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.

Every package in the GPT Project Skill System is described by a MANIFEST.json file. This page documents both schemas in full: the core manifest that lives in SYSTEM_CORE/ and the skill manifest that lives at the root of each skill package. It also covers fallback behavior, invalidity conditions that cause a hard fail, and a real-world skill manifest example.

​
Core manifest

The core manifest is located at /mnt/data/GPT.SKILLS/SYSTEM_CORE/MANIFEST.json after SKILL CORE UNPACK. It controls which files are mounted during core activation.

​
Current manifest

{
  "system_name": "gpt_project_skill_system",
  "version": "0.6",
  "primary_file": "ENTRY.md",
  "load_sequence": [
    "ENTRY.md",
    "README.md",
    "SEMANTICS.md",
    "OPERATIONAL_RULES.md"
  ],
  "optional_load_sequence": []
}

​
Core manifest fields

All five fields are required. A missing or type-invalid field causes a hard fail.
​
system_name
string
required
Must be exactly gpt_project_skill_system. Any other value is a manifest invalidity and causes hard fail.
​
version
string
required
Must be exactly "0.6" for core version 0.6. Any other value causes hard fail. This is a string, not a number.
​
primary_file
string
required
The entry point for the core. Must satisfy all of the following:
  • A direct filename — no subdirectory component.
  • Must be included in load_sequence.
  • Must have a .md or .json extension.
​
load_sequence
array of strings
required
Ordered list of files mounted during core activation. Must satisfy all of the following:
  • Non-empty array.
  • Each entry is a direct filename (no subdirectory).
  • Allowed extensions: .md and .json only.
  • No duplicate entries.
  • Must include primary_file.
  • Must not overlap with optional_load_sequence.
​
optional_load_sequence
array of strings
required
Ordered list of files mounted only if present and UTF-8 readable. May be an empty array. When a file in this list is missing or not UTF-8, the loader emits a technical warning, skips that file, and continues — this is a degraded load, not a hard fail. Must satisfy all of the following:
  • Each entry is a direct filename (no subdirectory).
  • Allowed extensions: .md and .json only.
  • No duplicate entries.
  • Must not overlap with load_sequence.

​
Skill manifest

The skill manifest is located at MANIFEST.json at the root of each skill ZIP and at /mnt/data/GPT.SKILLS/SKILLS/<name>/MANIFEST.json after unpack.

​
Required skill manifest fields

​
skill_name
string
required
The canonical name of the skill. Must satisfy all of the following:
  • Non-empty string.
  • Matches the pattern ^[a-z0-9_-]+$.
  • Equals the command slug <name> used in SKILL <name> UNPACK and SKILL <name> LOAD.
  • Equals the ZIP filename stem — the file must be named <skill_name>.zip.
  • Equals the installed folder name — the skill is extracted to /mnt/data/GPT.SKILLS/SKILLS/<skill_name>/.
All four identifiers — skill_name, command slug, ZIP filename, and installed folder — must match exactly.
​
version
string
required
The version of this skill package. Must be a non-empty string. No specific format is enforced, but the field must be present and non-empty.

​
Optional skill manifest fields

​
primary_file
string
The entry point for this skill. Must be a safe relative path. Must be present if and only if load_sequence is also present. Providing only one of primary_file or load_sequence is a manifest invalidity and causes hard fail.
  • Must be included in load_sequence.
  • Must point to a UTF-8 textual file, not a script or binary asset.
​
load_sequence
array of strings
Ordered list of AI-only textual files mounted during SKILL <name> LOAD. Must be present if and only if primary_file is also present. Must satisfy all of the following:
  • Non-empty array when present.
  • Each entry is a safe relative path.
  • Must include primary_file.
  • Every declared file must exist after unpack and be a file target (not a directory).
  • Every entry must be a UTF-8 textual file.
  • Must not include scripts intended for execution or binary assets.
  • Must not overlap with support_files.
​
support_files
array of strings
Supplemental reference files that are physically available after unpack but not mounted automatically. These are consulted only when needed during a session. Must satisfy:
  • Each entry is a safe relative path.
  • Every declared file must exist after unpack and be a file target.
  • Must not overlap with load_sequence.
​
tool_files
array of strings
Script or tool files physically available after unpack. These are never mounted automatically and never run automatically. The Python environment may execute them when explicitly invoked. Must satisfy:
  • Each entry is a safe relative path.
  • Every declared file must exist after unpack, be a file target, and be UTF-8 text.
  • Tools may use only the Python standard library or packages already available in the current GPT Python environment. No pip install. No network install.
​
asset_files
array of strings
Data, template, media, or other files physically available after unpack. Never mounted automatically. May be non-textual when used as files by the environment. Must satisfy:
  • Each entry is a safe relative path.
  • Every declared file must exist after unpack and be a file target.
​
capabilities
array of strings
Lightweight descriptive metadata listing what this skill can do. Does not create routing, authorization, dependency management, or automatic tool access. If present, must be an array of strings. An invalid type produces a warning and is ignored — it does not cause hard fail.
​
runtime_hints
object
Lightweight descriptive metadata as a JSON object. Not a control plane. Does not affect mount behavior, create dependencies, or authorize tool access. If present, must be a JSON object. An invalid type produces a warning and is ignored — it does not cause hard fail.

​
Fallback behavior

When both primary_file and load_sequence are omitted from the skill manifest, the loader activates fallback mode: 
{
  "primary_file": "SKILL.md",
  "load_sequence": ["SKILL.md"]
}
In fallback mode, a SKILL.md file at the root of the skill package is required. If it is missing, the load hard fails.
Fallback mode applies only to skill packages, not to the core. The core load_sequence is always required and may never be omitted.
Providing only one of primary_file or load_sequence is a manifest invalidity. They must either both be present or both be omitted.

​
Manifest invalidity — hard fail conditions

Any of the following conditions renders a manifest invalid and causes an immediate hard fail:
  • Missing required field (system_name, version, primary_file, load_sequence, optional_load_sequence for core; skill_name, version for skill)
  • Invalid field type (e.g., load_sequence is not an array of strings)
  • system_name is not gpt_project_skill_system (core only)
  • version is not "0.6" (core only)
  • Duplicate entries inside load_sequence
  • Duplicate entries inside optional_load_sequence
  • Overlap between load_sequence and optional_load_sequence (core)
  • Overlap between load_sequence and support_files (skill)
  • Any path that is absolute (begins with /)
  • Any path that contains ..
  • Any core mounted path that references outside SYSTEM_CORE/
  • Any core mounted path that contains a subdirectory component (core version 0.6: direct filenames only)
  • Unsupported mounted file extension (core: only .md and .json are allowed)
  • Providing only one of primary_file or load_sequence (skill)
Unknown manifest fields are ignored or warned. They are not mounted and do not cause hard fail.

​
Skill manifest example — skill-adapter

The following is the real manifest from the skill-adapter skill package: 
{
  "skill_name": "skill-adapter",
  "version": "0.1.0",
  "primary_file": "SKILL.md",
  "load_sequence": ["SKILL.md"],
  "support_files": ["references/adaptation_report_schema.md"],
  "tool_files": ["scripts/inspect_skill_candidate.py"],
  "asset_files": [],
  "capabilities": ["skill-adaptation", "compatibility-review", "manifest-drafting"],
  "runtime_hints": {
    "uses_python": true,
    "output": "candidate package plan plus compatibility report"
  }
}
In this manifest:
  • skill_name is skill-adapter, so the ZIP must be skill-adapter.zip and the command slug must be skill-adapter.
  • SKILL.md is explicitly declared as both primary_file and the sole entry in load_sequence — it is mounted during SKILL skill-adapter LOAD.
  • references/adaptation_report_schema.md is in support_files — available after unpack but not auto-mounted.
  • scripts/inspect_skill_candidate.py is in tool_files — physically available after unpack, never auto-run.
  • asset_files is an empty array — no asset files in this package.
  • capabilities and runtime_hints are descriptive metadata only.

Build docs developers (and LLMs) love

Get started for freeTalk to us