Ravenstack Keep

The live Keep is ui-v2/ — see AGENTS.md for how it's actually served, how to run it locally, and the box update/deploy loop. Old ui/ (48×48 tiles) is frozen — do not add to it or port code from it into ui-v2/; see AGENTS.md's Boundaries section.

A visual command layer + progressive agent forge + multi-model Round Table for a personal AI operations fortress (OpenClaw / ReClaw).

This is an open collaboration space. The substrate (gateway, ReClaw 2.0, Ollama, Ravenstack vault, live RAG, working handoffs) already exists. This repo holds the design, Agent Specs, Keep MCP plans, and task board so multiple AIs (and humans) can review, push back, and build together.

Start Here

For every AI (read in order)

1. AGENTS.md — Mandatory first read: current architecture, what's live vs. frozen, how to run and deploy the real thing.
2. RAVENSTACK-KEEP-BLUEPRINT-v0.2.md — Living design brief (historical context for why; AGENTS.md is the source of truth for current state).
3. reviews/ — Shared multi-AI review inbox. Leave your findings here so others can read them.
4. Phase 1 artifacts (historical — Phase 1 shipped; kept for design context):
   • docs/AGENT-SPEC-TEMPLATE.md — Agent Spec template (kill condition mandatory).
   • schemas/agent-spec.schema.json — JSON Schema for specs.
   • agents/oracle.md — First real Agent Spec (Oracle); JSON twin agents/oracle.agent-spec.json.
   • mcp/README.md + mcp/tools.md — Keep MCP Phase-1 contract (five tools; skeleton only).

Latest session handoff

• reviews/findings/2026-07-29-grok-phase1-session.md — What Grok built, Round Table feasibility, improvement backlog, next steps.
• reviews/INDEX.md — Index of all AI findings (newest first).

Status

Phase 1 artifacts landed (2026-07-29): Agent Spec template + schema, Oracle draft spec, Keep MCP tool contracts, multi-AI reviews/ folder. Ready for cross-AI review. No production Keep MCP server yet — contract only.

How to participate (AIs)

1. Read the entire v0.2 blueprint.
2. Read reviews/README.md and the latest findings under reviews/findings/.
3. Review Phase 1 files under docs/, schemas/, agents/, mcp/.
4. Write your own finding using reviews/findings/TEMPLATE.md; add a row to reviews/INDEX.md.
5. Answer open questions (blueprint §12 + open items in findings); claim concrete actions from blueprint §13.
6. Prefer small, scoped, value-producing steps. Local-first. Kill conditions mandatory. Human gates permanent. No draft-to-execute.

Key Links

• Primary OpenClaw skills source: https://github.com/VoltAgent/awesome-openclaw-skills
• ClawHub: https://clawhub.ai/
• Roundtable.sh (multi-model council): https://roundtable.sh/
• Agent Mind Bridge: https://github.com/brendanlucas01/Agent-Mind-Bridge
• Phase 1 PR: #1

---

Operator: Jason (Ravenstack / ReClaw)
Partner: Grok
License: TBD (currently open for collaboration)