Agent Knowledge Cycle (AKC)

A knowledge cycle for AI agents — agent behavior compounds, human judgment sharpens. AKC is a six-phase knowledge cycle (Research → Extract → Curate → Promote → Measure → Maintain) that keeps skills, rules, and documentation aligned with reality over time. Unlike one-directional optimization frameworks, AKC is bidirectional: as the human curates and promotes knowledge, their judgment about what makes good agent behavior also sharpens. AKC treats human cognitive resources — attention and judgment — as the central constraint that does not scale with the model, and shapes every phase to protect that budget. AKC is not a harness — it runs on top of harnesses (like Everything Claude Code) to keep them coherent as understanding evolves. AKC ships as specifications, JSON schemas, architecture decision records, design-pattern skills, and a dependency-free Python reference implementation. The cycle is genre-neutral: it is a mechanism, and content (behavioral patterns, domain expertise, or constitutional values) is the downstream project's concern.