Clearward

Mission

Ship structured organization systems for everyday life and small trades businesses without treating AI-generated workbooks as automatically trustworthy.

Problem

Templates become unmaintainable when inputs, calculations, and presentation collapse into the same cells, when identifiers are unstable, and when “it opened on my machine” is the only release check.

Constraints

• Cross-platform Excel and Google Sheets compatibility.
• No macros, no Power Query, no external data connections.
• Production artifacts strictly separate from internal engineering artifacts.
• AI coding agents operate under persistent AGENTS.md instructions, not ad hoc prompting.

My role

I directed AI coding agents (Codex, ChatGPT) with structured agent instructions covering permitted directories, product requirements, release procedures, browser and account boundaries, QA, packaging, customer-delivery, spending boundaries, credential handling, human-takeover, and verification rules.

Architecture

Structured tables with stable record identifiers, controlled input cells separated from formula-driven state, data validation, configuration tables, dashboards, and bounded formulas. Every release moves through automated QA, visual PDF review, human approval, packaging, and hashing.

Build process

1. Products evolved from templates into structured, software-like workbooks.
2. Pet Care System as a concrete pattern: 13 expected worksheets on 8 structured data tables.
3. Second office engine used to open, recalculate, and save files to surface compatibility issues.

Validation

Lessons

• Privileged automation, even for spreadsheets

  An AI agent should operate under an explicit authorization model, exactly like any other privileged automation — least privilege, explicit boundaries, mandatory human takeover on ambiguity, and verification instead of self-reported success.

Cybersecurity equivalent

• AGENTS.md control planemaps toAuthorization policy for privileged automation
• Human approval before packagingmaps toChange control / release authorization
• Package hashesmaps toIntegrity verification
• Internal vs customer artifactsmaps toInformation minimization

AI principles applied

• Human-in-the-loop

  Consequential actions require a person. The loop is a state machine, not a polite request in a prompt.

  Where learned: Publishing, packaging, containment, and deploy all use explicit gates.

  See in the AI map →

• Structured outputs

  Contracts the next stage can consume: HTML stages, morning reports, triage JSON, workbook tables.

  Where learned: If formatting can drift, downstream consumption silently breaks.

  See in the AI map →

• Validation and evaluation

  Happy path is not enough. Missing, malformed, stale, contradictory, injected, and timed-out inputs are the real suite.

  Where learned: Triage functional plus adversarial layers; Clearward 49 automated checks; Basaltborne CI layers.

  See in the AI map →

• AI-assisted development

  Agents draft under persistent instructions and phase gates. They do not merge, package, or deploy themselves.

  Where learned: Clearward AGENTS.md; Basaltborne phase-gated lifecycle with PR review of AI-authored changes.

  See in the AI map →

Security principles applied

• IAM / authorization

  Who can do what is assigned outside the model and enforced at the boundary that matters.

  Where learned: Basaltborne RLS and RPCs; Clearward human-only credential boundary; triage least-privilege account.

  See in the security map →

• Least privilege

  Separate collection, analysis, and remediation. Give the current task only what it needs.

  Where learned: Triage agent; publishing-agent allowlist; Clearward directory and spending boundaries.

  See in the security map →

• Validation

  Schema, hashes, required fields, RLS tests, and visual review — not the generator marking its own homework.

  Where learned: Clearward formula/package QA; Basaltborne Zod + pgTAP; TinyClaw source checks.

  See in the security map →

• Change control

  Publication, packaging, and deploy are privileged transitions with a human on the gate.

  Where learned: Clearward release architecture; Basaltborne draft vs published; Self Help human review.

  See in the security map →

• Secrets management

  Credentials are a human-only boundary. Service roles stay server-side. Prompts are not a vault.

  Where learned: Clearward AGENTS.md; Basaltborne server-only privilege.

  See in the security map →

Technologies

• Excel
• Google Sheets
• Open XML
• Codex
• ChatGPT
• Hash validation

Skills demonstrated

• Agent Instruction Design
• Prompt Engineering
• Workflow Decomposition
• Human-in-the-Loop Control
• Structured Data
• Formula Logic
• QA Automation
• File Packaging
• Hash Validation
• Release Management
• Secrets Handling
• Authorization Boundaries