ITU

KnowledgebaseReferenceAutomated checks

Automated checks

Every submitted skill bundle is validated automatically, and the full report travels with the submission into the review queue as evidence. Failing checks never auto-reject a submission: the reviewer weighs the report and decides. The only hard requirement is a parseable manifest with a valid name, without which the skill cannot be identified.

The checks

Fail-level findings weigh heaviest in review; warnings flag recommended practice. Neither blocks the queue.

CheckLevelDetail
SKILL.md present and parseableFailThe file exists at the skill root and opens with a valid YAML frontmatter block.
Manifest declares name, description, licenseFailAll three fields are required.
Manifest declares a providerFailUnder metadata.provider (or top-level provider for legacy manifests).
Skill name is a valid slugFailLowercase letters, digits, and hyphens; 3-64 characters.
SKILL.md carries real instructionsFailAt least 80 characters of body beyond the manifest.
OpenAPI spec present under openapi/WarnRecommended for API skills; absence is flagged, not blocking.
Each OpenAPI file is a valid 3.x documentFailEvery .yaml/.json under openapi/ must parse and declare openapi: 3.x.
Schemas present under schemas/WarnRecommended by the marketplace guidelines.
Each schema file is valid JSONFailEvery .json under schemas/ must parse.
Rulebooks present under rulebooks/WarnPolicy rules should be separable from code.
depends_on paths resolveFailEvery schema or rulebook the manifest names must exist inside the bundle.

Where you see the results

The full report appears in the provider’s My submissions view, in the reviewer’s pane, and - once published - in the skill’s public review trail. Fix a failed bundle by updating the repository and resubmitting; see Publish and update skills.