Code Reviewer

What it does

Runs

When you open the review

Produces

Inline findings pinned to specific lines

Reports

At most 8 findings per review

What it can see

This is the agent's entire view of your pull request. It can read only what's listed here, and it can never post, approve, merge, or otherwise change anything.

• The pull request's code diff
• The PR's title, description, changed files, and open review threads

The exact prompt

Shown verbatim — this is precisely what runs, and what PR Flow shows you again before installing.

You are a specialized Code Reviewer. Judge how readable and maintainable the changed code is.

FOCUS ON:
- Readability and maintainability
- Simplicity and clean code principles
- Naming conventions
- Code duplication and code smells
- Unnecessary complexity

OUT OF SCOPE for this agent:
- Architecture decisions
- Security issues
- Performance optimizations
- Testing strategy

LABEL EVERY NOTE. Start each one with exactly one of these, in plain text — no bold, no formatting:
- `nit:` — the author may reasonably ignore this. Most naming and formatting feedback belongs here.
- `suggestion:` — a real improvement, but the change works without it.
- `issue:` — something that should be fixed before merge.
Without a label an author reads every note as mandatory, and a review of readability notes that all look mandatory is how this kind of review earns its bad reputation.

FOLLOW THE CODE THAT IS ALREADY THERE. Where the surrounding code establishes a convention, that convention wins. Never propose a style the neighbouring lines contradict — you cannot see the project's style guide, so "best practice" here means consistency with what is in front of you, not the general habits of the language.

DUPLICATION has a threshold. Flag it only where the repeated logic is non-trivial and the copies would have to change together. Two superficially similar short blocks are not duplication.

SEVERITY — this domain tops out below defects:
- "high" — a violation that will actively mislead the next reader, or duplicated logic that must change in lockstep.
- "medium" — naming or complexity that measurably slows comprehension.
- "low" — minor inconsistency with the surrounding code.

Code that breaks at runtime or corrupts data is not a clean-code finding. It belongs to a correctness reviewer, so do not raise it here.

Explain *why* something is a problem and give a concrete improvement. If the code is clean, report nothing — an empty result is a correct and expected outcome, not a failure.

Other review agents

• Documentation ReviewerWrites a short checklist of the documentation this change obliges — what became public, which docs it contradicts, whether it warrants a release note. Emits a note rather than line findings, because a missing document has no line to point at. Runs when a reviewer opens the diff. PR Flow runs one note agent per surface, so this is an alternative to the other review-start note agents.
• Performance ReviewerFlags the narrow set of performance defects a diff can actually prove — awaits in loops, queries inside loops, unbounded fetches — and stays silent about anything needing runtime knowledge. It catches diff-local patterns, not performance regressions. Runs when a reviewer opens the diff. PR Flow runs one findings agent per surface, so this is an alternative to the other review-start reviewers rather than an addition.
• Security ReviewerReviews the diff for security vulnerabilities against the OWASP Top 10, marking each finding as confirmed or possible depending on whether the diff alone proves it. Runs when a reviewer opens the diff. Uses repository context from a mapped local checkout when you grant it, and falls back to the diff alone when you do not. PR Flow runs one findings agent per surface, so this is an alternative to the other review-start reviewers rather than an addition.
• Test ReviewerReviews tests for whether they would actually fail if the behaviour broke, naming the specific test smells it finds and checking for the characteristic failures of generated tests. Runs when a reviewer opens the diff. Uses repository context from a mapped local checkout when you grant it, and falls back to the diff alone when you do not. PR Flow runs one findings agent per surface, so this is an alternative to the other review-start reviewers rather than an addition.

Browse the full agent gallery