Skip to content

fix(ai-review): make ai-review prompts more concise (#184) - #188

Merged
imaustink merged 1 commit into
mainfrom
fix/ai-review-conciseness-184
Aug 1, 2026
Merged

fix(ai-review): make ai-review prompts more concise (#184)#188
imaustink merged 1 commit into
mainfrom
fix/ai-review-conciseness-184

Conversation

@k5s-bot

@k5s-bot k5s-bot Bot commented Aug 1, 2026

Copy link
Copy Markdown

What & why

Fixes #184 — the ai-review flow produced reviews that were too verbose.

The review is driven entirely by the promptTemplate in the two ai-review IntegrationRoutes, so the fix is to tighten those prompts to emphasize signal over volume while keeping the important detail.

Changes

PR review (integrationroute-github-pr-labeled-review.yaml)

  • Added an explicit conciseness directive: comment only on things that materially affect correctness, security, tests, or maintainability; raise each distinct issue exactly once; skip nitpicks, linter-style preferences, praise, and restating what the diff already shows; hold any overall summary to a few sentences.
  • Changed the per-comment fix guidance to require exactly one fix mechanism per change comment instead of both a suggestion block and a collapsible AI-agent prompt. Prefer the one-click suggestion block for localized edits; fall back to the collapsible prompt only when the fix is architectural or spans several files. Carrying both on every comment was the biggest source of per-comment bulk.

Issue review (integrationroute-github-issue-labeled-review.yaml)

  • Added a matching brevity directive: lead with the verdict and the few points that change the decision; skip nitpicks, praise, and restating the issue.

Scope / risk

Prompt-text-only changes inside the Helm templates. Helm escaping ({{`` `` {{var}} `` ``}}) and the suggestion / collapsible-<details> mechanics are preserved — the mechanics still work, an AI review just uses them more sparingly. No code, config, schema, or test changes; no test references the literal prompt text (the routing tests use their own inline promptTemplates).


Maintainer: apply the ai-review label to this PR to request an automated code review, or the ai-triage label to have the review feedback addressed and the branch brought back in sync with its base. (The automation can't label its own PR, so this needs a human.)

The ai-review flow produced extremely verbose reviews. Two changes to the
review prompt templates make them concise while keeping the important detail:

PR review (integrationroute-github-pr-labeled-review):
- Add a "keep the review concise" directive: comment only on things that
  materially affect correctness, security, tests, or maintainability; raise
  each issue once; skip nitpicks, praise, and restating the diff; hold any
  overall summary to a few sentences.
- Require exactly ONE fix mechanism per change comment instead of both a
  suggestion block AND a collapsible AI-agent prompt. Prefer the one-click
  suggestion block for localized edits; fall back to the collapsible prompt
  only when the fix is architectural/multi-file. This was the largest source
  of per-comment bulk.

Issue review (integrationroute-github-issue-labeled-review):
- Add a matching brevity directive: lead with the verdict and the few points
  that change the decision; skip nitpicks, praise, and restating the issue.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011JZVZWHQDM26vgxFwFTC8g
@k5s-bot k5s-bot Bot mentioned this pull request Aug 1, 2026
@imaustink
imaustink merged commit 383b8f8 into main Aug 1, 2026
5 checks passed
@imaustink
imaustink deleted the fix/ai-review-conciseness-184 branch August 1, 2026 21:24
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.

AI Review is Too Verbose

1 participant