-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathCITATION.cff
More file actions
28 lines (28 loc) · 2.23 KB
/
CITATION.cff
File metadata and controls
28 lines (28 loc) · 2.23 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
cff-version: 1.2.0
message: "If you use or reference this architecture, please cite it as below."
type: software
title: "Agent Knowledge Cycle (AKC)"
version: "2.0.0"
abstract: "A knowledge cycle for AI agents — one that grows with the people who shape it. AKC is a six-phase self-improvement loop (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: Research is signal-first, recurring decisions are promoted to rules, compliance is measured instead of re-audited manually, and dialogue is front-loaded because misaligned implementation costs more than the conversation that would have prevented it. 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, eight architecture decision records, three design-pattern skills, and a dependency-free Python reference implementation. v2.0.0 declares the cycle genre-neutral: the cycle is a mechanism, and content (behavioral patterns, domain expertise, or constitutional values) is the downstream project's concern. The security triplet that had sat in AKC through v1.x (Security by Absence, Single External Adapter, Untrusted Content Boundary) was extracted as genre-specific and no longer ships with AKC."
authors:
- family-names: Shimomoto
given-names: Tatsuya
alias: shimo4228
orcid: "https://orcid.org/0009-0002-6168-4162"
license: MIT
doi: 10.5281/zenodo.19200727
repository-code: https://github.com/shimo4228/agent-knowledge-cycle
url: https://doi.org/10.5281/zenodo.19200727
keywords:
- ai-agent
- self-improvement
- knowledge-cycle
- claude-code
- agent-knowledge-cycle
- memory-centric
- bidirectional-growth
- human-ai-collaboration
- signal-first
- cognitive-economy
- genre-neutrality
date-released: "2026-04-19"