The permitting team wants clearer PR-review prompts. Which design best reduces noise?
Select an answer to reveal the explanation.
Short Explanation
Give the bot a guest list: bugs and security get in; local style debates wait outside. Report-versus-skip criteria with examples beat a vague confidence dial.
Full Explanation
Clearer PR-review prompts for a permitting team reduce noise by listing report categories (bugs, security, correctness) versus skip categories (local style debates, whitespace preferences), with concrete examples of each. That report-versus-skip design tells Claude what belongs on a municipal PR and what should stay out of the comment stream, which outperforms an undifferentiated confidence dial.
Using only a global confidence slider with no report or skip categories fails because confidence does not encode civic priorities; a model can be highly confident about a style nit. Leaving criteria implicit so reviewers invent categories per pull request fails as process design—inconsistent private taxonomies make CI non-reproducible across engineers. Reporting every possible nit including whitespace preference debates fails by flooding the channel and recreating the trust problem that motivated clearer prompts.
Exam caveat: report/skip lists with examples are prompt design, not a substitute for linters that already enforce style mechanically—prefer skipping style the formatter owns. Operational check: publish the report/skip table in the review prompt or CLAUDE.md, include two examples per side, and sample permitting PRs to confirm style debates drop while bug/security comments remain.