{
  "//": "Single source of truth for the databricks plugin across all targets (Claude, Codex, Copilot, Cursor). Edit this file, then run `python3 scripts/skills.py generate`. Generated from here: the 4 plugin.json (under plugins/databricks/), the 4 marketplace.json catalogs (at the repo root), and the whole plugins/databricks/ bundle (a copy/transform of skills/hooks/commands/rules/assets that agents fetch scoped). CI fails on drift. Do not hand-edit the generated files. The release version lives in the sibling version.meta.json (current_version), not here.",
  "name": "databricks",
  "description": "Databricks skills for the CLI, Apps, Lakebase, Model Serving, Lakeflow Jobs, Spark Declarative Pipelines, Declarative Automation Bundles (DABs), and classic-to-serverless migration.",
  "category": "development",
  "author": {
    "name": "Databricks",
    "url": "https://databricks.com"
  },
  "homepage": "https://github.com/databricks/databricks-agent-skills",
  "repository": "https://github.com/databricks/databricks-agent-skills",
  "license": "LicenseRef-Databricks",

  "keywords_lead": ["databricks"],
  "keywords_tail": ["data-engineering"],

  "marketplace": {
    "name": "databricks-agent-skills",
    "description": "Databricks agent skills for AI coding assistants.",
    "owner": { "name": "Databricks" },
    "displayName": "Databricks Agent Skills",
    "//source": "How each catalog points at the built bundle. The four plugin.json live under <subdir>/<target-dir>/, so agents fetch only <subdir> (a scoped/sparse clone). ref_template is the git ref the ref-capable catalogs (Claude/Codex/Copilot) use; it is `main`, where the committed bundle lives, so installs follow the latest committed bundle. The catalogs are not pinned to release tags by design. Cursor cannot pin a ref anyway and tracks the default branch. NOTE: the Codex entry still needs its `policy` block + a remote git-subdir-at-a-tag smoke test (open question in the proposal); add `codex_policy` here once verified.",
    "source": {
      "subdir": "plugins/databricks",
      "repo": "databricks/databricks-agent-skills",
      "git_url": "https://github.com/databricks/databricks-agent-skills.git",
      "ref_template": "main"
    }
  },

  "targets": {
    "claude": {
      "dir": ".claude-plugin",
      "commands": "./commands/",
      "hooks_render": {
        "out": "hooks/hooks.json",
        "dialect": "nested",
        "env_root": "${CLAUDE_PLUGIN_ROOT}",
        "tool_matcher": "Bash",
        "description": true
      }
    },
    "codex": {
      "dir": ".codex-plugin",
      "hooks_render": {
        "out": "hooks/codex-hooks.json",
        "dialect": "nested",
        "env_root": "${PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}",
        "tool_matcher": "(?i)bash|shell|exec",
        "description": false
      },
      "interface": {
        "displayName": "Databricks",
        "shortDescription": "Databricks skills and hooks for Codex.",
        "category": "development",
        "brandColor": "#FF3621",
        "logo": "./assets/databricks.png",
        "composerIcon": "./assets/databricks.svg"
      }
    },
    "copilot": {
      "dir": ".github/plugin",
      "displayName": "Databricks",
      "hooks_render": {
        "out": "hooks/copilot-hooks.json",
        "dialect": "copilot",
        "tool_matcher": "(?i)bash|shell|terminal"
      }
    },
    "cursor": {
      "dir": ".cursor-plugin",
      "displayName": "Databricks Skills",
      "commands": "./commands/",
      "rules": "./rules/",
      "hooks_render": {
        "out": "hooks/cursor-hooks.json",
        "dialect": "cursor",
        "tool_matcher": "Shell",
        "platform_flag": "--platform cursor"
      }
    }
  },

  "//skills": "Per-skill plugin metadata. Insertion order defines the keyword order in plugin.json: keywords = keywords_lead + [each skill's keyword] + keywords_tail. Repo-owned (does NOT live in SKILL.md frontmatter, which syncs from the internal source). Prompt-routing config lives in the separate routing block below.",
  "skills": {
    "databricks-core": { "keyword": "cli" },
    "databricks-data-discovery": { "keyword": "data-discovery" },
    "databricks-apps": { "keyword": "apps" },
    "databricks-app-design": { "keyword": "app-design" },
    "databricks-lakebase": { "keyword": "lakebase" },
    "databricks-model-serving": { "keyword": "model-serving" },
    "databricks-jobs": { "keyword": "jobs" },
    "databricks-pipelines": { "keyword": "pipelines" },
    "databricks-dabs": { "keyword": "dabs" },
    "databricks-serverless-migration": { "keyword": "serverless" },
    "databricks-vector-search": { "keyword": "vector-search" },
    "databricks-agent-bricks": { "keyword": "agent-bricks" },
    "databricks-ai-functions": { "keyword": "ai-functions" },
    "databricks-aibi-dashboards": { "keyword": "aibi-dashboards" },
    "databricks-apps-python": { "keyword": "apps-python" },
    "databricks-dbsql": { "keyword": "dbsql" },
    "databricks-docs": { "keyword": "docs" },
    "databricks-execution-compute": { "keyword": "execution-compute" },
    "databricks-genie-agents": { "keyword": "genie-agents" },
    "databricks-iceberg": { "keyword": "iceberg" },
    "databricks-lakeflow-connect": { "keyword": "lakeflow-connect" },
    "databricks-metric-views": { "keyword": "metric-views" },
    "databricks-ml-training": { "keyword": "ml-training" },
    "databricks-mlflow-evaluation": { "keyword": "mlflow-evaluation" },
    "databricks-python-sdk": { "keyword": "python-sdk" },
    "databricks-spark-structured-streaming": { "keyword": "structured-streaming" },
    "databricks-synthetic-data-gen": { "keyword": "synthetic-data" },
    "databricks-unity-catalog": { "keyword": "unity-catalog" },
    "databricks-unstructured-pdf-generation": { "keyword": "pdf-generation" },
    "databricks-zerobus-ingest": { "keyword": "zerobus" }
  },

  "//routing": "Prompt-router + Cursor-rule routing config, rendered by scripts/skills.py. strong/ambiguous/suppress are regex patterns (backslashes are doubled for JSON). The router (hooks/databricks-router.py) loads the generated hooks/_routing_data.json; the Cursor rule (rules/databricks-routing.mdc) and the router instruction are both rendered from the single `table`, so the two routing tables cannot drift.",
  "routing": {
    "strong": [
      "\\bdatabricks\\b",
      "\\bunity\\s+catalog\\b",
      "\\blakeflow\\b",
      "\\blakebase\\b",
      "\\bdbfs\\b",
      "\\bdbutils\\b",
      "\\bdbsql\\b",
      "\\bdatabricks\\.yml\\b",
      "\\basset\\s+bundle\\b",
      "\\bdabs\\b",
      "\\b(?:lakeflow|spark)\\s+declarative\\s+pipelines?\\b",
      "\\bdelta\\s+live\\s+tables?\\b",
      "\\bmosaic\\s+ai\\b",
      "\\bdelta\\s+sharing\\b",
      "\\bcloudfiles\\b"
    ],
    "ambiguous": [
      "\\bgenie\\b",
      "\\bdelta\\s+(lake|tables?)\\b",
      "\\bdeclarative\\s+pipelines?\\b",
      "\\bmodel\\s+serving\\b",
      "\\bvector\\s+search\\b",
      "\\bmlflow\\b",
      "\\bpyspark\\b",
      "\\bspark\\s*\\.\\s*(sql|read|write|table)\\b",
      "\\bserverless\\s+(compute|warehouse|migration)\\b",
      "\\bmedallion\\s+(architecture|tables?)\\b",
      "\\bsql\\s+warehouse\\b",
      "\\bauto\\s+loader\\b",
      "\\bwhat\\s+tables?\\b",
      "\\bwhich\\s+(catalog|schema)\\b",
      "\\b(list|show|find)\\s+(the\\s+)?tables?\\b",
      "\\bwhere\\s+does\\s+.{1,40}\\s+(data|table|live)\\b",
      "\\binformation_schema\\b"
    ],
    "suppress": [
      "\\bbigquery\\b",
      "\\bredshift\\b",
      "\\bsynapse\\b",
      "\\bsnowflake\\b",
      "\\bgit\\s+(commit|push|pull|status|log|diff|branch|rebase|merge|clone|stash)\\b",
      "\\b(read|edit|open|write|create|delete)\\s+(the\\s+|this\\s+|a\\s+|that\\s+)?file\\b",
      "\\bunit\\s+tests?\\b",
      "\\bnpm\\b",
      "\\bpip\\s+install\\b",
      "\\bdocker\\b",
      "\\bkubernetes\\b",
      "\\bjenkins(?:file)?\\b"
    ],
    "//table": "Ordered product-skill routing table. Both the router instruction and the Cursor .mdc rule render their rows from here. A skill's `note` is appended after the skill name (Genie's experimental caveat).",
    "table": [
      { "label": "Jobs / Lakeflow / workflows", "skill": "databricks-jobs" },
      { "label": "Pipelines / Lakeflow Spark Declarative Pipelines (formerly DLT)", "skill": "databricks-pipelines" },
      { "label": "Apps / AppKit", "skill": "databricks-apps" },
      { "label": "Asset Bundles / DABs / databricks.yml", "skill": "databricks-dabs" },
      { "label": "Model Serving / endpoints", "skill": "databricks-model-serving" },
      { "label": "Lakebase / Postgres", "skill": "databricks-lakebase" },
      { "label": "Vector Search / RAG", "skill": "databricks-vector-search" },
      { "label": "Classic-to-serverless migration", "skill": "databricks-serverless-migration" },
      { "label": "Data discovery / finding & exploring data / natural-language data Q&A / SQL generation (Genie)", "skill": "databricks-data-discovery" },
      { "label": "Genie Agents (formerly Genie Spaces): building, managing & tuning a curated per-data Genie Agent", "skill": "databricks-genie-agents" },
      { "label": "Agent Bricks / Knowledge Assistants / Genie Spaces / Multi-Agent Supervisor", "skill": "databricks-agent-bricks" },
      { "label": "AI Functions (ai_query, ai_classify, ai_extract, ai_parse_document)", "skill": "databricks-ai-functions" },
      { "label": "AI/BI dashboards", "skill": "databricks-aibi-dashboards" },
      { "label": "Python data apps (Streamlit, Dash, Gradio, Flask, FastAPI)", "skill": "databricks-apps-python" },
      { "label": "Databricks SQL warehouses", "skill": "databricks-dbsql" },
      { "label": "Databricks documentation lookup (llms.txt)", "skill": "databricks-docs" },
      { "label": "Executing code on Databricks compute", "skill": "databricks-execution-compute" },
      { "label": "Apache Iceberg tables / UniForm / Iceberg REST Catalog", "skill": "databricks-iceberg" },
      { "label": "Lakeflow Connect managed ingestion connectors", "skill": "databricks-lakeflow-connect" },
      { "label": "Metric Views / governed metrics", "skill": "databricks-metric-views" },
      { "label": "ML model training", "skill": "databricks-ml-training" },
      { "label": "MLflow agent / GenAI evaluation", "skill": "databricks-mlflow-evaluation" },
      { "label": "Python SDK / Databricks Connect / REST API", "skill": "databricks-python-sdk" },
      { "label": "Spark Structured Streaming", "skill": "databricks-spark-structured-streaming" },
      { "label": "Synthetic / test data generation (Faker)", "skill": "databricks-synthetic-data-gen" },
      { "label": "Unity Catalog governance, access control & system tables", "skill": "databricks-unity-catalog" },
      { "label": "Synthetic PDF generation for RAG", "skill": "databricks-unstructured-pdf-generation" },
      { "label": "Zerobus streaming ingest", "skill": "databricks-zerobus-ingest" }
    ],
    "instruction_preamble": "[DATABRICKS] This request is Databricks-related. Handle it through the Databricks skills rather than ad hoc commands. Use the Skill tool to load `databricks-core` first (the parent skill: CLI, auth, profile selection, data exploration), then load the product skill that matches the request:",
    "instruction_closing": "Then follow the skill's guidance (it drives the `databricks` CLI). If no product skill fits, databricks-core alone is enough.",
    "reminder": "[DATABRICKS] Databricks-related prompt: keep routing through the Databricks skills (databricks-core plus the matching product skill); load any that are not already loaded.",
    "rule_description": "Routing for any Databricks task -- CLI, auth, profiles, data exploration, Jobs/Lakeflow, Spark Declarative Pipelines (formerly DLT), Apps/AppKit, Asset Bundles/DABs, Model Serving, Lakebase/Postgres, Vector Search/RAG, Genie, and classic-to-serverless migration. Apply when the request is Databricks-related so the Databricks skills are loaded instead of ad hoc commands.",
    "rule_preamble": [
      "This request is Databricks-related. Handle it through the Databricks skills",
      "rather than ad hoc commands. Load the `databricks-core` skill first (the parent:",
      "CLI, auth, profile selection, data exploration), then load the product skill",
      "that matches the request:"
    ],
    "rule_closing": [
      "Then follow the skill's guidance (it drives the `databricks` CLI). If no product",
      "skill fits, databricks-core alone is enough."
    ]
  },

  "//hooks": "Canonical hook wiring, rendered into each target's dialect by scripts/skills.py: hooks/hooks.json (Claude, auto-loaded), codex-hooks.json, copilot-hooks.json, cursor-hooks.json. The shared content (the scripts, their events, the description, the session-start matcher) lives here once; per-target dialect knobs (schema shape, event casing, env var, command form, tool matcher, whether the router is wired) live under targets.*.hooks_render. The router is wired only on Claude + Codex (the other two cannot inject context from a prompt-submit hook).",
  "hooks": {
    "description": "Databricks plugin hooks: route Databricks prompts into the skills, prime session context, and hint on auth failures.",
    "session_start_matcher": "startup|clear|compact",
    "entries": [
      { "id": "router", "event": "prompt_submit", "script": "databricks-router.py" },
      { "id": "context", "event": "session_start", "script": "databricks-context.py" },
      { "id": "auth", "event": "post_tool", "script": "databricks-auth-helper.py" }
    ]
  }
}