Exobrain (the repo itself)

Rendered from knowledge/exobrain/README.md — the text an agent reads.

Not a product. A meta-domain documenting how an exobrain repo is structured and how its mechanisms (scopes, skills, tools, propagation) work — written so an agent can read it and either run an existing exobrain or build a new one.

TL;DR

An exobrain is a knowledge base plus an agent connector. It holds two kinds of content — knowledge domains (current truth, kept current) and workspaces (time-bound efforts that outdate by design) — and connects coding agents (Claude Code, OpenClaw, Codex) to them via scripts/connect-agent.sh. Context is organized in scopes that overlay innermost-wins: global < [group] < person < host. A person needs no group; groups appear only when more than one person shares context. Skills are the unit of agent capability — directories declared in a scope’s skills.json and surfaced per tier (always / optional / unlisted / off).

Crucially, an exobrain is one implementation of a shared concept, not a fork of shared code. New capabilities propagate as patterns an agent re-synthesizes locally, never as merged code. See propagation.md.

Index

File	Covers
entities.md	The content types — knowledge domains, workspaces, people/groups — and the “promote, don’t link” rule
scopes.md	The scope stack (global < [group] < person < host), resolution, overrides, top-level person
agents.md	Agent-agnostic content — per-agent sidecars, markers, injection surfaces, non-interference
skills.md	The skills system — registry, tiers, resolution, linker, fetcher, agent filtering, authoring craft
tools.md	External tools — the per-tool-doc catalog under tools/, scope overlays/resolution, .exobrain.json state
domains.md	How to structure a knowledge domain — directory layout, scope design, WIP (.wip) domains, the TIMELINE.md convention
authoring.md	How to write well-formed knowledge domains — horizon, current-state, synthesis, citations
grill.md	The interview discipline — distilling and curating knowledge through adversarial, one-question-at-a-time interrogation
propagation.md	How exobrains share improvements — the read-target/feed model, the borrow workflow, invariants
mounts.md	Mounts — shared, content-only knowledge repositories read from a local checkout, each with a charter (audience, purpose, held domains, what never goes there): the index, routing, the isolation gate, freshness
landing.md	How a change reaches the default branch — attended and detached lands, the sweep, who repairs authoring findings, and how to check on a land
security.md	The security posture and the findings ledger — per-scope security.json registries, recording and accepting a finding, how the ledger reaches a person, the review
compat.md	The compatibility-shim ledger — how transitional code is marked, dated, and retired
machinery.md	Index of the concrete machinery — every script, git/session hook, registry, and quality gate, with the verify-a-connector-change flow

For an agent instantiating a new exobrain

Read propagation.md → Instantiating. The short version: these files describe principles, not a rigid layout. Choose the structure that fits your principal (with/without groups, which scopes are populated). Preserve the invariants (security, scope-resolution semantics, the validation contract); adapt everything else.