-competing canonical authorities

  • Validated 20 routing scenarios across 10 commercial-production and 10 shot-level prompting test cases

  • Structural validation returned "Skill is valid!" 4 times across canonical and installed copies

  • Zero file-set differences and zero hash differences verified between canonical source and installed Codex copies

  • Corrected ambiguous test prompts and resolved all relative references across both skill trees

Technologies

Stack and tools

Agent architectureSkill routingPrompt systemsAI governanceAuthority designValidationProduction workflows

Next engineering system

COCKROACHDB · OPEN SOURCE PR #173411

CockroachDB: Transactional Range Split Isolation

Have an ambitious AI system to build?

Let’s build something serious.

I work across AI architecture, product engineering, private inference, automation, SaaS infrastructure, and production hardening.

Discuss a projectConnect on LinkedInTechnical profile