AGENTIX · ENGINEERING UPGRADEPART 04 / KNOWLEDGE
Engineering Upgrade Knowledge
§ Part 04 · knowledge that compounds

Stop re-explaining
your domain.

Generic AI forgets your codebase between prompts. The framework gives it typed standing documents, domain rules, ADRs, compliance guardrails, module contracts, layered so the right knowledge loads at the right moment. New developers ramp in days, and the domain gets sharper with every feature instead of being re-explained each time.

FIG.01, the knowledge hierarchy · context narrows as work narrows
Project The product and its standing truth, ADRs, domain rules, compliance guardrails, glossary.
Epic A body of work with its own goal and rollup of state from the features beneath it.
Feature The unit of delivery, its requirement, plan, design, stories, PR and cost ledger in one folder.
Story A vertical slice with Given/When/Then acceptance criteria, built one at a time.
An agent working a story loads its story file, the feature plan, the epic goal and the project's standing rules, in scope, nothing more. Context scales with the work, not with the size of the repo.
§ Typed standing documents

ADRs

Architecture decisions, captured once and cited by every agent, so the same trade-off is never re-litigated.

Compliance guardrails

PHI rules, tenant isolation and module boundaries, written as enforced standing documents, not wiki prose.

Module contracts

What each module owns and exposes, so agents respect boundaries instead of reaching across them.

Domain rules + glossary

The language of the business, encoded once, so "encounter" or "care plan" means the same thing everywhere.

"Most prompts need 5% of the context but pay for 100%."

Scoped context is the lever, the wrong context hallucinates, too much context wastes tokens. Loading exactly the standing documents a task needs is what makes the framework cheaper and more accurate at the same time.

§ Six rules enforced at the framework level, not the wiki
PHI protectionTenant isolationModule boundariesCQRS disciplineDB migration safetyStructured logging
FIG.02, the feature folder · every artifact for one feature, in one place
care-platform / feature tree
docs/epics/<epic>/features/<slug>/
├─ feature.md # phase tracker + cost rollup
├─ requirement.md # approved scope
├─ plan.md # architecture + file checklist
├─ design.md # screens + states (UI features)
├─ stories/01…NN.md # vertical slices, in order
├─ pr.md # review verdicts + PR link
├─ .cost.jsonl # per-agent token ledger
The folder is the feature. State rolls up bottom-up, story → feature.md → epic, so the overview is always current without anyone updating a board. This is the ledger that ROI reporting → reads.
Days to ramp a new developer, not months, the domain is in the repo, not in someone's head.
Every feature sharpens the knowledge base, decisions captured once compound instead of being re-explained each prompt.