databricks-setup-local

Q: How do I install databricks-setup-local?

Run npx skills add https://github.com/databricks/databricks-agent-skills --skill databricks-setup-local in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/databricks-setup-local/ (personal) or .claude/skills/databricks-setup-local/ (project).

by databricksAI & Agents

Previews, provisions, or diagnoses a uv-managed local Python .venv with databricks environments setup-local. Use when the user wants to set up or fix one for Databricks Connect, cluster or serverless compute, --job-task, or a bundle target, or when setup-local fails.

Installs: 108
Last 24h: —
Trending rank: —
Security: Not audited

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/databricks/databricks-agent-skills --skill databricks-setup-local

01
Install
Run the command in your project. The CLI asks which agents to install for.
02
Describe the task
Ask your agent for something this skill covers. It loads databricks-setup-local automatically when relevant.
03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.

SKILL.md

The instructions your agent reads when the skill is triggered.

Databricks Setup-Local

REQUIRED FIRST: Use databricks-core for CLI, authentication, and profile selection. Never use a default profile. For an existing environment, use databricks-execution-compute.

Workflow

1. Check CLI and authentication

The CLI must be >= v1.16.0 -- the release that added --no-dbconnect and the E_PROVISION_CONFLICT failure this skill routes. Compare deterministically -- never eyeball the version (1.9.0 is older than 1.16.0, and a lexical string compare gets this wrong). Run this gate first and do not proceed if it exits non-zero:

# Subshell so a failed gate reports non-zero without closing a persistent shell.
(
  # Read stdout only -- upgrade nags on stderr carry their own semver and would win the parse.
  if ! raw="$(databricks version 2>/dev/null)"; then
    echo "STOP: databricks CLI is missing or failed to run -- install or upgrade it via databricks-core first"
    exit 1
  fi
  have="$(printf '%s\n' "$raw" | grep -oE 'v[0-9]+\.[0-9]+\.[0-9]+' | head -n1 | tr -d v)"
  [ -n "$have" ] || have="$(printf '%s\n' "$raw" | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -n1)"
  if [ -z "$have" ]; then
    echo "STOP: could not read a version from 'databricks version' -- upgrade via databricks-core first"
    exit 1
  fi
  case "$raw" in
    *-dev*)
      echo "NOTE: dev build $have -- the floor cannot be checked; ask the user to confirm this build has setup-local"
      ;;
    *)
      if [ "$(printf '%s\n%s\n' "1.16.0" "$have" | sort -V | head -n1)" != "1.16.0" ]; then
        echo "STOP: databricks CLI $have is older than v1.16.0, the floor this skill requires -- older CLIs can run setup-local but not as documented here; upgrade via databricks-core first"
        exit 1
      fi
      ;;
  esac
)

Only after the gate passes, authenticate with the selected profile:

databricks auth describe --profile <PROFILE>

Prefer the latest stable CLI; no online lookup is required. If the gate exits non-zero (older than v1.16.0, missing CLI, or no readable version), or setup-local is absent from help, reports unknown command, or rejects a flag this skill tells you to pass, stop. If it prints NOTE: dev build, the floor is unverifiable -- confirm with the user before continuing. Use databricks-core to upgrade with approval and verify; never recreate setup-local manually.

Use the selected profile for every workspace command. Do not convert another package manager without approval.

2. Confirm the project directory

Confirm the root containing (or intended to contain) pyproject.toml, .venv, and uv.lock; ask if multiple roots are plausible. Use it for preview and apply. It must be greenfield or uv-managed, but need not be writable for preview.

3. Select one target

Choose exactly one branch:

Cluster: use --cluster-id <ID> or --cluster-name <NAME>. If unknown, list clusters with the selected profile and ask; see examples.
Serverless: use --serverless-version <N>. No version-list command exists; ask if unspecified.
Job task: use --job-task <JOB_ID>.<TASK_KEY>. If the task is unknown, run databricks jobs get <JOB_ID> --profile <PROFILE> --output json, present task keys, and ask.
Bundle: a project with databricks.yml. Use databricks-dabs to inspect its root and selected target. Omit compute flags only when that target resolves supported classic or serverless compute; otherwise ask. Add --target <BUNDLE_TARGET> for a named target.

Never combine compute flags. If no branch resolves, ask the user.

4. Preview

Dry-run first; it writes and installs nothing:

databricks environments setup-local --profile <PROFILE> <TARGET_ARGS> --dry-run --output json

For bundles, <TARGET_ARGS> is empty or --target <BUNDLE_TARGET>. Default to normal mode. Use --no-dbconnect only when the user explicitly does not want this command managing databricks-connect; the result's mode still reports constraints-only for it. See JSON output and examples.

5. Obtain approval and apply

Before apply, verify the directory is writable and run uv --version. Ask before installing uv; never silently set DATABRICKS_LOCALENV_AUTO_INSTALL_UV=1 or run a remote installer.

Show the target, versions, warnings, plan.diff, and directory. Explain that apply may:

back up and rewrite pyproject.toml;
install Python and dependencies;
update .venv and uv.lock.

For --serverless-version N in a bundle, also disclose the post-apply job YAML synchronization described below so approval covers both mutations.

Apply only after the user requested provisioning or approves that plan for the named directory. Preserve the directory, profile, target, and mode.

If the directory, profile, target, mode, or project files change after preview, rerun --dry-run, show the new plan, and obtain approval again. Treat its resolved Python, databricks-connect, and managed constraints as authoritative; do not substitute guessed versions. Reconcile user-owned dependency conflicts separately, with approval.

6. Handle the result

ok: true: for --serverless-version N in a bundle, update every existing job environments[].spec.environment_version in its YAML sources to "N", then validate the bundle. Report if none exist; do not invent one. Skip this for cluster and job-task targets. Report target, versions, warnings, and venvPath; prefer uv run <cmd> or derive the platform-specific interpreter from venvPath.
ok: false: use troubleshooting. Ask before diagnostic runs that mutate files and before filing an external issue.
No JSON: inspect stderr for a pre-pipeline CLI, authentication, directory, or cache error.

FAQ

What does the databricks-setup-local skill do?

How do I install databricks-setup-local?

Run npx skills add https://github.com/databricks/databricks-agent-skills --skill databricks-setup-local in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/databricks-setup-local/ (personal) or .claude/skills/databricks-setup-local/ (project).

Which AI agents can use databricks-setup-local?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is databricks-setup-local safe to install?

No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.

Who maintains databricks-setup-local?

It is published by databricks in databricks/databricks-agent-skills. It has 108 recorded installs.