Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/chatgpt-skill-system/llms.txt

Use this file to discover all available pages before exploring further.

GPT Project Skill System brings a structured skill-package workflow to ChatGPT Projects. Instead of copying instructions by hand or loading everything at once, you package each skill as a ZIP file, upload it to your Project, and explicitly unpack and load only the skills you need for the current session. When the session ends, so does the active skill set — no permanent install, no hidden state, no background routing.

Quickstart

Set up the core, load your first skill, and start a focused session in minutes.

Core Concepts

Understand how sessions, skills, packages, and lifecycle commands fit together.

Skill Package Format

Learn the ZIP structure, MANIFEST.json fields, and file category rules for building skills.

Command Reference

Every lifecycle command — UNPACK, LOAD, and their validation rules.

​
How it works

1

Upload the core and skill packages

Add dist/GPT.SKILLS.zip and your skill ZIPs (e.g. skill-adapter.zip) to your ChatGPT Project. Add the contents of CUSTOM_INSTRUCTIONS.txt to the Project’s custom instructions.
2

Boot the core

In a new chat, run SKILL CORE UNPACK. This extracts the core to /mnt/data/GPT.SKILLS/, reads MANIFEST.json, and activates the system for the current session.
3

Unpack and load a skill

Run SKILL skill-adapter UNPACK to install the skill, then SKILL skill-adapter LOAD to mount its instructions into the session. The skill is now semantically active.
4

Work, then close

Use the loaded skills for your focused task. When the work focus changes, open a new chat and repeat a targeted boot/load cycle to keep context clean.

​
What this is not

The skill set is active only in the current chat session. There is no permanent installation, no automatic routing, no hooks, and no persistent state. Every session starts clean.
The system intentionally excludes: permanent skill installation, automatic skill discovery, automatic loading, hook execution, watchers, daemons, dependency installation, and marketplace behavior. This is by design — a controlled, predictable session is the goal.

​
Key features

Session-scoped activation

Skills are active only inside the current chat. Closing the chat ends the active skill set with no residual state.

Explicit lifecycle commands

Nothing happens automatically. Every unpack and load is an explicit operator decision.

ZIP-based packages

Skills are self-contained ZIP files with a declarative MANIFEST.json. No install step, no registry.

Cooperative composition

Multiple skills can be active in the same session when they support the same work focus.

Build validation

The included build script validates package shape, manifests, paths, and UTF-8 safety before producing any ZIP.

skill-adapter included

The bundled skill-adapter skill helps you inspect and adapt external skill packages for compatibility.

Build docs developers (and LLMs) love

Get started for freeTalk to us