KCC FX

The KCC Agentic Framework

The doctrine has a runnable counterpart: a portable, harness-neutral spec-driven framework that turns the operating model into agents, skills, and a lifecycle you can execute.

FrameworkRepositorySpec-drivenHarness-neutral
Created 2026-06-08 · v0.4.0

What the repository is

KCC is doctrine; the kcc-agentic-framework repository is the runnable form of it. It is a portable, harness-neutral spec-driven workflow framework: the same kernel, capabilities, and lifecycle described here, packaged as agents and slash-command skills that a coding harness can execute.

Harness-neutral means one neutral source generates outputs for several runners. The same capability source produces Claude Code agents and skills, Codex / OpenCode `AGENTS.md` catalogs, an OpenCode configuration, and a generic runner manifest - so the doctrine is not locked to one vendor.

One kernel. Many capabilities. Many cells - now with a runnable harness for each.

What is inside

The framework mirrors the three-layer architecture. The kernel and capabilities live under a neutral source tree; generated harness files are produced from them rather than hand-edited.

AreaContents
KernelAgent contract surfaces, lifecycle gates, decision-trace format, maturity criteria, governance protocols
CapabilitiesReusable agents and skills (interrogators, architect, planner, implementer, verifier, butler, token guard)
AgentsThe roles that do the work, each with a declared contract and model class
SkillsThe slash commands that orchestrate agents across the lifecycle
Memory + TracesPersistent memory store and per-run decision traces (the cognitive substrate)

The agents you get

  • Interrogators - idea, technical, security, infrastructure, and UX/UI interrogators that produce typed decision briefs
  • Architect + Architecture Critic - produce and review ADRs, C4 and flow diagrams, guardrails, and quality gates
  • Planner + Implementer - turn specs into dependency-waved plans and then into code and tests
  • Verifier - checks the implementation against every acceptance criterion and writes a PASS/FAIL report
  • Meta-agents - the Butler (memory + trace custody) and Token Guard (budget estimates)

How it connects to the doctrine

Every agent in the framework implements the nine contract surfaces. Every expensive step passes a cost envelope gate. Every important decision leaves a decision trace, and low confidence trips the human gate. The framework is the doctrine made executable - see Using the Framework to run it.