https://github.com/pr0ck0/ai-config
Agent config for ontology-driven LLM development. One source, minimal surface, secure by default.
https://github.com/pr0ck0/ai-config
agents agents-md ai-config claude-code codex cursor developer-tools gemini knowledge-graph llm ontology openai owl rdf secure-by-default sparql tdd
Last synced: 2 months ago
JSON representation
Agent config for ontology-driven LLM development. One source, minimal surface, secure by default.
- Host: GitHub
- URL: https://github.com/pr0ck0/ai-config
- Owner: PR0CK0
- License: mit
- Created: 2026-04-03T19:09:01.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-04-03T19:41:29.000Z (3 months ago)
- Last Synced: 2026-04-04T13:51:02.264Z (3 months ago)
- Topics: agents, agents-md, ai-config, claude-code, codex, cursor, developer-tools, gemini, knowledge-graph, llm, ontology, openai, owl, rdf, secure-by-default, sparql, tdd
- Language: Shell
- Size: 39.1 KB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Agents: AGENTS.md
Awesome Lists containing this project
README
# ai-config
> Agent config for ontology-driven LLM development. One source, minimal surface, secure by default.
[](LICENSE)
[](https://claude.ai/code)
[](https://cursor.sh)
[](https://github.com/google-gemini/gemini-cli)
[](https://github.com/openai/codex)
One file. One source. No drift.
---
## Philosophy
Rules here are **organizational scar tissue** — grown from real incidents, not upfront design. A rule gets added the second a mistake repeats. Nothing is added speculatively.
- **One source.** Edit `AGENTS.md` only. `build.sh` generates all tool-specific aliases.
- **Minimal surface.** Short rules shape judgment. Long rules get ignored — and the middle ones get ignored most. Research shows LLMs reliably attend to the beginning and end of long contexts, with significant degradation in between ([Liu et al., 2023 — "Lost in the Middle"](https://arxiv.org/abs/2307.03172)). Keep `AGENTS.md` short and front-load what matters.
- **Ontology-grounded.** Models, schemas, and knowledge graphs are first-class.
- **TDD.** Every function ships with a test. Red → green → refactor, always.
- **Secure.** No hardcoded paths, tokens, or secrets. Ever.
---
## Structure
```
ai-config/
├── AGENTS.md ← single source of truth — edit this only
├── build.sh ← generates all tool folders + wires git hooks
├── claude/
│ ├── CLAUDE.md ← generated
│ ├── settings.json ← model, hooks, statusline config
│ ├── agents/ ← 8 specialist sub-agents
│ ├── hooks/ ← session management, context injection, statusline
│ └── commands/ ← /-check, /-ship, /-review, /-discover, /-session, /-sync
├── cursor/
│ ├── user-rules.md ← generated — paste into Settings → Rules for AI
│ └── rules/
│ └── global.mdc ← generated — drop into .cursor/rules/ per project
├── gemini/
│ └── GEMINI.md ← generated
└── codex/
└── AGENTS.md ← generated
```
Each tool folder is built exhaustively. You only need the generated config file for your tool — the rest (agents, hooks, commands) is Claude Code-specific.
---
## Install
**1. Clone and build:**
```bash
git clone https://github.com/PR0CK0/ai-config.git ~/ai-config
cd ~/ai-config
./build.sh # generates all tool folders + wires pre-push hook
```
**2. Claude Code — rules only (drop-in):**
```bash
cp ~/ai-config/claude/CLAUDE.md ~/.claude/CLAUDE.md
```
**Claude Code — full setup (agents, hooks, slash commands, statusline):**
```bash
cp ~/ai-config/claude/CLAUDE.md ~/.claude/CLAUDE.md
cp ~/ai-config/claude/settings.json ~/.claude/settings.json
cp -r ~/ai-config/claude/agents ~/.claude/agents
cp -r ~/ai-config/claude/hooks ~/.claude/hooks
cp -r ~/ai-config/claude/commands ~/.claude/commands
chmod +x ~/.claude/hooks/statusline.sh
```
**3. Cursor — global rules:**
Paste `cursor/user-rules.md` content into Cursor → Settings → Rules for AI.
**Cursor — per-project:**
```bash
mkdir -p .cursor/rules && cp ~/ai-config/cursor/rules/global.mdc .cursor/rules/
```
**4. Gemini CLI:**
```bash
mkdir -p ~/.gemini && cp ~/ai-config/gemini/GEMINI.md ~/.gemini/GEMINI.md
```
**5. Codex CLI:**
```bash
cp ~/ai-config/codex/AGENTS.md ~/AGENTS.md # global
# or copy to project root for per-project
```
**Per-project pointer (any tool):**
```bash
echo "READ ~/ai-config/AGENTS.md BEFORE ANYTHING" > AGENTS.md
```
**Symlinks instead of copies:**
```bash
./build.sh --symlink
```
> `build.sh` wires `.githooks/pre-push` automatically — if `AGENTS.md` changes, the hook rebuilds all aliases and blocks the push until generated files are committed.
---
## Inspiration
| Source | What it contributed |
|--------|---------------------|
| [steipete/agent-scripts](https://github.com/steipete/agent-scripts) | Git safety rules, atomic commit discipline, the pointer pattern |
| [steipete — Just Talk To It](https://steipete.me/posts/just-talk-to-it) | "Organizational scar tissue" philosophy; blast-radius thinking |
| [forrestchang/andrej-karpathy-skills](https://github.com/forrestchang/andrej-karpathy-skills) | Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution |
| Home-grown | Ontologist + modeler agents, session hooks, slash commands, statusline |
---
## License
MIT