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.

A skill package is a flat ZIP file containing a MANIFEST.json at the root and one or more declared skill files. The manifest identifies the skill, declares the entry point, and maps every file into one of four functional categories. The loader reads MANIFEST.json first and uses it to drive all subsequent unpack and mount behavior.

​
Two package modes

The presence or absence of primary_file and load_sequence in the manifest controls which loading mode the runtime uses.

​
Fallback mode

When the manifest omits both primary_file and load_sequence, the loader applies fallback defaults: 
{
  "primary_file": "SKILL.md",
  "load_sequence": ["SKILL.md"]
}
Use fallback mode when the root SKILL.md is the only entry point and no extra mounted files are needed. It is the simplest possible shape. Fallback manifest: 
{
  "skill_name": "summarizer",
  "version": "1.0"
}
Fallback file tree: 
summarizer.zip
  MANIFEST.json
  SKILL.md
Root SKILL.md is required in fallback mode. If you use explicit manifest mode, SKILL.md at the root is not required — your primary_file may be any safe relative UTF-8 textual file.

​
Explicit manifest mode

When the manifest declares both primary_file and load_sequence, the loader uses those values directly. Use explicit mode when the entry point is nested, when extra mounted files are needed, or when you want unambiguous declaration of what gets loaded. Explicit manifest: 
{
  "skill_name": "csv-profiler",
  "version": "1.0",
  "primary_file": "SKILL.md",
  "load_sequence": ["SKILL.md"],
  "support_files": ["docs/columns.md"],
  "tool_files": ["tools/profile_csv.py"],
  "asset_files": ["templates/template.csv"]
}
Explicit file tree: 
csv-profiler.zip
  MANIFEST.json
  SKILL.md
  docs/
    columns.md
  tools/
    profile_csv.py
  templates/
    template.csv
Providing only one of primary_file or load_sequence is a hard-fail validation error. Both must be present or both must be omitted — there is no partial declaration.

​
ZIP structure rules

The ZIP must be flat at root: MANIFEST.json must appear directly inside the ZIP with no surrounding wrapper directory. Internal subfolders are permitted as long as their paths are safe. Valid structure: 
csv-profiler.zip
  MANIFEST.json          ← at ZIP root
  SKILL.md
  docs/
    columns.md           ← safe internal folder, allowed
  tools/
    profile_csv.py
  templates/
    template.csv
Invalid — wrapper directory: 
csv-profiler.zip
  csv-profiler/          ← single wrapper folder, hard fail
    MANIFEST.json
    SKILL.md
Additional packaging rules enforced at build and unpack time:
  • All ZIP entry paths must use POSIX / separators — backslashes (\) in any entry name are a hard fail.
  • Nested ZIP files inside a skill package are not allowed.
  • Every declared file path must be a file target, not a directory.

​
File categories

Every file in a skill package falls into one of four functional categories, each with distinct mount and availability semantics.

​
load_sequence

Files in load_sequence are semantically mounted into the active skill context during SKILL <name> LOAD. They are the only files that become active context automatically.
  • Must include primary_file.
  • All entries must be AI-only UTF-8 textual files (.md, .txt, and similar text formats).
  • Must not include scripts, executables, or binary assets.
  • May include nested files with safe relative paths (for example, docs/intro.md).
  • Every entry must exist in the ZIP and must be a file target, not a directory.

​
support_files

Support files are supplemental reference files. They are physically available after unpack but are not mounted into the active context automatically — they are consulted only when needed.
  • Every entry must be a safe relative path and must exist after unpack.
  • Must not overlap with load_sequence (overlap is a hard fail).
  • Use support_files for reference material that should be reachable but should not pollute the automatic context.

​
tool_files

Tool files are Python scripts physically available to the Python environment after unpack. They are never mounted and never run automatically.
  • Every entry must exist, be a file target, and be readable as UTF-8.
  • Tool files may use only Python standard library or packages already available in the current GPT Python environment.
  • pip install and network installs are prohibited. If a dependency is missing, the tool must degrade or fail with a clear warning.
  • The skill should provide a non-script fallback where possible.

​
asset_files

Asset files are data, template, media, or other physical files that the task environment can use directly. They are never mounted automatically.
  • Every entry must be a safe relative path and must exist after unpack.
  • May be non-textual when the environment can consume them as files (for example, CSV templates, images).

​
Name identity rule

The skill name must be consistent across all four locations. A mismatch at any point is a hard fail.
LocationMust equal
Command slugskill_name
ZIP filename<skill_name>.zip
Installed folder/mnt/data/GPT.SKILLS/SKILLS/<skill_name>/
skill_name in MANIFEST.jsonthe slug
Example — consistent identity: 
SKILL summarizer UNPACK    ← command slug: summarizer
summarizer.zip             ← ZIP filename matches
/mnt/data/GPT.SKILLS/SKILLS/summarizer/   ← installed folder matches

{
  "skill_name": "summarizer",
  "version": "1.0"
}
The skill_name must also match the pattern ^[a-z0-9_-]+$. Spaces, uppercase letters, dots, slashes, and .. are all invalid slugs.

​
Full manifest example

The following is the complete manifest for the csv-profiler skill demonstrating all four file categories and optional metadata fields: 
{
  "skill_name": "csv-profiler",
  "version": "1.0",
  "primary_file": "SKILL.md",
  "load_sequence": [
    "SKILL.md"
  ],
  "support_files": [
    "docs/columns.md"
  ],
  "tool_files": [
    "tools/profile_csv.py"
  ],
  "asset_files": [
    "templates/template.csv"
  ],
  "capabilities": ["csv analysis"],
  "runtime_hints": {
    "uses_python": true
  }
}

Build docs developers (and LLMs) love

Get started for freeTalk to us