Improve target update conflict comment template
Addresses feedback from @rasamhossain in about-gitlab-com-fork!57 (comment):
- The message has to be clear and concise - sound like human
- Needs to be human readable
- What are the next actions on this MR
- Approved by the specialist to say this is an agent output
Changes to the conflict comment template in system-prompt.md:
| Before | After |
|---|---|
| Verbose per-conflict blocks with linguistic analysis + redundant summary table | Single scannable table per file with truncated values |
Separate Specialist: line above table (ambiguous with multiple specialists) |
Edited by column in table (handles multi-specialist cases naturally) |
| No explicit next actions | Checkbox next steps (remove keys or confirm with specialist) |
| No agent attribution | Footer linking to GDATP project |
| No tone guidance for the agent | Explicit formatting rules: "Be concise. Write like a human reviewer, not an essay." |
| Generic "resolve each conflict" | Clear verdict line distinguishing "English unchanged" (regression) vs "English changed" (needs review) |
Example output (after):
⚠️ Target update conflicts detected — This MR overwrites 7 specialist edit(s) in 1 file(s).ja-jp —
content/ja-jp/assessments/.../results-exploring.yml
Key Specialist version This MR Edited by seo.titleアセスメント結果 評価結果 Emi Kimura seo.descriptionエクスプロアリング（導入検討）ステージの… AIの近代化のステージを探る… Emi Kimura AssessmentsResultsHero.headerエクスプロアリング（導入検討） 探検 Emi Kimura English source unchanged since the specialist edits — merging would be a regression.
Next steps:
- Remove the conflicting keys from this MR and re-submit, or
- Confirm with the language specialist that the new values are intentional
Generated by GitLab Translation Agent. Conflicts detected via git history query.
Edited by Rasam Hossain