Kads — Development skill for Claude Code
KADS (KRUZZZZY's Agent Delegation Surface): assign any coding agent to each role (find, triage, fix, review, integrate) and let it run.
How to install Kads
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open KRUZZZZY/kads and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Kads does
KADS (KRUZZZZY's Agent Delegation Surface): assign any coding agent to each role (find, triage, fix, review, integrate) and let it run. Mechanical checks, negative controls, evidence-only review, merge trains.
Alternatives in Development
- MCP Builder — Guide for creating high-quality MCP servers to integrate external APIs and services 94.1k ★
- Claude Integration Guide — FOR CLAUDE CODE: When a user asks you to integrate components from this showcase repository into their project 9.3k ★
- Mono Color Skill — One-ink editorial print image skill — warm paper, halftone photography, active negative space, and restrained 3k ★
README
KADS: KRUZZZZY's Agent Delegation Surface
**An automated delegation surface for coding-agent harnesses.** You define the roles (finder, triager, fixer, reviewer, integrator), point each role at whichever agent harness and model you want, and KADS runs the work between them unattended: queue, routing, budgets, escalation, mechanical checks, independent review and merge trains. A human is needed only when every agent on the escalation ladder has failed a job.
It does not contain a model or an agent of its own. It is the control plane *around* agents. Today it drives Claude Code, OpenAI Codex, OpenCode, MiniMax Code, a DeepSeek harness and local Ollama models side by side, each as one or more "lanes" with its own quota, pacing and failure handling.
KADS began as the fix pipeline behind Zene Studio and still runs it.
**Status: reference implementation, not yet repo-agnostic.** It was built for one repository, [Zene Studio](https://github.com/KRUZZZZY/zene-studio) (a C++/Qt DAW). Paths, build and test commands, gates and lane definitions are still wired to that repo and one workstation. The goal is to change that: open KADS, pick an agent per role, point it at a repo, and get a bug tracker and queue for that repo. The work towards that is tracked in [issue #1](https://github.com/KRUZZZZY/kads/issues/1). Until it lands, read this as a working design with working code you can lift from.
What the surface does
┌──────────── roles (each delegated to any harness + model) ────────────┐
repo ──► │ Finder ──► Triage ──► Queue/Router ──► Fixer ──► Checks ──► Reviewer │ ──► Merge train ──► new base
│ read-only dedup, difficulty, own submit, evidence- │ build + test all
│ sweeps realness, urgency, worktree negative only │ together; doctor
│ ranking per-model stats control verdicts │ agent on repeats
└─────────
Related Skills
Auditor Executor Protocol
A two-role Auditor/Executor protocol for multi-phase agent-executed work, plus a small CLI (auditkit) that sca
Tmux Master
Tmux master skill to allow claude code and codex to play the orchaster role to fork tmux session reversed for
Bug Triage
Classify bugs, assign severity and priority, and propose next actions.
Readable
Render the Readable Surface — what this system IS, organised by worry, in plain language, for the reader who d
Excel Hygiene
A 'second pair of eyes' Claude Code skill for any .xlsx — deterministic mechanical checks (openpyxl) plus Pank
Dsh Plugin Product Subagents
Role-based Codex / Claude Code / ACP subagent providers for the DeepSeek Harness — continuable children, durab
Related Agents
Zp Surface Probe
ZeroProtocol active surface probe for ONE host. Use to find out what is actually alive on a single in-scope ho
Socratic Questioner
Mandatory pre-implementation interrogation. Use before any non-trivial coding task to surface assumptions, fin
Questioner
Mandatory pre-implementation interrogation. Use before any non-trivial coding task to surface assumptions, fin