CAK R&D exists to discover agent-native software abstractions: forms of skills, procedural memory, contracts, verifiers, traces, and governance that fit AI agent failure modes instead of copying human software defaults by habit.
R&D precedes architecture for major design decisions. Implementation should follow only after a Research Decision Record (RDR), a small experiment, or an explicit implementation spike identifies what should be built and what should not be built.
Current status: CAK has executable spikes such as the experimental ContractSpec checker MVP, but those spikes are research artifacts until they are validated against concrete questions, alternatives, and kill criteria.
Start here:
- R&D Charter
- Research Run Protocol
- Source Ledger template
- Debate Protocol
- Research Quality Gate
- Pattern References
- Skill Architecture Deep Research Synthesis
- Research Runs
- Research Decision Record template
- Research Question Map
- Literature Anchors
RDRs are downstream of research runs. A useful RDR compresses source discovery, evidence tracking, pattern extraction, adversarial review, debate, quality gates, unknowns, minimal experiment candidates, and kill criteria.
Pattern references are idea sources, not dependencies. They can guide source discovery and pattern extraction, but they do not support claims until inspected and recorded in a source ledger.
Research runs should search for stronger analogues beyond the seed list. The seed references are starting points, not a boundary around the research space.
The skill architecture synthesis captures the current working hypothesis that agent-native skills may be compiled bridges between evidence/provenance IR and runtime-control IR. It is a synthesis note, not a decision-grade RDR.