Non-recursive skill file listing reports subfolders with nodeType "ITEM"

[andrii-novikov]

Name and Version

ai-dial-core development (Docker image epam/ai-dial-core:development), observed 2026-09-04.

What steps will reproduce the bug?

1. Create a skill that contains subdirectories — e.g. upload a ZIP through the UI with this layout:

   SKILL.md
   LICENSE.txt
   agents/analyzer.md
   assets/eval_review.html
   references/schemas.md
   scripts/utils.py
   

2. List its files non-recursively:

   GET /v2/metadata/skills/{bucket}/{path}/files/
   

3. Inspect nodeType on the entries that correspond to directories.

What is the expected behavior?

Directory entries are reported with "nodeType": "FOLDER", so that a client can tell files from directories by the field that exists for exactly that purpose.

This is what /v1/metadata/files/{bucket}/{path} does — its folder entries come back as FOLDER.

What do you see instead?

Every entry is "nodeType": "ITEM", directories included. Only the trailing / on url distinguishes them:

{
  "name": "files",
  "parentPath": "skill-creator",
  "bucket": "{bucket}",
  "url": "skills/{bucket}/skill-creator/files/",
  "nodeType": "FOLDER",
  "resourceType": "SKILL",
  "items": [
    {
      "name": "SKILL.md",
      "url": "skills/{bucket}/skill-creator/files/SKILL.md",
      "nodeType": "ITEM",
      "resourceType": "SKILL",
      "updatedAt": 1788530552986
    },
    {
      "name": "agents",
      "url": "skills/{bucket}/skill-creator/files/agents/",
      "nodeType": "ITEM",
      "resourceType": "SKILL"
    },
    {
      "name": "scripts",
      "url": "skills/{bucket}/skill-creator/files/scripts/",
      "nodeType": "ITEM",
      "resourceType": "SKILL"
    }
  ]
}

agents and scripts are directories. Both are reported as ITEM, and neither carries updatedAt (unlike the file entries beside them).

A client that trusts nodeType — the documented way to tell the two apart — will treat a directory as a file.

Additional information

• recursive=true is unaffected. That listing is flattened: it returns leaf files at every depth with the structure in parentPath, and contains no directory entries at all, so there is nothing to mislabel.
• Empty directories are omitted from both modes.
• The python client currently works around this by deriving node_type from the trailing / of url for skill file listings, so that its users do not have to know about the discrepancy. We would like to drop that workaround once this is fixed. See feat: add read-side resource for DIAL Core /v2/skills ai-dial-client-python#138.
• API reference for the endpoint: https://dialx.ai/dial_api