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.

The GPT Project Skill System is a session-scoped skill loader for ChatGPT Projects. This page explains what the system is, the problem it solves, how its core ideas fit together, and what it explicitly does not do — so you can decide whether it fits your workflow before reading anything else.

​
The problem it solves

ChatGPT Projects can use uploaded files as working context, but they provide no native skill-package workflow. There is no built-in way to bundle a set of reusable instructions, reference files, and tool scripts into a portable unit, upload that unit to a Project, and mount it into a session with a single explicit command. Without a workflow like this, the only option is copying instructions by hand every time you start a new work session — which is fragile, error-prone, and hard to version. GPT Project Skill System adds that workflow.

​
Three core ideas

The system is built around three simple ideas:
  1. Package a skill as a ZIP file — instructions, support files, scripts, and assets all in one place.
  2. Unpack that ZIP into the Project’s runtime environment (/mnt/data/GPT.SKILLS/).
  3. Load the declared instructions into the current session only when you choose to use them.
A skill is not active because its ZIP file is present in the Project. It becomes active only after it has been explicitly unpacked and loaded.

​
What a skill package can include

A skill package is a self-contained ZIP file. It can carry more than one file:
File typePurpose
Instruction filesThe primary content mounted into the session
Support / reference filesSupplemental material available after unpack, consulted only when needed
Python-compatible tool scriptsScripts physically available after unpack, never run automatically
Assets, templates, examplesData or template files physically usable from /mnt
Only files declared in a skill’s load_sequence are semantically mounted as active context. Support files, tool scripts, and assets are physically available after unpack but are not automatic active context.

​
The session-first model

The active skill set exists only inside the current chat session. The system is designed around short, controlled, focused sessions — not a permanent always-on capability library. When your work focus changes materially, the right move is to open a new chat in the same Project and repeat a targeted boot/load cycle with only the skills that belong in the new session. This keeps the active semantic set small and prevents context drift. SKILLS/ under the unpacked core is a physical slot for installed skills, not a semantic registry. Physical presence there does not mean a skill is active.

​
The core package

The core itself is packaged as dist/GPT.SKILLS.zip. When unpacked it produces: 
/mnt/data/GPT.SKILLS/
  SYSTEM_CORE/
  SKILLS/
The SYSTEM_CORE/ directory holds five logical files that govern bootstrap, operating model, and lifecycle rules: 
MANIFEST.json          ← controls what is read and in what order
ENTRY.md               ← boots and activates the core
README.md              ← descriptive overview
SEMANTICS.md           ← operating model
OPERATIONAL_RULES.md   ← lifecycle commands and validation
MANIFEST.json is the authoritative source of load order. The core version at time of writing is 0.6.

​
What the system does NOT provide

The following capabilities are outside the scope of this system and will not be added by loading any skill or modifying any file.
The system does not provide:
  • Permanent skill installation — active skills exist only in the current chat
  • Automatic skill discovery — no package is scanned unless you name it explicitly
  • Automatic skill loading — uploading a ZIP does not load it
  • Automatic routing — the model does not redirect requests to skills on its own
  • Hooks — no lifecycle hooks fire on any event
  • Watchers — no file or state is watched
  • Daemons — no background process runs
  • Dependency installation — no pip install, no network install
  • Marketplace behavior — no registry, no discovery feed
  • Persistent runtime state — nothing survives the end of a chat session

​
Explore further

Quickstart

Upload the core, add custom instructions, load your first skill, and run a focused session in under five minutes.

Core Concepts

Understand the session model, core vs. skill split, lifecycle phases, mount meaning, and cooperative composition.

Skill Package Format

Learn the two supported package shapes, manifest fields, path rules, and what the build script validates.

Command Reference

Full reference for SKILL CORE UNPACK, SKILL <name> UNPACK, and SKILL <name> LOAD.

Build docs developers (and LLMs) love

Get started for freeTalk to us