skill
Claude-Code-Base-Harness
A curated Claude Code configuration harness — rules, skills, agents, hooks, an Executive output style, and a base UI. Plain files: copy what you need into ~/.claude/ or a project's .claude/. No plugin
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/MaGe1993/Claude-Code-Base-Harness itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 23, 2026
Rules · Skills · Hooks — copy what you need, adapt and go.
%%{init: {'theme':'base','themeVariables':{'primaryColor':'#2FBF5E','primaryTextColor':'#ffffff','primaryBorderColor':'#1E9447','lineColor':'#0D9488','fontSize':'14px'},'flowchart':{'curve':'cardinal'}}}%%
flowchart TD
classDef product fill:#0284C7,stroke:#0369A1,color:#fff,font-weight:bold
classDef content fill:#2FBF5E,stroke:#1E9447,color:#fff
classDef event fill:#0D9488,stroke:#0A7A6F,color:#fff
classDef store fill:#0f172a,stroke:#2FBF5E,color:#a5b4fc
classDef install fill:#1e293b,stroke:#334155,color:#94a3b8
classDef zone fill:#0f172a0F,stroke:#334155,color:#94a3b8
harness(["📦 Claude-Code-Base-Harness — what's inside"]):::product
harness --> skills[["🛠️ Skills ×26"]]:::content
harness --> rules[["📏 Rules ×8"]]:::content
harness --> agents[["🤖 Agents ×3"]]:::content
harness --> hooks{{"⚡ Hooks ×8"}}:::event
harness --> outstyle["🎭 Executive style"]:::content
skills --> baseui[("🎨 base-ui CSS")]:::store
subgraph od["▸ loaded on demand"]
skills
rules
agents
end
subgraph wired["▸ wired via settings.json"]
hooks
outstyle
end
class od,wired zone
harness -.->|"download & copy"| install[/"⬇️ ~/.claude or .claude"/]:::install
linkStyle default stroke:#94a3b8,stroke-width:1.5px
linkStyle 0,1,2,3,4 stroke:#2FBF5E,stroke-width:2px
linkStyle 5 stroke:#0D9488,stroke-width:2px
linkStyle 6 stroke:#334155,stroke-width:1.2px,stroke-dasharray:4\,3
What is it? · What you get · Installation · Support · License
A curated, opinionated set of Claude Code configuration artefacts — rules, skills, agents,
hooks, output styles, and a base UI — that gives any project a correct Claude Code setup
without authoring it from scratch. No plugin, no install: plain files you copy into
~/.claude/ (all projects) or a project's .claude/, then adapt.
🛠️ Skills
| Skill | What it does |
|---|---|
audit-legal | EU AI Act / GDPR / German-law compliance triage |
audit-security | OWASP Top 10:2025 source-code security audit |
echarts | Apache ECharts 5 charts in base-ui HTML |
explain | Depth-calibrated explanations, simple or deep |
file-changelog | Author CHANGELOG.md (Keep a Changelog 1.1.0) |
file-claude | Author and audit CLAUDE.md; migrate auto-memory into it |
file-license | Author LICENSE and third-party attribution |
file-project | Author PROJECT.md — the AI-facing architecture spec |
file-readme | Author README.md — relevance-ordered, visual, GitHub-native |
frontend | Standalone HTML via the base-ui design system |
hooks | Author lifecycle hooks with companion scripts |
lean-review | Read-only quality review of any artifact |
loop | Set up and run recurring Claude Code loops |
masterprompt | Build a maximal expert masterprompt on any topic |
mcp | Author and repair .mcp.json across all transports |
mermaid | Generate styled, GitHub-renderable diagrams |
new-project | Scaffold a new Claude Code project end to end |
outputstyle | Author output style files; ships the Executive style |
proofread | Pre-ship verification of a finished deliverable |
rules | Author rules; routes to hooks or settings when those fit better |
settings | Configure permissions, model selection, and env vars |
skills | Author and maintain skills from a need or workflow |
sota | Grounded state-of-the-art web research |
subagents | Author specialist subagent definition files |
tasks | Guide to the Claude Code task system |
toon | Convert JSON or YAML to token-efficient TOON |
📏 Rules
| Rule | What it covers |
|---|---|
command-execution (always on) | Claude runs commands itself; never asks you to |
error-handling | No silent catches, fail-fast, backoff, structured logging |
json-style | JSON and JSONC conventions |
md-style | Markdown conventions for harness files |
security | Credential handling, input validation, injection, supply chain |
toon-style | Token-efficient tabular data for LLM prompts |
visual-verify | Screenshot-verify visual output; no code-only sign-off |
yaml-style | YAML and frontmatter conventions |
⚡ Hooks
Each hook ships as a .ps1 + .sh pair behind one dispatch.mjs launcher — Windows runs
the .ps1, macOS/Linux the .sh, nothing to swap by hand.
| Hook | Event | What it does |
|---|---|---|
check-credentials | PreToolUse | Blocks writes containing detected keys, tokens, or secrets |
check-memory | SessionStart | Detects auto-memory files, prompts migration into CLAUDE.md |
cleanup-sessions | SessionStart | Prunes stale session artifacts under ~/.claude/ |
dangerous-cmd-guard | PreToolUse | Blocks catastrophic shell commands on both shells |
format-on-write | PostToolUse | Auto-formats written files with installed formatters |
post-compact | SessionStart | Re-injects PROJECT.md into context after compaction |
pre-compact | PreCompact | Blocks compaction only when PROJECT.md is missing or stale |
session-cost-logger | Stop · PreCompact · SessionEnd | Logs token usage per turn |
🤖 Agents
| Agent | Role |
|---|---|
ag-cqo | Chief Quality Officer — audit and Approved/Rejected stage-gate |
ag-cso | Chief Strategy Officer — outward state-of-the-art assessment |
ag-cto | Chief Technology Officer — writes, verifies, and self-reviews code |
🎭 Output style
| Style | Mode |
|---|---|
Executive | Bottom-line first, terse, decision-ready |
🎨 Base UI
| Asset | What it is |
|---|---|
base-ui | Green-blue glassmorphism CSS for AI-generated HTML, no build step |
🚀 Quick start — terminal
Fastest path — clone and copy into your project:
git clone https://github.com/MaGe1993/Claude-Code-Base-Harness.git
cp -r Claude-Code-Base-Harness/.claude Claude-Code-Base-Harness/CLAUDE.md your-project/
Then open a Claude Code session in your project and ask:
Read
CLAUDE.mdand.claude/skills/skills.index.toon, then tell me what's available.
[!TIP] Adapt
CLAUDE.mdto your project — fill in your build commands and paths, remove what does not apply. Everything else works as copied.
🛠️ Manual install — no terminal
- On the GitHub page, click Code → Download ZIP and extract it.
- Choose the scope:
- One project only → the project's root folder.
- All your projects →
~/.claude/(your home.claudefolder).
- Copy the extracted
.claudefolder and theCLAUDE.mdfile into the target with your file explorer. - Open a Claude Code session there and ask it to read
CLAUDE.md.
[!IMPORTANT] If a
settings.jsonalready exists in the target, merge thehooksblock manually — each hook event is an array; append rather than overwrite the whole file.
Every artifact is standalone. Copy individual pieces, then trim CLAUDE.md to reference
only what you copied:
| Want | Copy |
|---|---|
| Just the skills | .claude/skills/ |
| Just the format rules | .claude/rules/md-style.md, json-style.md, yaml-style.md |
| Just the output style | .claude/output-styles/executive.md + the outputStyle key |
| Just the hooks | .claude/hooks/ (incl. dispatch.mjs) + the hooks block |
| Just the base UI | .claude/skills/frontend/assets/base-ui/ |
Pull with git pull origin main (or re-download the ZIP). Check CHANGELOG.md
for breaking changes, re-read the skill and agent index files in open sessions, and merge any
new CLAUDE.md routing rows or settings.json hook registrations manually — do not overwrite
your customizations.
If this harness saves you time, a GitHub star helps others find it.
Enjoying it? Buy me a coffee on Ko-fi — it keeps the harness growing.
MIT — © 2026 Dr. Marius Gensler. See LICENSE.