Documentation
¶
Overview ¶
Package review provides AI-powered code review for git diffs. It parses structured feedback from a provider and returns categorized issues, quality scores, and actionable suggestions. It also provides ReviewDiff for lightweight post-task annotation reviews.
Index ¶
Constants ¶
const ( VerdictPass = "PASS" VerdictFail = "FAIL" )
Verdict values for post-task reviews.
const ( SeverityCritical = "critical" // bugs, security holes, data loss SeverityMajor = "major" // logic errors, missing error handling SeverityMinor = "minor" // style, naming, duplication SeveritySuggestion = "suggestion" // optional improvements )
Severity levels for code review issues.
const Author = "ai-reviewer"
Author is the annotation author used for AI code review annotations.
Variables ¶
This section is empty.
Functions ¶
func ExtractVerdict ¶
ExtractVerdict scans review text for "VERDICT: PASS" or "VERDICT: FAIL". Defaults to VerdictPass when no verdict line is found.
func FormatMarkdown ¶
FormatMarkdown renders a Review as a Markdown document.
func GetDiff ¶
GetDiff runs `git diff HEAD~1` in workDir and returns the diff output. Returns an empty string (no error) when there is no parent commit or no diff.
func ReviewDiff ¶
func ReviewDiff(ctx context.Context, p provider.Provider, model string, timeout time.Duration, diff, taskTitle string) (string, error)
ReviewDiff calls the provider with a structured code review prompt for the given diff and task title. It returns the full review text (markdown) and an error. The text ends with a "VERDICT: PASS" or "VERDICT: FAIL" line. If the diff is empty the function returns a short note without calling the provider.
Types ¶
type Issue ¶
type Issue struct {
// Severity is one of: critical, major, minor, suggestion.
Severity string `json:"severity"`
// File is the path of the affected file (may be empty for cross-cutting issues).
File string `json:"file"`
// Line is the approximate line number (0 means unknown).
Line int `json:"line"`
// Title is a short one-line description of the issue.
Title string `json:"title"`
// Detail explains the problem in full.
Detail string `json:"detail"`
// Fix is a concrete suggestion or code snippet to resolve the issue.
Fix string `json:"fix"`
}
Issue is a single code review finding.
type Review ¶
type Review struct {
// Score is an overall code quality score from 1–10.
Score float64 `json:"score"`
// Summary is a 2–3 sentence high-level assessment.
Summary string `json:"summary"`
// Issues is the list of findings ordered by severity (critical first).
Issues []Issue `json:"issues"`
// Praise lists specific things that are well done.
Praise []string `json:"praise"`
// Suggestions lists general improvement ideas that don't map to a single issue.
Suggestions []string `json:"suggestions"`
// TestFeedback is targeted feedback on test coverage and quality.
TestFeedback string `json:"test_feedback"`
// SecurityNotes lists any security-relevant observations.
SecurityNotes []string `json:"security_notes"`
}
Review is the structured output of an AI code review.
func Perform ¶
func Perform(ctx context.Context, prov provider.Provider, model string, timeout time.Duration, diff, taskContext, goal string) (*Review, error)
Perform sends the provided diff to the AI provider and returns a structured review. taskContext is optional — if set it's included in the prompt so the AI can judge whether the code actually solves the intended PM task.
Source Files