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.

MANIFEST.json is the identity and configuration file for every skill package. Every skill package must include one at the ZIP root — it is the first thing the loader reads during both SKILL <name> UNPACK and SKILL <name> LOAD. The manifest drives validation, physical unpack behavior, and semantic mount behavior. Without a valid MANIFEST.json, the skill cannot be installed or loaded.

​
Required fields

​
skill_name
string
required
The canonical name of the skill. Must match the pattern ^[a-z0-9_-]+$ — lowercase letters, digits, hyphens, and underscores only. Must exactly equal the command slug, the ZIP filename (without .zip), and the installed folder name under /mnt/data/GPT.SKILLS/SKILLS/. A mismatch at any of those four locations is a hard fail.Valid examples: summarizer, legal-review, csv_profilerInvalid examples: My Skill (space), Summarizer (uppercase), skill.name (dot), ../x (traversal)
​
version
string
required
A non-empty string that identifies the skill version. No specific format is enforced — "1.0", "0.1.0", and "2024-06" are all acceptable. An empty string or a missing field is a hard fail.

​
Load behavior fields

primary_file and load_sequence form a pair. Both must be present or both must be omitted. Providing only one is a hard fail.
​
primary_file
string
Safe relative path to the skill’s entry point file. Must be included in load_sequence. When both primary_file and load_sequence are omitted, the loader falls back to "SKILL.md" for this field. Required if load_sequence is present.
​
load_sequence
string[]
Ordered list of safe relative UTF-8 textual files to mount semantically during SKILL <name> LOAD. These are the only files that become active context automatically. Must include primary_file. Must not include scripts or binary assets. Must not overlap with support_files. Required if primary_file is present.When both primary_file and load_sequence are omitted, the loader falls back to ["SKILL.md"] for this field.

​
Physical availability fields

These fields declare files that are extracted from the ZIP and physically available after SKILL <name> UNPACK, but are not automatically mounted into the active context.
​
support_files
string[]
Supplemental reference files consulted only when needed. Physically available after unpack but not mounted automatically. Must not overlap with load_sequence — overlap is a hard fail. Every entry must be a safe relative path pointing to an existing file target.
​
tool_files
string[]
Python-compatible scripts physically available to the Code Interpreter environment after unpack. Never mounted and never run automatically. Every entry must be a safe relative path, must exist, and must be readable as UTF-8. Tool files may only use Python standard library or packages already available in the GPT Python environment — pip install and network installs are prohibited. If a required package is missing, tool execution must degrade or fail with a clear warning.
​
asset_files
string[]
Data, template, or media files physically available after unpack. Never mounted automatically. May be non-textual when the environment can use them as files (for example, CSV templates, images). Every entry must be a safe relative path pointing to an existing file target.

​
Metadata fields

These fields are descriptive only. They carry no runtime enforcement and do not alter mount behavior.
​
capabilities
string[]
Lightweight human-readable descriptions of what the skill can do. Examples: ["csv analysis", "column profiling"]. No format is enforced. Does not create routing, authorization, or validation guarantees.
​
runtime_hints
object
Lightweight hints about the skill’s runtime context. Example: {"uses_python": true, "output": "structured report"}. Does not create dependency management, authorization, routing, or validation guarantees. Keep it lightweight — do not use runtime_hints to express behavior the loader cannot mechanically verify.
Unknown manifest fields are ignored or warned at load time. They are never mounted and do not affect skill behavior. Unknown fields do not cause a hard fail — only explicitly invalid fields do.

​
Validation rules

The following conditions cause a hard fail during unpack or load. A hard fail stops the operation immediately with an error message.

Missing required field

Either skill_name or version is absent from the manifest.

Invalid skill_name

skill_name does not match ^[a-z0-9_-]+$, is empty, or does not match the ZIP filename, command slug, or installed folder.

Empty version

version is present but is an empty string.

primary_file not in load_sequence

primary_file is declared but the value is not listed in load_sequence.

Partial load pair

Exactly one of primary_file or load_sequence is present. Both must be provided or both must be omitted.

load_sequence / support_files overlap

A path appears in both load_sequence and support_files.

Unsafe paths

Any declared path is absolute, contains .., uses backslashes, contains empty segments, or contains a drive letter.

Script or binary in load_sequence

A file with an executable or binary extension (.py, .sh, .js, .exe, and similar) appears in load_sequence.

​
Complete examples

​
Minimal fallback manifest

The simplest valid manifest — SKILL.md at the ZIP root is used as both primary_file and the sole load_sequence entry by fallback: 
{
  "skill_name": "summarizer",
  "version": "1.0"
}

​
Full explicit manifest

All fields declared explicitly, including all four file categories and optional metadata: 
{
  "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",
    "column profiling"
  ],
  "runtime_hints": {
    "uses_python": true,
    "output": "structured csv report"
  }
}

Build docs developers (and LLMs) love

Get started for freeTalk to us