Skip to content

fix: make LLM review resilient to individual question failures #550

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
academo merged 1 commit into from
Mar 26, 2026
Merged

fix: make LLM review resilient to individual question failures #550

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 17 additions & 1 deletion pkg/analysis/passes/llmreview/llmreview.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,10 @@ var (
Name: "llm-review-skipped",
Severity: analysis.SuspectedProblem,
}
llmReviewPassed = &analysis.Rule{
Name: "llm-review-passed",
Severity: analysis.OK,
}
)

// blockingAnalyzers contains validators that, if they report errors, should cause
Expand Down Expand Up @@ -66,7 +70,7 @@ var Analyzer = &analysis.Analyzer{
Name: "llmreview",
Requires: append([]*analysis.Analyzer{sourcecode.Analyzer}, blockingAnalyzers...),
Run: run,
Rules: []*analysis.Rule{llmIssueFound, llmReviewSkipped},
Rules: []*analysis.Rule{llmIssueFound, llmReviewSkipped, llmReviewPassed},
ReadmeInfo: analysis.ReadmeInfo{
Name: "LLM Review",
Description: "Runs the code through an LLM to check for security issues or disallowed usage.",
Expand Down Expand Up @@ -213,6 +217,7 @@ func run(pass *analysis.Pass) (any, error) {
return nil, nil
}

issuesFound := 0
for _, answer := range mandatoryAnswers {
if answer.ShortAnswer != answer.ExpectedShortAnswer {
detail := buildDetailString(answer)
Expand All @@ -222,6 +227,7 @@ func run(pass *analysis.Pass) (any, error) {
fmt.Sprintf("LLM flagged: %s", answer.Question),
detail,
)
issuesFound++
}
}

Expand All @@ -242,9 +248,19 @@ func run(pass *analysis.Pass) (any, error) {
fmt.Sprintf("LLM suggestion: %s", answer.Question),
detail,
)
issuesFound++
}
}

if issuesFound == 0 && llmReviewPassed.ReportAll {
pass.ReportResult(
pass.AnalyzerName,
llmReviewPassed,
"LLM review completed without concerns",
"",
)
}

return nil, nil
}

Expand Down
2 changes: 1 addition & 1 deletion pkg/llmclient/agentic_client.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@ const (
maxLLMRetries = 3
maxConsecutiveNoTools = 5
retryDelay = 2 * time.Second
llmCallTimeout = 20 * time.Second
llmCallTimeout = 90 * time.Second

budgetNudgePrompt = `You have only %d tool calls remaining. Wrap up your investigation and call submit_answer now with whatever information you have gathered so far.`

Expand Down
62 changes: 24 additions & 38 deletions pkg/llmvalidate/llmvalidate.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,46 +191,32 @@ func (c *Client) AskLLMAboutCode(
for _, question := range questions {
userPrompt := fmt.Sprintf("Answer this question based on the source code files provided in the system prompt: %s", question.Question)

var answer LLMAnswer
var lastErr error

for retries := 3; retries > 0; retries-- {
agenticAnswers, err := agenticClient.CallLLM(c.ctx, []string{userPrompt}, absCodePath)
if err != nil {
lastErr = err
logme.DebugFln("Error calling LLM (retries left: %d): %v", retries-1, err)
continue
}

if len(agenticAnswers) == 0 {
lastErr = fmt.Errorf("No answer returned from LLM for question: %s", question.Question)
logme.DebugFln("No answer returned (retries left: %d)", retries-1)
continue
}

agenticAnswer := agenticAnswers[0]

// Check if the agentic client reported an error for this question
if agenticAnswer.Error != "" {
lastErr = fmt.Errorf("LLM error for question %q: %s", question.Question, agenticAnswer.Error)
logme.DebugFln("LLM error (retries left: %d): %s", retries-1, agenticAnswer.Error)
continue
}

answer = LLMAnswer{
Question: agenticAnswer.Question,
Answer: agenticAnswer.Answer,
ShortAnswer: agenticAnswer.ShortAnswer,
Files: agenticAnswer.Files,
CodeSnippet: mergeNewlines(agenticAnswer.CodeSnippet),
ExpectedShortAnswer: question.ExpectedAnswer,
}
lastErr = nil
break
agenticAnswers, err := agenticClient.CallLLM(c.ctx, []string{userPrompt}, absCodePath)
Copy link
Copy Markdown
Collaborator Author

@academo academo Mar 26, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

the main change here is removing the for retries := 3; retries > 0; retries-- loop

this was retrying 3 times but CallLLM already has retrying handling internally (3 times)

if err != nil {
logme.DebugFln("Error calling LLM for question %q: %v. Skipping.", question.Question, err)
continue
}

if len(agenticAnswers) == 0 {
logme.DebugFln("No answer returned from LLM for question: %s. Skipping.", question.Question)
continue
}

agenticAnswer := agenticAnswers[0]

// Check if the agentic client reported an error for this question
if agenticAnswer.Error != "" {
logme.DebugFln("LLM error for question %q: %s. Skipping.", question.Question, agenticAnswer.Error)
continue
}

if lastErr != nil {
return nil, fmt.Errorf("Failed to generate answer after 3 retries: %w", lastErr)
answer := LLMAnswer{
Question: agenticAnswer.Question,
Answer: agenticAnswer.Answer,
ShortAnswer: agenticAnswer.ShortAnswer,
Files: agenticAnswer.Files,
CodeSnippet: mergeNewlines(agenticAnswer.CodeSnippet),
ExpectedShortAnswer: question.ExpectedAnswer,
}

logme.DebugFln("Answer: %v", prettyprint.SPrint(answer))
Expand Down
Loading