AI Review is Too Verbose

[imaustink]

When the agent is triggered via the ai-review tag flow, the resulting review is extremely verbose. We need to make it more concise while still providing the important details.

[k5s-bot]

🤖 Starting work on this now. Watch live or take over the session here: https://claude.ai/code/session_011JZVZWHQDM26vgxFwFTC8g

[k5s-bot]

Done. Here's a summary of the work.

What I found

Issue #184 ("AI Review is Too Verbose") concerns the ai-review label flow. That flow is driven by two IntegrationRoute Helm templates in charts/community-components/templates/, whose promptTemplate fields are the entire instruction set the SWE agent follows when reviewing. The verbosity came from the prompts themselves — most notably a recent change (commit 498c43a) that made every change comment carry both a one-click suggestion block and a collapsible AI-agent fix prompt, with no guidance to prioritize or limit comments.

Changes (prompt text only)

• integrationroute-github-pr-labeled-review.yaml — added a conciseness directive (comment only on material issues, raise each once, skip nitpicks/praise/diff-restating, keep any summary to a few sentences) and changed the fix guidance to require exactly one fix mechanism per comment: prefer the suggestion block for localized edits, fall back to the collapsible prompt only for architectural/multi-file fixes.
• integrationroute-github-issue-labeled-review.yaml — added a matching brevity directive (lead with the verdict and decision-changing points; skip nitpicks/praise/restating).

Helm escaping and the suggestion/<details> mechanics are preserved. No code, schema, or test changes — and no test asserts the literal prompt text (routing tests use their own inline templates), so nothing breaks.

PR

https://github.com/imaustink/agent-controller…

---

✅ Opened pull request: imaustink/agent-controller#188