Seeds

Definition

A Seed is an initial context document that establishes governance ground truth at the start of an operational session or project. Seeds are append-only, human-authored, and represent the authoritative baseline from which all subsequent session context flows.

A Seed is not editable after it is committed to the corpus. If the governing reality it describes changes, a new Seed is authored and committed. The original Seed remains in the corpus as the record of what was true at the time it was written.

The Ground Truth Function

Seeds answer the question: what do we know, at governance level, before any work begins?

Typical Seeds include:

• The canonical system architecture document
• The approved technology decisions (ADRs)
• The organizational engineering standards that apply to this project
• Security and compliance baselines
• The scope and purpose of the project

These documents are authored by humans with the appropriate authority. They are not generated by the AI model. They are the governed inputs that the model consumes — not outputs it produces.

Immutability

The immutability of Seeds is a governance property, not a technical limitation. Seeds could be modified — the corpus allows new records to be written. But the governance convention is that Seeds are committed at project initialization and replaced only by explicit, versioned new Seeds when the governing reality changes.

This immutability is what makes Seeds reliable as a baseline. A model consuming a Seed can rely on it as the authoritative record of what was true when it was written.

Seeds vs. Packs

Packs provide domain-specific operational context. Seeds provide governance ground truth.

The difference: Seeds are project-level or org-level governance documents. Packs are domain-level technical context documents. A project might have one set of Seeds (the project charter, the architecture document) and apply multiple Packs over time as work moves across domains.

In Yanzi

Seeds are authored as markdown documents and committed to the Yanzi corpus with the yanzi context seed add command. They are loaded into session context during deterministic context composition before any domain Packs or session-specific context is applied.