docs/

General design documents and reusable principles for LLM/Agent systems.

Use this directory for documents that answer:

docs/ is intentionally flat at the beginning. Create subdirectories only when several real documents accumulate around the same topic.

Current documents

Start here:

Agent Adoption Guide — Low-context protocol selection, reading order, minimal bindings, and bootstrap record.
Protocol manifest — Machine-readable versions, maturity, level namespaces, dependencies, artifacts, and evidence.

Practiced protocols:

Agent-first Memory Architecture — practiced; portable memory lifecycle with trust calibration, reversible withdrawal, and explicitly authorized erasure.
Agent-first Auto-Walk Architecture — practiced; exploratory association over consolidated memory or any structured corpus.
Agent-first Skill Lifecycle Architecture — practiced; imported and relationship-formed procedural artifacts share evidence-, authority-, version-, and rollback-aware lifecycle rules.
Agent-first Council Architecture — practiced; single-agent and multi-agent deliberation, revision, evaluation, and termination.
Agent-first Steward Architecture — practiced; the 1:1:N relationship and coordination protocol.

Design-only protocols:

Agent-first Active Workspace Architecture — design-only; short-lived, source-aware task state for calibrating current attention, with an optional audited-completion profile.
Agent-first Inner Speech Architecture — design-only; bounded runtime self-guidance, including the Inside-Outsider stance.

Advanced:

Composable Agent Cognition Protocols — design-only; composition rules for explicit artifact exchange. Read after selecting two or more protocols.
Governed Shared Memory Composition Profile — design-only; perspective-preserving shared standing over existing artifacts with declared audience and admission authority.
Governed Artifact Portability and Recovery Guide — design-only; backup, restore, and cross-binding migration semantics without copying authority or runtime state.
Governed Artifact Replication and Exchange Guide — design-only; ongoing one-way replication and producer-attributed append-only exchange without creating another writer.

Memory use cases — Practical implementations of the memory architecture.
Auto-Walk use cases — Practical implementations of Auto-Walk.
Skill Lifecycle use cases — Imported and relationship-formed procedures with explicit lifecycle evidence and gaps.
Council use cases — Same-runtime sub-agent and heterogeneous-model Council bindings.
Steward use cases — Real 1:1:N Steward bindings coordinating participants and managed resources.