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.

Activating a skill in a GPT Project session is a deliberate two-step process: unpack, then load. Unpack installs the skill’s files physically into the runtime environment. Load reads those files and makes the skill semantically active in the current chat. Neither step happens automatically — a skill ZIP present in your Project is completely inert until you explicitly run both commands.
Skill ZIPs uploaded to a ChatGPT Project are inert by default. The system does not pre-scan, pre-validate, or auto-load any skill package. A skill becomes active only after an explicit SKILL <name> UNPACK followed by SKILL <name> LOAD.

​
Unpacking a skill

The SKILL <name> UNPACK command installs a skill package physically into the skill directory. It does not make the skill active. When you run SKILL <name> UNPACK, the runtime:
  1. Validates <name> against the slug rules (see Slug rules below).
  2. Verifies that the skill ZIP filename in the Project sources is exactly <name>.zip.
  3. Verifies that the ZIP contains a root MANIFEST.json with no single wrapper directory surrounding the package.
  4. Verifies that the skill_name field inside the manifest equals <name>.
  5. Extracts the ZIP into /mnt/data/GPT.SKILLS/SKILLS/<name>/, overwriting any previously extracted copy for the same skill name.
  6. Validates the skill package enough to support later load.
Example: 
SKILL skill-adapter UNPACK
After this command succeeds, the skill’s files are physically present at /mnt/data/GPT.SKILLS/SKILLS/skill-adapter/. The skill is installed but not yet active. No instructions have been read. No capabilities have been enabled.

​
Loading a skill

The SKILL <name> LOAD command mounts an already-installed skill and makes it semantically active for the current session. When you run SKILL <name> LOAD, the runtime:
  1. Confirms the skill has already been unpacked into /mnt/data/GPT.SKILLS/SKILLS/<name>/. It does not auto-unpack.
  2. Reads the skill’s MANIFEST.json.
  3. Reads and validates every file in the manifest’s load_sequence. If the manifest omits primary_file and load_sequence, the loader falls back to reading root SKILL.md.
  4. Makes the skill semantically active — the model treats the mounted files as live context for the session.
Example: 
SKILL skill-adapter LOAD
After this command succeeds, the skill is active. Its declared load_sequence files have been read and the model uses them when relevant to the current task.

​
Alternative syntax

Both commands also accept natural language phrasing. These are equivalent to the structured commands above: 
Unpack the skill-adapter skill.
Load the skill-adapter skill.
Use whichever form feels natural for your workflow. The structured SKILL <name> UNPACK / SKILL <name> LOAD form is recommended when precision matters or when scripting a session opening sequence.

​
Loading multiple skills

Multiple skills can be active in the same session as long as they support the same work focus. Load them one at a time: 
SKILL skill-adapter UNPACK
SKILL skill-adapter LOAD

SKILL my-other-skill UNPACK
SKILL my-other-skill LOAD
Active skills compose cooperatively, not sequentially. The model weighs the entire pool of active skills against the current task and applies whichever skill or combination of skills is most relevant. There is no routing layer and no priority queue — the model uses judgment.
Keep the active skill set small and focused. Loading skills “just in case” adds semantic noise and degrades task focus. Load only the skills your current work actually requires.

​
Reloading a skill

Running SKILL <name> LOAD for a skill that is already active performs a full semantic reload of that skill. The runtime re-reads the manifest and all load_sequence files from scratch. This is useful if you have re-unpacked the skill during the same session (for example, after modifying and re-uploading the ZIP) and want the updated content to take effect without opening a new chat. 
SKILL skill-adapter UNPACK
SKILL skill-adapter LOAD

​
Slug rules

The <name> argument in every skill command must be a valid slug. The slug is the canonical identity of the skill — it must match the ZIP filename, the skill_name in the manifest, and the installed folder path. Allowed characters: 
a-z  0-9  _  -
Rules:
  • Must not be empty.
  • No uppercase letters.
  • No spaces.
  • No dots (.).
  • No forward slashes (/) or backslashes (\).
  • No path traversal sequences (..).
Valid slug examples: 
skill-adapter
my-skill
code_review
v2-helper
Invalid slug examples: 
Skill-Adapter      ← uppercase letters
my skill           ← contains a space
my.skill           ← contains a dot
../skill           ← path traversal
skills/adapter     ← contains a slash
A hard fail is raised immediately if <name> violates any of these rules.

Build docs developers (and LLMs) love

Get started for freeTalk to us