Protocol Files

The protocol itself is a small set of markdown files at the repo root. They are written for an AI agent to execute, so they're terse and operational — this page is a map. Each links to the source on GitHub.

Always read

File	Role
AGENTS.md	The hub. Every vendor's agent reads it first; it routes to everything else. A target's AGENTS.md is the memory hub (from templates/AGENTS.md); the tool's root AGENTS.md is the dual-mode operator dispatcher.
README.md	Human-facing quickstart + design philosophy.

Read on demand

File	Role
DECAY.md	The authoritative decay rules — tiers, windows, what never decays, supersession.
REVIEW.md	The review ritual, step by step, with the safe-write Safety section.
MERGE.md	The tiered, human-gated memory/ conflict-resolution protocol.
SKILLS.md	The skills layer — authoring, syncing, deleting, the adapter recipe.

Operator-only

File	Role
ENABLE.md	The 10-step enablement flow (Modes A / B / C). Lives in the tool, never installed into a target.
MIGRATE.md	Vendor-file detection + promotion for Mode C.
UPGRADE.md	The version ladder + per-rung instructions + source-of-truth map.

Per-repo state (generated)

Path	Role
memory/continuity.md	The live projection — facts, decisions, open threads.
memory/vision.md	The north star (core, invariant-verified).
memory/sessions/	Immutable dated session logs — the event ledger.
memory/archive/	Faded facts + a greppable INDEX.md.
memory/decay-policy.md	This repo's tunable decay parameters.
.agent/version.md	Install manifest — gates in-place upgrades.
.agent/schema.md	The file-format reference for the memory layer.

Going deeper

For the why behind these, read the Whitepaper and the design notes under Background in the nav.