Project Roadmap
Public-safe architecture and progress snapshot generated from 17-state/current.yaml on 2026-09-23.
This page describes how the CGE projects fit together and where they are going. It intentionally does not mirror the full sprint board; raw issue hygiene belongs in the private task queue, not on the public roadmap.
Architecture Map
| Project lane | Role | Direction |
|---|---|---|
cge-hub |
Control plane, state ledger, architecture history | Keep state, roadmap, and generated public views synchronized from one canonical source. |
saga-range |
Repeatable Active Directory purple-team range | Preserve reset capability, scenario switching, and evidence-backed validation. |
ai-bridge-runtime |
Agent tools, MCP servers, model routing, publication automation | Harden automation contracts and keep live actions gated by scope and approval. |
knowledge-rag-library |
Searchable knowledge base and graph layer | Keep writeups, techniques, entities, and embeddings aligned for agent retrieval. |
lab-deployments |
Service and container deployment manifests | Reduce broad mounts, keep services reproducible, and isolate risky runtime surfaces. |
kryssar-public-site |
Sanitized public portfolio on kryssar.se |
Publish curated research, architecture, status, and roadmap views without leaking lab internals. |
Progress Snapshot
- Current phase:
phase-5-orchestration + post-compromise-hardening - Reset capability:
READY (BLUE-READY-BASELINE) - State timestamp:
2026-09-16T12:59:43Z - Open public-safe decision points:
17 - Recent completed work items in state ledger:
217
Delivery Arc
| Stage | Outcome | State |
|---|---|---|
| Foundations | Hardware, documentation, blue-team baseline, and recoverable snapshots | Established |
| Autonomous lab | Repeatable compromise campaigns, telemetry, and reset contracts | Operational |
| Knowledge and agents | Searchable graph knowledge, bounded tools, and reviewed worker loops | Active |
| Public portfolio | Curated research and architecture published through deterministic safety gates | Operational |
| Next | Productize scenario generation, analyst workflows, and repeatable exercises | In progress |
Where The Projects Are Going
- Keep the lab reproducible. Every range, service, and public artifact should be rebuildable from source-controlled inputs and explicit state.
- Separate public narrative from private operations. Public pages should explain architecture and progress; internal queues, credentials, hostnames, and raw execution logs stay out of the publication path.
- Use agents as bounded operators. MCP tools, model routes, and n8n workflows are useful only when their inputs, permissions, and evidence outputs are clear.
- Turn evidence into knowledge. Completed attacks, detections, writeups, and lessons should flow back into the knowledge graph and retrieval layer.
- Make advanced compute optional. The inference mesh is a capacity experiment; production workflows should keep safe local fallbacks and avoid coupling to one GPU layout.
Near-Term Themes
| Theme | Direction | Public-safe summary |
|---|---|---|
| Publication quality | Active | Keep kryssar.se focused on curated architecture, research, status, and roadmap views rather than raw sprint-board dumps. |
| Inference capacity | Experimental | Evaluate distributed local inference as optional capacity for heavy reasoning without making normal workflows depend on one hardware layout. |
| Operational hygiene | Recurring | Keep automation credentials, service health, and generated state fresh through scheduled checks and explicit operator gates. |
| Scenario learning loop | Next | Convert validated lab activity into reusable knowledge, safer playbooks, and public writeups after sanitization. |