Skip to content

fix: small correctness/security cleanups in experimental skill scripts#534

Open
jamesbroadhead wants to merge 1 commit into
databricks-solutions:experimentalfrom
jamesbroadhead:ace-review-fixes-experimental
Open

fix: small correctness/security cleanups in experimental skill scripts#534
jamesbroadhead wants to merge 1 commit into
databricks-solutions:experimentalfrom
jamesbroadhead:ace-review-fixes-experimental

Conversation

@jamesbroadhead
Copy link
Copy Markdown

Summary

Batch of small, contained fixes surfaced by a parallel GPT 5.4 + Gemini 3.1 Pro code review. None of these are introduced by recent changes — they are pre-existing issues that ought to land regardless. Each fix is local and review-by-eye.

databricks-agent-bricks/scripts/mas_manager.py

  • _delete now tolerates empty / non-JSON DELETE response bodies (was raising JSONDecodeError on successful deletes).
  • add_examples_batch short-circuits on empty input (ThreadPoolExecutor rejects max_workers=0).
  • _build_agent_list validates uc_function_name has 3 dotted parts (was IndexError on malformed) and rejects missing endpoint_name with a clear error (was sending {"name": null} to the API).
  • main() uses a _parse_json_arg helper that exits cleanly on malformed JSON CLI args, replacing raw json.loads tracebacks.

databricks-execution-compute/scripts/compute.py

  • manage-cluster --action get now keys off the SDK's typed NotFound exception instead of substring-matching "does not exist" in the message. Also flips success to false for the DELETED state so callers gating on success don't treat a missing cluster as a successful lookup.

databricks-app-python/examples/llm_config.py (security)

  • OAuth error no longer interpolates the full token-endpoint payload (which can contain id_token / refresh material). Logs the present key names instead.
  • DATABRICKS_MODEL validation error drops the response.text[:300] echo so server bodies don't end up in operator-visible error text.

databricks-app-python/examples/fm-minimal-chat.py

  • Docstring + app.yaml examples reference the actual filename (fm-minimal-chat.py), not 2-minimal-chat-app.py.

databricks-app-python/examples/fm-parallel-calls.py

  • Guard Speedup division on total_time > 0 to avoid ZeroDivisionError on fast paths.
  • Convert the trailing standalone triple-quoted string (dead code) to real # comments.

databricks-python-sdk/examples/5-serving-and-vector-search.py

  • Replace the [0.1, 0.2, 0.3, ...] literal-Ellipsis vector with a named placeholder + comment explaining it's a stand-in (the literal would fail server-side if copied).

Companion PR

Equivalent PR for main (smaller — mas_manager.py and compute.py don't exist under databricks-skills/ on main): https://github.com/databricks-solutions/ai-dev-kit/pull/new/ace-review-fixes-main (will be opened next)

Test plan

  • python3 -m py_compile on all six modified files
  • CI green
  • Spot-check mas_manager.py delete_mas against a real workspace (DELETE response.json fix)

This pull request and its description were written by Isaac.

Batch of small fixes surfaced by a parallel GPT 5.4 + Gemini 3.1 Pro
review. None of these are introduced by recent changes; they are
pre-existing issues that ought to land regardless.

databricks-agent-bricks/scripts/mas_manager.py:
- `_delete` now tolerates empty / non-JSON DELETE response bodies
  instead of raising JSONDecodeError on success.
- `add_examples_batch` short-circuits on empty input (ThreadPoolExecutor
  rejects max_workers=0).
- `_build_agent_list` validates `uc_function_name` has 3 dotted parts
  and rejects missing `endpoint_name` with a clear error, instead of
  IndexError / `{"name": null}` on the API.
- `main()` uses a `_parse_json_arg` helper that exits cleanly on
  malformed JSON CLI args, replacing raw `json.loads` tracebacks.

databricks-execution-compute/scripts/compute.py:
- `manage-cluster --action get` now keys off the SDK's `NotFound`
  exception type instead of substring-matching "does not exist" in the
  message. Also returns `success: false` for the DELETED state, so
  callers gating on `success` don't treat a missing cluster as a
  successful lookup.

databricks-app-python/examples/llm_config.py:
- OAuth error no longer interpolates the full token-endpoint payload
  (which can contain `id_token` / refresh material). Logs the present
  key names instead.
- DATABRICKS_MODEL validation error drops the `response.text[:300]`
  echo so server bodies don't end up in operator-visible error text.

databricks-app-python/examples/fm-minimal-chat.py:
- Docstring + `app.yaml` examples reference the actual filename
  (`fm-minimal-chat.py`), not `2-minimal-chat-app.py`.

databricks-app-python/examples/fm-parallel-calls.py:
- Guard `Speedup` division on `total_time > 0` to avoid
  ZeroDivisionError on fast paths.
- Convert the trailing standalone triple-quoted string (dead code) to
  real `#` comments.

databricks-python-sdk/examples/5-serving-and-vector-search.py:
- Replace the `[0.1, 0.2, 0.3, ...]` literal-Ellipsis vector with a
  named placeholder + comment explaining it's a stand-in.

This pull request was AI-assisted by Isaac.

Co-authored-by: Isaac
@jamesbroadhead
Copy link
Copy Markdown
Author

Hi @calreynolds — could you take a look at this when you have a moment? Small set of fixes from a code review. Companion PR for main is #535.

— this comment was written by Claude

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant