{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://skill-map.ai/spec/v1/tags/markdown.schema.json",
  "title": "NodeTagsReport",
  "description": "Canonical tagger-report shape (2026-07-21, the taxonomy leg of the summarizer-split direction; reframed 2026-07-25). An Action whose report schema `$ref`s a schema under the `tags/` namespace is a TAGGER: `sm record` detects that reference (mirror of the `summaries/` detection signal) and surfaces the report's `tags[]` as a PROPOSAL on the completion event (`job-events.md`, `tagsProposed`), writing nothing. Tags are human curation (`architecture.md` §Storage rule), so the operator reviews the proposal in the ordinary tags editor and saves it through the usual consent-gated `.sm` write. See `job-lifecycle.md` §Tags proposal.",
  "type": "object",
  "allOf": [
    {
      "$ref": "../report-base.schema.json"
    }
  ],
  "required": [
    "tags"
  ],
  "properties": {
    "tags": {
      "type": "array",
      "minItems": 0,
      "maxItems": 8,
      "items": {
        "type": "string",
        "minLength": 2,
        "maxLength": 30,
        "description": "Short lowercase kebab-case topical tag, written in the language of the file's content."
      },
      "description": "The topical tags the model proposes for the node (up to 8). With the node's CURRENT tags injected into the prompt (`job-lifecycle.md` §Current-tags injection for taggers), this list is what the model finds MISSING, so an EMPTY array is a legitimate answer: the existing vocabulary already covers the file. Lowercase kebab-case, in the language of the file's content."
    }
  }
}