Revisão de Pull Requests Assistida por Inteligência Artificial

um Estudo de Caso no CodeLab-IFPR

Authors

  • João Lucas Pinheiro de Oliveira Instituto Federal do Paraná (IFPR) - Campus Paranavaí Author
  • Ana Beatriz Tavares Malaquias Instituto Federal do Paraná (IFPR) - Campus Paranavaí Author
  • Gustavo Randi da Cruz Instituto Federal do Paraná (IFPR) - Campus Paranavaí Author
  • Eliandro Gomes da Silva Instituto Federal do Paraná (IFPR) - Campus Paranavaí Author
  • Marcelo Figueiredo Terenciani Instituto Federal do Paraná (IFPR) - Campus Paranavaí Author
  • Frank Willian Cardoso de Oliveira Instituto Federal do Paraná (IFPR) - Campus Paranavaí Author

Keywords:

Supervisão humana, Critérios de aceite, Pull Request, Agentes de IA, Revisão de código, Human oversight, Acceptance criteria, Pull request, AI agents, Code review

Abstract

CodeLab-IFPR is an extension laboratory where Software Engineering students build systems for community partners. In one of them, Coopervai, which supports a glass recycling cooperative, reviewing Pull Requests (PRs) became a bottleneck: checking each PR against its corresponding User Story is slow and error-prone, as acceptance criteria are easily overlooked. This experience report describes how the review process was structured with Claude Code, under one rule: the AI does the checking work and the decision remains human. The setup combines three skills (review orchestration, visual validation of the changed screens, and the project's business context), six read-only specialist agents (PHP, Laravel, Blade/Alpine, React, MySQL, and security/LGPD), two guardrails enforced by the tool itself (no public action without the reviewer's "approved N" reply, and blocked reading of secret files), and a script that measures the cost of each review. The verdict has three levels and every satisfied criterion cites evidence. As of September 18, 2026, the core flow had been used on real PRs, the guardrails had been tested in simulated scenarios and in real blocks, and the heuristic pre-analysis of screens is treated as support for the reviewer, not as a replacement.

References

Aikido Security (2025). PromptPwnd: prompt injection em GitHub Actions com agentes de IA. https://www.aikido.dev/blog/promptpwnd-github-actions-ai-agents. Acesso em: 30 set. 2026.

Anthropic (2026). Claude Code: documentação de skills, subagentes e hooks. https://code.claude.com/docs. Acesso em: 30 set. 2026.

Brasil (2018). Lei nº 13.709, de 14 de agosto de 2018: Lei Geral de Proteção de Dados Pessoais (LGPD). Diário Oficial da União.

Guerino, G. C. et al. (2025). Can GPT-4o evaluate usability like human experts? A comparative study on issue identification in heuristic evaluation. In Human-Computer Interaction – INTERACT 2025. Springer.

Nielsen, J. (1994). Enhancing the explanatory power of usability heuristics. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '94), pages 152-158. ACM.

Published

2026-11-10

How to Cite

Pinheiro de Oliveira, J. L., Tavares Malaquias, A. B., Randi da Cruz, G., Gomes da Silva, E., Figueiredo Terenciani, M., & Cardoso de Oliveira, F. W. (2026). Revisão de Pull Requests Assistida por Inteligência Artificial: um Estudo de Caso no CodeLab-IFPR. Information Technology Week, 1(1). https://tecnoif.com.br/periodicos/index.php/setif/article/view/609

Similar Articles

1-10 of 320

You may also start an advanced similarity search for this article.

Most read articles by the same author(s)

<< < 1 2 3 4 5 6 7 8 9 > >>