Content
87%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a well-structured, token-efficient overview that adds genuinely non-obvious PySpark-specific knowledge and routes detail into 8 real, clearly-signaled reference files. Its one weakness is workflow clarity: the build/test sequence is present but lacks an explicit validation checkpoint and feedback loop, which the rubric flags for batch and database-style operations.
Suggestions
Add an explicit validate→fix→retry checkpoint in the Project Setup / Testing flow: state that `uv run pytest` and `uv run ruff check` must pass before running `uv build`, so the build is gated on validation rather than just listed alongside it.
Provide a short verification checklist for completing a data source (e.g., confirm `name()` returns the format string, run unit tests with mocked external calls, then an integration test against the real system) to make the multi-step process and its checkpoints explicit.
Add an explicit checkpoint for batch/streaming write paths — e.g., validate that streaming offsets are non-overlapping before committing, and retry-with-backoff on transient write failures — since external write operations are exactly where the rubric expects a feedback loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence — it explicitly declines to repeat general Python best practices ('still apply but aren't repeated here') and concentrates on non-obvious PySpark-specific gotchas (executor-side imports, serialization-driven flat inheritance), with every section earning its tokens; the single 'Spark 4.0+' mention is load-bearing (the API's availability floor) rather than decorative. | 3 / 3 |
Actionability | Provides fully executable, copy-paste-ready guidance — `uv init`/`uv add`/`uv run pytest`/`uv build` commands, a concrete project layout, a runnable pytest example that patches `requests.post`, and exact class signatures like `class YourBatchWriter(YourWriter, DataSourceWriter)`; the full implementation skeleton is intentionally deferred to a reference, which is appropriate structure rather than a defect. | 3 / 3 |
Workflow Clarity | Steps are clearly sequenced (architecture → project setup → key decisions → testing) and validation commands exist (`uv run pytest`, `uv run ruff check`), but there is no explicit validate→fix→retry feedback loop or gating checkpoint; because the skill targets batch and database-style external operations, the rubric caps workflow_clarity at 2 when feedback loops are missing. | 2 / 3 |
Progressive Disclosure | The body is a concise overview that points to 8 one-level-deep reference files, each clearly signaled with a 'Read when...' trigger; all referenced files were verified to exist, and the few cross-links between references are lateral sibling pointers (e.g. production→authentication) rather than deeply nested chains, matching the 'clear overview with well-signaled one-level-deep references' anchor. | 3 / 3 |
Total | 11 / 12 Passed |