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 explicit lifecycle commands that control how the core and individual skill packages are bootstrapped, installed, and activated within a session. This page covers each command’s purpose, step-by-step behavior, hard fail conditions, and valid invocation syntax.

​
Command target boundaries

Each lifecycle command may operate only on its declared target. The runtime must not pre-scan, pre-validate, compile, inspect, or infer capability from packages that were not named by an explicit lifecycle command.
Skill ZIP files present among Project sources are inert until an explicit SKILL <name> UNPACK command is issued. Presence in Project sources does not imply install, validation, load, activation, mount, pre-scan, or tool availability.
CommandTarget
SKILL CORE UNPACKGPT.SKILLS.zip only
SKILL <name> UNPACK<name>.zip only
SKILL <name> LOADInstalled skill at /mnt/data/GPT.SKILLS/SKILLS/<name>/ only
SKILL CORE UNPACK must not unpack, inspect, validate, load, compile, or activate any skill package. SKILL <name> UNPACK targets only its named ZIP. SKILL <name> LOAD targets only the already-installed skill folder — it does not auto-unpack.

​
SKILL CORE UNPACK

Boots the core from GPT.SKILLS.zip and activates it for the current session.

​
Behavior

The command executes the following steps in order:
  1. Removes /mnt/data/GPT.SKILLS/ before extraction.
  2. Removes any legacy /mnt/data/sistema_madre/ before extraction.
  3. Extracts GPT.SKILLS.zip into /mnt/data/GPT.SKILLS/.
  4. Does not fall back to /mnt/data/sistema_madre/.
  5. Validates /mnt/data/GPT.SKILLS/SYSTEM_CORE/MANIFEST.json.
  6. Validates and mounts all files declared in the manifest.
  7. Activates the core for the current session.

​
Hard fail conditions

Hard fail if core unpack produces any extraction root other than /mnt/data/GPT.SKILLS/. There is no fallback path.

​
Re-running the command

Re-running SKILL CORE UNPACK in the same chat resets the active semantic state. Any skills that were previously loaded are no longer considered active after the command completes. The session begins fresh from the newly unpacked core.

​
Natural-language aliases

The command also accepts natural-language phrasings such as:
  • "Unpack GPT.SKILLS"
  • "Boot the skill system"
  • "Initialize GPT.SKILLS"

​
Example

SKILL CORE UNPACK

​
SKILL <name> UNPACK

Installs a skill package physically into the skill directory. This command does not make the skill semantically active — it only prepares it for a later SKILL <name> LOAD.

​
Slug rules for <name>

The <name> slug must satisfy all of the following:
  • Matches the pattern ^[a-z0-9_-]+$
  • Is not empty
  • Contains no spaces
  • Contains no . characters
  • Contains no / or \ characters
  • Is not and does not contain ..
Valid slugs:
  csv-profiler
  skill_adapter
  summarizer
  my-tool-v2

Invalid slugs:
  My_Skill       (uppercase letter)
  csv.profiler   (contains dot)
  csv/profiler   (contains slash)
  ../escape      (path traversal)
  skill name     (contains space)
                 (empty)

​
Behavior

  1. Validates <name> as a skill slug (pattern + character rules above).
  2. Verifies that the skill ZIP filename is exactly <name>.zip.
  3. Verifies that the skill ZIP has MANIFEST.json at its root with no single wrapper directory.
  4. Verifies that skill_name in the skill manifest equals <name>.
  5. Extracts the skill ZIP into /mnt/data/GPT.SKILLS/SKILLS/<name>/.
  6. Overwrites the previous extracted copy if one exists for the same skill name.
  7. Validates the package sufficiently to support a later SKILL <name> LOAD.

​
ZIP flat-root requirement

The skill ZIP must be flat at root:
  • MANIFEST.json must be at the ZIP root level — not inside a subdirectory.
  • No single wrapper directory may surround the package.
  • Safe internal folders within the package are allowed.
csv-profiler.zip
  MANIFEST.json       ← at ZIP root ✓
  SKILL.md
  docs/
    columns.md
  tools/
    profile_csv.py

​
Hard fail conditions

ConditionEffect
<name> violates the slug ruleHard fail
ZIP filename is not <name>.zipHard fail
skill_name in skill manifest does not equal <name>Hard fail
Install path would not be /mnt/data/GPT.SKILLS/SKILLS/<name>/Hard fail
ZIP contains a single wrapper directory around the packageHard fail

​
After unpack

The skill is physically installed but not semantically active. A skill folder present under /mnt/data/GPT.SKILLS/SKILLS/ means the skill is installed. It does not mean the skill is loaded or usable in the current session. Only SKILL <name> LOAD activates it.

​
Example

SKILL csv-profiler UNPACK

​
SKILL <name> LOAD

Mounts an already-installed skill into the current session, making it semantically active.

​
Behavior

  1. Does not auto-unpack the skill. The skill must have been previously unpacked.
  2. Fails if the skill has not been extracted into /mnt/data/GPT.SKILLS/SKILLS/<name>/.
  3. Reads the skill MANIFEST.json.
  4. Reads the files in the skill’s load_sequence, or falls back to SKILL.md if load_sequence is omitted.
  5. Makes the skill semantically active in the current session.
SKILL <name> LOAD requires a prior SKILL <name> UNPACK in the same or a previous session. There is no automatic install path — if the skill folder is absent, the command hard fails.

​
Reloading

Calling SKILL <name> LOAD on an already-loaded skill performs a full semantic reload — the manifest and load sequence are re-read from disk and the skill’s semantic state is refreshed.

​
After load

The skill is semantically active in the current session only. Active means the model has read the mounted skill files and will use them when relevant to the current task. Active does not mean the skill intercepts requests, runs hooks, watches files, or persists outside the current session.
Session scope is strict: the active semantic set exists only in the current chat. Nothing persists across chats or across sessions. To reactivate skills in a new chat, run SKILL CORE UNPACK followed by the required SKILL <name> LOAD commands.

​
Example

SKILL csv-profiler LOAD

​
Complete session example

The following shows a typical boot-and-load sequence for a session that uses two skills: 
# Step 1 — boot the core
SKILL CORE UNPACK

# Step 2 — install skill packages (only needed once per session or after update)
SKILL csv-profiler UNPACK
SKILL summarizer UNPACK

# Step 3 — activate the skills for this session
SKILL csv-profiler LOAD
SKILL summarizer LOAD
Load only the skills you need for the current work focus. Excess active context degrades session focus. When the work changes materially, open a new chat and repeat the targeted boot/load cycle.

Build docs developers (and LLMs) love

Get started for freeTalk to us