A steadier way to run agent code reviews
A small team used several agents to review each code change and merge only when they all agreed. The same change could pass on Monday and then get blocked on Wednesday, even though the code had not changed. The unstable part was the judging agent, not the .
Making the review prompt bigger and adding more inline rules made the problem worse. In a , the rules became less reliable and the agents drifted back toward loose judgment. The useful fix was to move the review rules out of the prompt and into a skill that every agent loaded on every run.
Each agent reviewed independently without seeing the others’ , and any single “not sure” stopped the merge instead of being averaged into approval. Keeping the rules as a file beside the code made the consensus much less shaky.
Key points
- Several agents reviewing the same code change can still give different results on different runs.
- Adding more rules to a long prompt did not make the review more reliable.
- Putting the review rules in a fixed skill made the agents’ judgment steadier.
- The agents reviewed independently and could not see each other’s .
- One unsure result stopped the merge instead of being averaged into a yes.