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 exposes three lifecycle commands that you type directly in a ChatGPT Project chat. Each command has a single declared target, performs strict validation before acting, and hard-fails on any error — it never silently continues past an invalid state. Understanding each command’s scope and boundaries is essential before adding skills to a Project.

​
SKILL CORE UNPACK

Boots the session core from GPT.SKILLS.zip. Syntax 
SKILL CORE UNPACK
Target: GPT.SKILLS.zip only.
1

Clean previous state

Removes /mnt/data/GPT.SKILLS/ before extraction. Also removes any legacy /mnt/data/sistema_madre/ directory if present.
2

Extract the core archive

Extracts GPT.SKILLS.zip into /mnt/data/GPT.SKILLS/. The extracted root must be exactly /mnt/data/GPT.SKILLS/ — no other root is accepted.
3

Validate the core manifest

Reads and validates SYSTEM_CORE/MANIFEST.json. Required fields: system_name, version, primary_file, load_sequence, optional_load_sequence. For core version 0.6, system_name must be exactly gpt_project_skill_system and version must be exactly "0.6".
4

Validate and mount load_sequence files

Validates every file declared in load_sequence. Each must exist in the extracted tree, must be a file target (not a directory), and must be readable as UTF-8. Files in optional_load_sequence are probed and mounted if present and valid; missing or non-UTF-8 optional files produce a degraded-load warning and are skipped.
5

Activate the core

Marks the core as semantically active for the current session.
Hard fail conditions
  • The extracted root is anything other than /mnt/data/GPT.SKILLS/
  • SYSTEM_CORE/MANIFEST.json is missing, not valid JSON, or fails field validation
  • Any file in load_sequence is missing, is a directory target, or is not UTF-8
  • Any ZIP entry uses \ as a path separator
Rerunning SKILL CORE UNPACK in the same chat resets the active semantic state and clears all previously loaded skills. Previously loaded skills are no longer considered active after core unpack completes.
Target boundary: SKILL CORE UNPACK must not inspect, unpack, validate, load, compile, or activate any skill package. Skill ZIP files present in Project sources are inert until an explicit SKILL <name> UNPACK is issued. Presence in Project sources does not imply pre-scan, validation, or tool availability.

​
SKILL <name> UNPACK

Installs a skill package physically into the skill directory. Syntax 
SKILL <name> UNPACK
Example 
SKILL skill-adapter UNPACK
Target: <name>.zip only.
1

Validate the slug

Validates <name> against the slug rule: only [a-z0-9_-] characters, not empty, no spaces, no ., no /, no \, no ...
2

Verify the ZIP filename

Confirms the ZIP filename is exactly <name>.zip. A filename mismatch is an immediate hard fail.
3

Verify the package structure

Confirms MANIFEST.json is present at ZIP root and that no single wrapper directory surrounds the package contents.
4

Verify skill_name

Reads skill_name from the skill manifest and confirms it equals <name>. The ZIP filename, command slug, installed folder name, and skill_name in the manifest must all match.
5

Extract into the skill directory

Extracts the skill ZIP into /mnt/data/GPT.SKILLS/SKILLS/<name>/. If a previous extracted copy exists for the same skill, it is overwritten.
6

Validate for later load

Validates the skill package enough to support a subsequent SKILL <name> LOAD — checks declared file presence and path safety.
Slug rule 
^[a-z0-9_-]+$
No spaces, dots, slashes, backslashes, or .. sequences are permitted. Hard fail conditions
  • <name> violates the slug rule
  • The ZIP filename is not <name>.zip
  • skill_name in the manifest does not equal <name>
  • The install path is not /mnt/data/GPT.SKILLS/SKILLS/<name>/
  • The ZIP contains a single wrapper directory at root instead of MANIFEST.json
SKILL <name> UNPACK installs the skill physically into /mnt/data/GPT.SKILLS/SKILLS/<name>/. It does not activate the skill semantically. A skill folder present under SKILLS/ is installed, not active. Only SKILL <name> LOAD makes an installed skill active in the current session.

​
SKILL <name> LOAD

Mounts an installed skill and makes it semantically active in the current session. Syntax 
SKILL <name> LOAD
Example 
SKILL skill-adapter LOAD
Target: The installed skill at /mnt/data/GPT.SKILLS/SKILLS/<name>/ only.
1

Confirm the skill is installed

Fails immediately if the skill has not been unpacked into /mnt/data/GPT.SKILLS/SKILLS/<name>/. SKILL <name> LOAD does not auto-unpack.
2

Read the skill manifest

Reads MANIFEST.json from the installed skill directory. If primary_file and load_sequence are omitted, the loader applies fallback mode: primary_file = "SKILL.md", load_sequence = ["SKILL.md"].
3

Read load_sequence files

Reads every file declared in load_sequence (or the fallback sequence). Each file must exist, be a file target, and be readable as UTF-8.
4

Activate the skill

Makes the skill semantically active in the current session. From this point the model has read the mounted skill files and can apply them when relevant to the current task.
Reloading the same skill with SKILL <name> LOAD performs a full semantic reload — the skill context is re-read from the installed files.

​
Command target boundaries

Each command may operate only on its declared target. The table below summarises the strict scope of each command.
CommandDeclared targetMay it touch skill ZIPs?
SKILL CORE UNPACKGPT.SKILLS.zipNo — hard boundary
SKILL <name> UNPACK<name>.zipIts own ZIP only
SKILL <name> LOADInstalled skill at SKILLS/<name>/No ZIP access
The runtime must not pre-scan, pre-validate, compile, inspect, or infer capability from skill packages that were not named by an explicit lifecycle command. Presence of a skill ZIP in Project sources is inert until SKILL <name> UNPACK is explicitly issued.

​
Natural language aliases

You can also trigger the unpack and load commands using natural language. The following phrases are accepted as equivalents to the formal command syntax: 
Unpack the skill-adapter skill.
Load the skill-adapter skill.
These are recognized as natural language equivalents to SKILL skill-adapter UNPACK and SKILL skill-adapter LOAD respectively.

Path Rules

ZIP entry separator and safe relative path requirements.

Runtime Limits

What the system explicitly does not provide.

Manifest Reference

Every manifest field for core and skill packages.

Session Lifecycle

How to open, boot, load, and close a skill session.

Build docs developers (and LLMs) love

Get started for freeTalk to us