--- domain: custodian repo: the-custodian updated: "2026-05-17" --- # INTENT > This file explains why the-custodian exists, what role it plays in the > ecosystem, and what it must not absorb as neighboring repositories mature. --- ## Why it exists The Custodian exists to hold the long-lived governance, memory, and coordination substrate for a local-first agent ecosystem spanning multiple project domains. It gives Bernd and trusted agents a stable place to preserve values, constitution, domain charters, workplans, and cross-domain orientation without turning any single implementation service into the source of identity. Its deeper purpose is stewardship: keeping the system coherent across years of tool changes, repo splits, agent sessions, and domain growth. --- ## The governing principle the-custodian owns **meaning, boundaries, and continuity**. It should answer: 1. **What matters?** Values, constitution, charters, and governance constraints. 2. **What is being pursued?** Workplans, domain roadmaps, and coordination notes. 3. **What must be remembered?** Append-only memory, decisions, provenance, and session handoff context. It should not become the implementation home for every service it coordinates. When a subsystem becomes operational code with its own runtime, tests, and deployment surface, it should live in its own repository and report back through State Hub and workplans. --- ## What it is the-custodian is the **governance and continuity substrate** for the Custodian ecosystem. It contains: - canon: constitution, foundational values, standards, domain charters, concept seeds, and roadmaps - memory: working notes and episodic logs that preserve session continuity - workplans: repo-backed plans for Custodian-owned coordination work - runtime scaffolding: agent policies, prompts, and integration guidance - integration pointers: lightweight references to operational services such as the standalone State Hub --- ## What it is not | Concern | Owner | |---|---| | Live State Hub implementation, migrations, dashboard, tests | `state-hub` | | Event-triggered maintenance task creation | `activity-core` | | General task lifecycle backend | `issue-core` | | Repository capability profiling | `repo-scoping` | | Domain-specific products and experiments | Their domain repositories | | External publication, contracts, payments, or legal authority | Human approval only | The repository may describe or coordinate these systems, but it should avoid reabsorbing their runtime code. --- ## What it enables When this repository is doing its job, a human or agent can: - understand the purpose and rules of the whole ecosystem without reading every implementation repo - start a session with clear governance and State Hub handoff rules - trace work from values and domain charters into active workplans - preserve decisions, provenance, and memory in durable, reviewable files - split operational subsystems into focused repos without losing continuity --- ## Design values **Local-first continuity.** The core governance record must remain useful even when services are offline or external vendors change. **Human-approved authority.** The Custodian can draft, coordinate, and remember, but irreversible decisions remain human-gated. **Small operational surface.** This repo may coordinate services, but live services should own their own code, tests, and deployment paths. **Append-only memory by default.** Session and episodic records should be preserved rather than silently rewritten. **Reviewable canon.** Canon changes should be proposed with provenance and review, not silently applied as operational side effects.