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 enforces strict path and ZIP entry rules for both core and skill packages. This page defines what constitutes a safe relative path, explains the differences between core and skill path policies, documents ZIP entry separator requirements, and lists the slug pattern that skill names must satisfy.

​
Safe relative path definition

A path is safe and relative if it satisfies all of the following conditions:
  • Is not absolute — does not begin with /.
  • Does not contain .. — no traversal components.
  • Does not use backslash (\) path separators.
  • Does not contain empty path segments — no // or trailing / that implies a segment.
  • Does not contain drive letters — for example, no C: prefix.
  • Does not rely on traversal — no sequences that navigate outside the package root.
  • Does not use a wrapper directory as the sole ZIP root — MANIFEST.json must be reachable at ZIP root level.
Valid safe relative paths:
  MANIFEST.json
  SKILL.md
  docs/columns.md
  tools/helper.py
  references/schema.md
  templates/report.csv
  SYSTEM_CORE/MANIFEST.json

Invalid paths:
  /absolute/path.md          ← absolute (leading slash)
  ../escape.md               ← traversal with ..
  SYSTEM_CORE\..\escape.md   ← backslash and traversal
  C:\absolute\path.md        ← drive letter
  docs//columns.md           ← empty path segment
  C:relative.md              ← drive letter prefix

​
Core path policy

Core mounted paths are subject to a stricter policy than skill paths. In core version 0.6, all of the following restrictions apply to paths declared in the core manifest:
  • All declared paths are relative to SYSTEM_CORE/ — the loader resolves them as SYSTEM_CORE/<filename>.
  • Mounted paths must be direct filenames — no subdirectory component is permitted.
  • Allowed mounted extensions: .md and .json only.
  • The following are all disallowed in mounted core paths:
    • Absolute paths
    • ..
    • References that would resolve outside SYSTEM_CORE/
    • Subdirectory separators in mounted path strings
Files that are unpacked into the core ZIP but not declared in the manifest may exist on disk — they are simply not mounted. Presence on disk does not imply availability in the session context.
SKILLS/README.md is not mountable and must not appear in SYSTEM_CORE/MANIFEST.json. It must not be used as a list of active skills.

​
Skill path policy

Skill packages use a more permissive policy than the core. The key differences:
  • Skill packages may preserve original safe relative internal paths, including nested folder structures.
  • load_sequence may include SKILL.md and any safe relative UTF-8 textual file, including files in subdirectories.
  • support_files, tool_files, and asset_files may point to any safe relative path within the package.
The one restriction that applies to all file categories equally is that every path must be a safe relative path as defined above. 
Allowed in core manifest:
  ENTRY.md            ← direct filename, .md ✓
  README.md           ← direct filename, .md ✓
  MANIFEST.json       ← direct filename, .json ✓

Not allowed in core manifest:
  docs/overview.md    ← subdirectory component ✗
  ../escape.md        ← traversal ✗
  SKILLS/README.md    ← outside SYSTEM_CORE ✗
  helper.py           ← unsupported extension ✗

​
ZIP entry separator requirements

All ZIP entries in both core and skill packages must use POSIX / path separators.
  • ZIP entries must not contain \ (backslash).
  • The runtime must fail fast if a core or skill package contains ZIP entries with \.
  • The runtime must not silently normalize invalid ZIP entries — there is no automatic conversion from \ to /.
Valid ZIP entries:
  SYSTEM_CORE/MANIFEST.json
  SYSTEM_CORE/ENTRY.md
  SKILLS/README.md
  MANIFEST.json
  SKILL.md
  docs/columns.md
  tools/helper.py

Invalid ZIP entries:
  SYSTEM_CORE\MANIFEST.json     ← backslash separator ✗
  ..\escape.md                  ← traversal ✗
  C:\absolute\path.md           ← drive letter + backslash ✗
  /absolute/path.md             ← leading slash ✗

​
Slug rules for skill names

The <name> value used in SKILL <name> UNPACK and SKILL <name> LOAD must be a valid skill slug.
​
<name>
string
required
The skill slug used in lifecycle commands. Must match the pattern ^[a-z0-9_-]+$ and satisfy all of the following:
  • Not empty.
  • Lowercase ASCII letters, digits, hyphens, and underscores only.
  • No spaces.
  • No dots (.).
  • No forward slashes (/).
  • No backslashes (\).
  • Is not and does not contain ...
Valid slugs:
  csv-profiler
  skill_adapter
  summarizer
  my-tool-v2
  data123

Invalid slugs:
  My_Skill        ← uppercase letter
  csv.profiler    ← contains dot
  csv/profiler    ← contains forward slash
  csv\profiler    ← contains backslash
  ../escape       ← path traversal
  skill name      ← contains space
                  ← empty
  -leading-dash   ← (technically matches pattern, but see notes below)
The slug must match the skill_name field in the skill manifest, the ZIP filename (without the .zip extension), and the installed folder name under /mnt/data/GPT.SKILLS/SKILLS/. All four must be identical.

​
Wrapper directory rule

A skill ZIP must not use a single wrapper directory as the ZIP root. MANIFEST.json must appear at the ZIP root level — it must not be nested inside a folder that shares the package name. 
csv-profiler.zip
  MANIFEST.json       ← at ZIP root ✓
  SKILL.md
  docs/
    columns.md
  tools/
    profile_csv.py
The presence of a single wrapper directory causes a hard fail during SKILL <name> UNPACK. Internal subdirectories within the package are fully supported — the restriction applies only to a single top-level wrapper that surrounds the entire package.

Build docs developers (and LLMs) love

Get started for freeTalk to us