By observation of several rounds or PR's in different repos the agents use the local code review skill only before the first PR lands, but as soo as PR review issues are discovered they go back into fix commit PR comment fix commit PR comment over and over, this wastes enormous amounts of wall time and actions runtime and agent tokens.
The skill and agent guard should be written such that every code change is locally reviewed before presented to a github code review agent. That includes the first round of commits bore opening the PR and any code change before it is committed to an open already open PR.
Feature or fix branch: deep review before commit.
PR to dev or main branch: deep review of source branch before opening PR to dev or main.
Fix to dev or main branch with existing open PR: Fix branch deep review before commit to fix branch.
E.g. #1098 is a good example of hundreds of review cycles to land one PR and only one deep review was done before the PR to dev was opened, after that lots of commits to dev untested, and after dev to main PR lots of new fix branches to dev with lots of commits to fix branches untested, then fix branches merged with dev, and more issues repeat.
Refer to [/home/pieter/.claude/projects/-home-pieter-repos-ProjectTemplate/memory/MEMORY.md] including [/home/pieter/.claude/projects/-home-pieter-repos-ProjectTemplate/memory/bash-function-called-via-or-and-suspends-set-e-for-whole-body.md] and [/home/pieter/.claude/projects/-home-pieter-repos-ProjectTemplate/memory/qodo-and-coderabbit-review-behavior-on-projecttemplate.md] for learnings from that review round.
By observation of several rounds or PR's in different repos the agents use the local code review skill only before the first PR lands, but as soo as PR review issues are discovered they go back into fix commit PR comment fix commit PR comment over and over, this wastes enormous amounts of wall time and actions runtime and agent tokens.
The skill and agent guard should be written such that every code change is locally reviewed before presented to a github code review agent. That includes the first round of commits bore opening the PR and any code change before it is committed to an open already open PR.
Feature or fix branch: deep review before commit.
PR to dev or main branch: deep review of source branch before opening PR to dev or main.
Fix to dev or main branch with existing open PR: Fix branch deep review before commit to fix branch.
E.g. #1098 is a good example of hundreds of review cycles to land one PR and only one deep review was done before the PR to dev was opened, after that lots of commits to dev untested, and after dev to main PR lots of new fix branches to dev with lots of commits to fix branches untested, then fix branches merged with dev, and more issues repeat.
Refer to [/home/pieter/.claude/projects/-home-pieter-repos-ProjectTemplate/memory/MEMORY.md] including [/home/pieter/.claude/projects/-home-pieter-repos-ProjectTemplate/memory/bash-function-called-via-or-and-suspends-set-e-for-whole-body.md] and [/home/pieter/.claude/projects/-home-pieter-repos-ProjectTemplate/memory/qodo-and-coderabbit-review-behavior-on-projecttemplate.md] for learnings from that review round.