Arbora ecosystem
Build agents and connect source systems through Arbora, with Medulla as the durable knowledge engine. The ecosystem is developed inside Medulla as changesets, a code graph, and requirement records, served through Arbora MCP.
Start with work waiting to be built
The public specification report lists owner-native specs, the share of requirements each has delivered, and the requirements still open. Expand a specification to pick one. Each owner repository holds the full design, implementation tasks, and test contract. A status claim needs public evidence; publication alone does not mean implementation or acceptance.
Contribute with the shared skills
The path is spec-driven, from constitution to release. Each step names its skill and the files it writes.
- Constitution. Read the constitution. Every plan checks against it.
- Route. Pick the owner repository with the ecosystem boundaries and
arbora-development. Choose an open (SPECIFIED) row in Canopy. - Specify.
speckit-specifyandspeckit-clarifywritespecs/<id>/spec.md: user stories, Given/When/Then scenarios, andFR-andSC-labels mapped to row IDs. - Clarify. Resolve every
[NEEDS CLARIFICATION]marker inspec.mdbefore you plan. - Plan.
speckit-planwritesplan.md. Addcontracts/for cross-repository seams,data-model.md,quickstart.md, andresearch.mdwhen unknowns remain. - Tasks.
speckit-taskswritestasks.md. Each line reads- [ ] T001 [P] [US1] <ROW-ID>. - Analyze and checklist.
speckit-analyzechecks consistency and writeschecklists/requirements.md. - Implement.
sdd-implementerbuilds each task. Run the repository gates. - Land. Open a pull request whose commits carry a
Spec: <ROW-ID>trailer and a bound test. Do not editstatus.jsonby hand. - Release. Landed rows ship in the next release bundle.
Standards: constitution, spec standard, spec templates, and Canopy. Install the skills with pip install universal-skills, then install-skills --all-detected --symlink. Read the full contribution path.
Core projects
Seven core projects. Arbora is the one door: served operations, admission, authentication, and MCP and A2A serving. It hosts Overstory. Meristem is the in-run harness and holds no store. Medulla owns durable knowledge. Scion SDK holds connector contracts and transports. Tuig and Axil are operator surfaces on the same boundary.
- Arbora — the one door (formerly graph-os).
- Medulla — durable knowledge and reasoning; one engine with one server executable (
medulla-server) and themedullapackage. - Scion SDK — connector contracts and transports.
- Meristem — in-run agent harness.
- Overstory — browser experience.
- Tuig — terminal operator surface for the Arbora REST interface.
- Axil — desktop GUI.
Related contributor repositories
- Repository Manager — workspace lifecycle.
- Pipelines — shared CI and release workflows.
- Emerald Exchange — exchange integration.
- Tunnel Manager — tunnel lifecycle.
- Universal Skills — shared spec, development, architecture, and verification skills.
- Skill Graphs — distills knowledge into skills that Medulla imports natively, layering the ontology over each knowledge graph; generated by the skill-graph-builder skill, which can source content with the web crawler and web scraper skills.