Verify every per-template duplicated/synced file is byte-identical to its source of truth. Use when reviewing a PR that touches a template or a shared source file (.scripts/source/, .claude/skills/), or before pushing such a change, to catch drift — a copy edited directly, or a sync not re-run. Repo-dev tool — NOT synced into templates.
95%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Templates in this repo must be self-contained: Databricks Apps deploys each
template as its own directory (databricks workspace export-dir), so a template
can't import shared code from outside its folder, and symlinks don't survive the
export. Shared files are therefore copied into every template by the sync
scripts, with the single source of truth in .scripts/source/ and .claude/skills/:
.scripts/sync-scripts.py → the shared Python scripts listed in SCRIPTS_TO_SYNC
(quickstart.py, start_app.py, evaluate_agent.py, preflight.py, …) plus
.github/workflows/deploy.yml (with {{BUNDLE_NAME}} substitution). See the
script itself for the authoritative, current list of what it copies..scripts/sync-skills.py → each template's .claude/skills/..scripts/sync-scripts.py also generates the node crash-test fixture's
plain-JS diagnostics
(.scripts/agent-integration-tests/crash-examples/node_crash_app/diagnostics.mjs)
from .scripts/source/diagnostics.ts by stripping TS types. That fixture runs as
raw node index.mjs (no TS build) and must be self-contained for deploy, so it
can't import the shared .ts — generating it keeps its plain-JS twin from drifting.The risk this guards is drift: someone edits a copy directly, or edits the source but forgets to re-run the sync, so the per-template copies silently diverge.
From the repo root:
python .scripts/sync-scripts.py && python .scripts/sync-skills.py && git diff --exit-codeThe sync scripts are Python-stdlib-only (no venv needed) and safe to re-run — they
only rewrite the copies from source. sync-scripts.py additionally shells out to
node (present throughout this Node-heavy repo) to regenerate the crash fixture's
diagnostics twin; it fails loudly if node is missing rather than skipping the
check. They create .scripts/__pycache__/, which is gitignored and does not affect
the check.
Never hand-edit a synced copy. Edit the source — .scripts/source/<file> or
.claude/skills/<skill>/ — then re-run the sync and commit the regenerated copies:
python .scripts/sync-scripts.py
python .scripts/sync-skills.py
git add -A && git commit -m "chore: re-sync <file> from source"The same check runs in CI via .github/workflows/sync-check.yml (it re-runs both
sync scripts and fails on any diff). Until GitHub Actions is enabled on this repo,
run this check locally during code review of any template change or any edit to a
shared source file.
databricks/app-templates
3826450
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.