Claude Code
Reads a Claude Code harness — instruction files (CLAUDE.md), auto memory, tools (MCP, skills), permissions and hooks — and draws it as one assembly map, by folder and by how new agents are spawned.
Rates a harness at levels 1–5 in six areas and shows them as a hexagon chart: instruction design, memory and location, tool scope, pre-blocking, post-verification and observability, and delegation design.
README (Korean), translated
Level 5 is reached only when there is a record of actual confirmation — for example, pre-blocking reaches 5 only if there is a record of a deny rule actually blocking a command.
README (Korean), translated
How to start
From the README (in Korean), “에이전트에게 맡기기”: paste this prompt into a coding agent such as Claude Code, and change the folder paths to the ones where you usually start work.
https://github.com/Dominic-DK/harness-map 의 README 를 읽고 "직접 실행" 순서대로 실행해 줘.
내가 자주 작업을 시작하는 폴더는 ~/work/a 와 ~/work/b 야.
3단계의 에이전트 점검은 안내에 적힌 안전 규칙을 지켜서 직접 수행하고 기록해 줘.
끝나면 만들어진 harness-map.html 을 브라우저로 열고, 결과를 쉬운 문어체로 다섯 문장 안팎으로 요약해 줘.
Running it directly needs Python 3 and no extra packages.
GitHub · MIT license · README in Korean · Repo created 2026-10-02
Claude Code
Keep lightweight subagents from getting dumber — load only the skills, plugins and MCP servers each project needs. A Tuist-style module manager and dashboard for Claude Code harnesses.
“200k models such as Haiku and Sonnet — the ones you typically spawn as subagents or claude -p workers — get about 8,000 characters. … On the author's machine the list was 89 skills and 33,688 characters, four times the budget.”
README
“On the author's machine, archiving unused user items and hiding the descriptions of 29 unused built-in and plugin skills brought a Haiku session's skill list back inside the budget — every skill still listed by name.”
README
How to start
Quick start (README):
git clone https://github.com/Dominic-DK/harnist && cd harnist
pip install pyyaml # the only dependency (or: uv run harnist.py ...)
python3 harnist.py demo --open
demo builds a fake workspace in a temp folder — projects, a registry, a global ~/.claude, session history and measurements — and opens the dashboard read-only. Nothing on your machine is touched.
GitHub · MIT license · README in English and Korean · Repo created 2026-10-06
MCP
askew-mcp is the local connector for Askew: an MCP server (stdio) that lets Claude Code, Claude Desktop, Cursor, Codex or any MCP client run Shortcuts on your iPhone, send you notifications, and read what your phone sends back.
Running Shortcuts on the iPhone needs a connector key from the Askew app, which is in preparation (README: “The Askew app is currently in waitlist”). Local action search and recipe build need no key; recipe build requires macOS.
How to start
Add it to Claude Code (README). The README also has configs for Claude Desktop, Cursor and Codex CLI. Requires Node 22 or newer.
claude mcp add askew -s user -e ASKEW_CONNECTOR_KEY=akc_XXXX -- npx -y askew-mcp
GitHub · npm · MIT license · Repo created 2026-09-17
MCP
Let any AI agent read your Mac. Calendar, Reminders, Notes, Contacts, Clipboard, Location, Weather, now playing — through Apple Shortcuts. Read-only, local, no account, no server. Works with Claude Code, Claude Desktop, Cursor, Codex, or any MCP client.
How to start
Setup (README):
npx -y mac-shortcuts-mcp setup # install the read recipes (once)
Then add it to your MCP client:
{
"mcpServers": {
"mac": { "command": "npx", "args": ["-y", "mac-shortcuts-mcp"] }
}
}
macOS only. The bundled recipes are read-only. mac_run_shortcut can run any installed Shortcut, including one that writes data or sends messages.
GitHub · npm · MIT license · Repo created 2026-09-19
LLM review
Two sealed loops that make LLM review produce evidence instead of opinion. Adversarial discovery until exhaustion, and reference-derived rubrics with a pass line declared up front.
GitHub · MIT license · Repo created 2026-09-07
