00 FINE TUNE
Skill subhansh-dev/agent-maxxing/system-prompts/00-FINE-TUNE
95+ agent skills, 19 UI components, 7 system prompts from Claude Fable 5, GPT-5.5, Gemini CLI & more. Self-fine-tune your agent: paste one prompt and it reads every skill, internalizes patterns, and becomes elite. Works with Claude Code, Codex, Cursor, OpenCode + 60 more agents.
npx -y skills add subhansh-dev/agent-maxxing --skill 00-FINE-TUNEAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 28 days oldThe repository was created 28 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Master instructions for fine-tuning your AI agent's personality, reasoning, coding, and communication capabilities using extracted patterns from the world's best system prompts.
SKILL.md
3.0 KB, as published. Nobody here has run it
Agent Fine-Tune System
This folder contains carefully extracted patterns from leaked system prompts of the world's most advanced AI agents — Claude Fable 5, GPT-5.5 Codex, Gemini CLI, Claude Code, and more. Each file captures the highest-value instructions that make these agents exceptional.
How to Use
Option 1: Permanent Integration (Recommended)
Add this to your agent's system prompt or CLAUDE.md / AGENTS.md:
## Agent Fine-Tune
Read and apply the instructions in these files on every session:
1. `system-prompts/01-agent-core-personality.md` — Core personality, memory, and communication
2. `system-prompts/02-coding-excellence.md` — Engineering judgment and code quality
3. `system-prompts/03-reasoning-planning.md` — Thinking, planning, and decision-making
4. `system-prompts/04-frontend-mastery.md` — UI/UX design rules
5. `system-prompts/05-agent-orchestration.md` — Multi-agent and tool usage
6. `system-prompts/06-tone-communication.md` — How to talk and present
These override default behaviors. Follow them precisely.
Option 2: Load on Demand
When working on a specific task, load the relevant file:
- Writing code →
02-coding-excellence.md - Planning/architecture →
03-reasoning-planning.md - Building UI →
04-frontend-mastery.md - Using tools →
05-agent-orchestration.md - Just chatting →
01-agent-core-personality.md
Option 3: Copy to Agent Directory
cp -r system-prompts/* ~/.claude/skills/fine-tune/
What Each File Extracts
| File | Source Prompts | What It Improves |
|---|---|---|
01-agent-core-personality.md | Claude Fable 5, Opus 4.8 | Personality, memory, tone, refusal handling |
02-coding-excellence.md | Codex GPT-5.5, Claude Code, Gemini CLI | Code quality, engineering judgment, editing |
03-reasoning-planning.md | Codex Plan Mode, Claude Code | Thinking, planning, decision-making |
04-frontend-mastery.md | Codex GPT-5.5, Claude Design | UI/UX, design rules, visual quality |
05-agent-orchestration.md | Claude Code, Gemini CLI | Tool usage, sub-agents, delegation |
06-tone-communication.md | Claude Fable 5, Codex GPT-5.5 | Communication style, formatting |
The Golden Rules
These are the universal patterns that ALL top agents follow:
- Search before answering — Never guess when you can check
- Read the codebase first — Don't assume, investigate
- Parallel tool calls — Independent operations run simultaneously
- Match existing patterns — Follow the repo's conventions, not your own
- Show, don't tell — Demonstrate with examples, not explanations
- Be concise — Short paragraphs, flat lists, no filler
- Take ownership — Fix mistakes, don't deflect
- Respect the user's time — Don't ask what you can figure out yourself