fix: repair unescaped quotes in LLM JSON responses (#45)

fix: repair unescaped quotes in LLM JSON responses

Add repairJSON fallback that handles unescaped quotes in LLM string
values using first-valid-candidate heuristic with structural lookahead.

Reviewed-by: sonnet-review-bot
Reviewed-by: gpt-review-bot
Reviewed-by: security-review-bot

.gitea/workflows/ci.yml

@@ -28,12 +28,33 @@ jobs:
         include:
           - name: sonnet
             token_secret: SONNET_REVIEW_TOKEN
-            model: gpt-5
+            provider: anthropic
+            llm_path: /anthropic/v1
+            model: claude-sonnet-4-6
           - name: gpt
             token_secret: GPT_REVIEW_TOKEN
+            provider: openai
+            llm_path: /openai/v1
+            model: gpt-5
+          - name: gpt41
+            token_secret: GPT_REVIEW_TOKEN
+            provider: openai
+            llm_path: /openai/v1
             model: gpt-4.1
+          - name: gpt5-mini
+            token_secret: GPT_REVIEW_TOKEN
+            provider: openai
+            llm_path: /openai/v1
+            model: gpt-5-mini
+          - name: gpt41-mini
+            token_secret: GPT_REVIEW_TOKEN
+            provider: openai
+            llm_path: /openai/v1
+            model: gpt-4.1-mini
           - name: security
             token_secret: SECURITY_REVIEW_TOKEN
+            provider: openai
+            llm_path: /openai/v1
             model: gpt-5
             system_prompt_file: SECURITY_REVIEW.md
     steps:
@@ -49,9 +70,10 @@ jobs:
           PR_NUMBER: ${{ github.event.pull_request.number }}
           REVIEWER_TOKEN: ${{ secrets[matrix.token_secret] }}
           REVIEWER_NAME: ${{ matrix.name }}
-          LLM_BASE_URL: ${{ secrets.LLM_BASE_URL }}
+          LLM_BASE_URL: ${{ secrets.LLM_BASE_URL }}${{ matrix.llm_path }}
           LLM_API_KEY: ${{ secrets.LLM_API_KEY }}
           LLM_MODEL: ${{ matrix.model }}
+          LLM_PROVIDER: ${{ matrix.provider }}
           CONVENTIONS_FILE: "CONVENTIONS.md"
           PATTERNS_REPO: "rodin/go-patterns"
           PATTERNS_FILES: "README.md,patterns/"

cmd/review-bot/main.go

@@ -319,31 +319,32 @@ func main() {
 	// 1. POST new review first (gets non-stale approval badge on HEAD)
 	// 2. Then supersede old review with link to the new one
 	// Order matters: post first so we have the new review's URL for the supersede message.
-	var existingReview *gitea.Review
-	var existingCommentID int64
+	var oldReviews []gitea.Review
 	if *reviewerName != "" {
 		existingReviews, err := giteaClient.ListReviews(ctx, owner, repoName, prNumber)
 		if err != nil {
 			slog.Warn("could not list existing reviews", "pr", prNumber, "error", err)
 		} else {
-			sharedToken := hasSharedToken(existingReviews, sentinel)
-			if !sharedToken {
-				existingReview = findOwnReview(existingReviews, sentinel)
-				if existingReview != nil {
-					cid, err := giteaClient.GetTimelineReviewCommentID(ctx, owner, repoName, prNumber, sentinel)
-					if err != nil {
-						slog.Warn("could not find old review comment ID for supersede", "error", err)
-						existingReview = nil // can't supersede without comment ID
-					} else {
-						existingCommentID = cid
-					}
-				}
-			} else {
+			if hasSharedToken(existingReviews, sentinel) {
 				slog.Warn("shared token mode: skipping supersede to avoid clobbering sibling review")
+			} else {
+				oldReviews = findAllOwnReviews(existingReviews, sentinel)
 			}
 		}
 	}
 
+	// Self-request as reviewer (ensures we appear in required-reviewer checks)
+	authUser, err := giteaClient.GetAuthenticatedUser(ctx)
+	if err != nil {
+		slog.Warn("could not determine authenticated user for reviewer self-request", "error", err)
+	} else if authUser != "" {
+		if err := giteaClient.RequestReviewer(ctx, owner, repoName, prNumber, authUser); err != nil {
+			slog.Warn("could not self-request as reviewer", "user", authUser, "error", err)
+		} else {
+			slog.Debug("self-requested as reviewer", "user", authUser, "pr", prNumber)
+		}
+	}
+
 	// POST new review
 	slog.Info("posting review", "event", event, "pr", prNumber)
 	posted, err := giteaClient.PostReview(ctx, owner, repoName, prNumber, event, reviewBody, inlineComments)
@@ -353,14 +354,46 @@ func main() {
 	}
 	slog.Info("review posted", "review_id", posted.ID, "user", posted.User.Login, "pr", prNumber)
 
-	// Supersede old review with link to the new one
-	if existingReview != nil && existingCommentID > 0 {
+	// Supersede all old reviews with link to the new one
+	if len(oldReviews) > 0 {
 		newReviewURL := fmt.Sprintf("%s/%s/%s/pulls/%d#pullrequestreview-%d", strings.TrimRight(*giteaURL, "/"), owner, repoName, prNumber, posted.ID)
-		supersededBody := buildSupersededBody(existingReview.Body, existingReview.CommitID, newReviewURL, sentinel)
-		if err := giteaClient.EditComment(ctx, owner, repoName, existingCommentID, supersededBody); err != nil {
-			slog.Warn("could not mark old review as superseded", "comment_id", existingCommentID, "error", err)
-		} else {
-			slog.Info("marked old review as superseded", "old_state", existingReview.State, "new_review_id", posted.ID, "pr", prNumber)
+		for _, oldReview := range oldReviews {
+			cid, err := giteaClient.GetTimelineReviewCommentIDForReview(ctx, owner, repoName, prNumber, oldReview.ID)
+			if err != nil {
+				slog.Warn("could not find comment ID for old review", "review_id", oldReview.ID, "error", err)
+				continue
+			}
+			supersededBody := buildSupersededBody(oldReview.Body, oldReview.CommitID, newReviewURL, sentinel)
+			if err := giteaClient.EditComment(ctx, owner, repoName, cid, supersededBody); err != nil {
+				slog.Warn("could not mark old review as superseded", "review_id", oldReview.ID, "comment_id", cid, "error", err)
+				continue
+			}
+			slog.Info("marked old review as superseded", "review_id", oldReview.ID, "new_review_id", posted.ID, "pr", prNumber)
+
+			// Resolve old review's inline comments
+			oldComments, err := giteaClient.ListReviewComments(ctx, owner, repoName, prNumber, oldReview.ID)
+			if err != nil {
+				slog.Warn("could not list old review comments for resolution", "review_id", oldReview.ID, "error", err)
+				continue
+			}
+			resolved, failed := 0, 0
+			for _, c := range oldComments {
+				if c.ID == 0 {
+					continue
+				}
+				if err := giteaClient.ResolveComment(ctx, owner, repoName, c.ID); err != nil {
+					slog.Debug("could not resolve inline comment", "comment_id", c.ID, "error", err)
+					failed++
+				} else {
+					resolved++
+				}
+			}
+			if resolved > 0 {
+				slog.Info("resolved old inline comments", "review_id", oldReview.ID, "count", resolved, "pr", prNumber)
+			}
+			if failed > 0 {
+				slog.Warn("some inline comments could not be resolved", "review_id", oldReview.ID, "failed", failed, "pr", prNumber)
+			}
 		}
 	}
 
@@ -586,21 +619,34 @@ func extractSentinelName(body string) string {
 	return rest[:end]
 }
 
-// findOwnReview locates a review matching the given sentinel in its body.
+// findOwnReview locates the most recent non-superseded review matching the sentinel.
 func findOwnReview(reviews []gitea.Review, sentinel string) *gitea.Review {
 	var best *gitea.Review
 	for i := range reviews {
 		if !strings.Contains(reviews[i].Body, sentinel) {
 			continue
 		}
-		// Skip superseded reviews (they contain our sentinel in the collapsed body)
 		if strings.Contains(reviews[i].Body, "~~Original review~~") {
 			continue
 		}
-		// Take the highest ID (most recent)
 		if best == nil || reviews[i].ID > best.ID {
 			best = &reviews[i]
 		}
 	}
 	return best
 }
+
+// findAllOwnReviews returns all non-superseded reviews matching the sentinel.
+func findAllOwnReviews(reviews []gitea.Review, sentinel string) []gitea.Review {
+	var result []gitea.Review
+	for i := range reviews {
+		if !strings.Contains(reviews[i].Body, sentinel) {
+			continue
+		}
+		if strings.Contains(reviews[i].Body, "~~Original review~~") {
+			continue
+		}
+		result = append(result, reviews[i])
+	}
+	return result
+}

cmd/review-bot/main_test.go

@@ -841,3 +841,24 @@ func cleanEnv() []string {
 	}
 	return env
 }
+
+func TestFindAllOwnReviews(t *testing.T) {
+	reviews := []gitea.Review{
+		{ID: 1, Body: "<!-- review-bot:sonnet -->\nfirst review"},
+		{ID: 2, Body: "<!-- review-bot:gpt -->\nother bot"},
+		{ID: 3, Body: "<!-- review-bot:sonnet -->\nsecond review"},
+		{ID: 4, Body: "~~Original review~~\n<!-- review-bot:sonnet -->\nsuperseded"},
+		{ID: 5, Body: "<!-- review-bot:sonnet -->\nthird review"},
+	}
+
+	got := findAllOwnReviews(reviews, "<!-- review-bot:sonnet -->")
+	if len(got) != 3 {
+		t.Fatalf("findAllOwnReviews() returned %d, want 3", len(got))
+	}
+	wantIDs := []int64{1, 3, 5}
+	for i, r := range got {
+		if r.ID != wantIDs[i] {
+			t.Errorf("got[%d].ID = %d, want %d", i, r.ID, wantIDs[i])
+		}
+	}
+}

gitea/client.go

@@ -82,6 +82,7 @@ type ChangedFile struct {
 
 // ReviewComment represents an inline comment to attach to a review.
 type ReviewComment struct {
+	ID          int64  `json:"id,omitempty"`
 	Path        string `json:"path"`
 	NewPosition int64  `json:"new_position"`
 	Body        string `json:"body"`
@@ -425,6 +426,68 @@ func (c *Client) GetTimelineReviewCommentID(ctx context.Context, owner, repo str
 	return 0, fmt.Errorf("no timeline event found with sentinel")
 }
 
+// GetTimelineReviewCommentIDForReview finds the timeline comment ID for a
+// specific review by matching its body content in the timeline.
+func (c *Client) GetTimelineReviewCommentIDForReview(ctx context.Context, owner, repo string, number int, reviewID int64) (int64, error) {
+	// Use the reviews API to get the review body, then find in timeline
+	reqURL := fmt.Sprintf("%s/api/v1/repos/%s/%s/pulls/%d/reviews/%d",
+		c.baseURL,
+		url.PathEscape(owner),
+		url.PathEscape(repo),
+		number,
+		reviewID)
+	body, err := c.doGet(ctx, reqURL)
+	if err != nil {
+		return 0, fmt.Errorf("get review %d: %w", reviewID, err)
+	}
+	var review struct {
+		Body string `json:"body"`
+		User struct {
+			Login string `json:"login"`
+		} `json:"user"`
+	}
+	if err := json.Unmarshal(body, &review); err != nil {
+		return 0, fmt.Errorf("parse review %d: %w", reviewID, err)
+	}
+	if review.Body == "" {
+		return 0, fmt.Errorf("review %d has empty body", reviewID)
+	}
+
+	// Use a prefix for matching (handles minor trailing whitespace differences)
+	matchPrefix := review.Body
+	if len(matchPrefix) > 200 {
+		matchPrefix = matchPrefix[:200]
+	}
+
+	const pageSize = 50
+	for page := 1; ; page++ {
+		timelineURL := fmt.Sprintf("%s/api/v1/repos/%s/%s/issues/%d/timeline?limit=%d&page=%d",
+			c.baseURL,
+			url.PathEscape(owner),
+			url.PathEscape(repo),
+			number,
+			pageSize,
+			page)
+		tlBody, err := c.doGet(ctx, timelineURL)
+		if err != nil {
+			return 0, fmt.Errorf("get timeline (page %d): %w", page, err)
+		}
+		var events []TimelineEvent
+		if err := json.Unmarshal(tlBody, &events); err != nil {
+			return 0, fmt.Errorf("parse timeline (page %d): %w", page, err)
+		}
+		for _, ev := range events {
+			if ev.Type == "review" && ev.User.Login == review.User.Login && strings.HasPrefix(ev.Body, matchPrefix) {
+				return ev.ID, nil
+			}
+		}
+		if len(events) < pageSize {
+			break
+		}
+	}
+	return 0, fmt.Errorf("no timeline event found for review %d", reviewID)
+}
+
 // EditComment updates the body of an issue/review comment.
 func (c *Client) EditComment(ctx context.Context, owner, repo string, commentID int64, newBody string) error {
 	reqURL := fmt.Sprintf("%s/api/v1/repos/%s/%s/issues/comments/%d",
@@ -460,3 +523,113 @@ func (c *Client) EditComment(ctx context.Context, owner, repo string, commentID
 	}
 	return nil
 }
+
+// GetAuthenticatedUser returns the login of the user authenticated by the token.
+func (c *Client) GetAuthenticatedUser(ctx context.Context) (string, error) {
+	reqURL := fmt.Sprintf("%s/api/v1/user", c.baseURL)
+	body, err := c.doGet(ctx, reqURL)
+	if err != nil {
+		return "", fmt.Errorf("get authenticated user: %w", err)
+	}
+	var result struct {
+		Login string `json:"login"`
+	}
+	if err := json.Unmarshal(body, &result); err != nil {
+		return "", fmt.Errorf("parse user response: %w", err)
+	}
+	return result.Login, nil
+}
+
+// RequestReviewer adds the given user as a requested reviewer on a pull request.
+// This is idempotent — requesting an already-requested reviewer is a no-op.
+func (c *Client) RequestReviewer(ctx context.Context, owner, repo string, number int, reviewer string) error {
+	reqURL := fmt.Sprintf("%s/api/v1/repos/%s/%s/pulls/%d/requested_reviewers",
+		c.baseURL,
+		url.PathEscape(owner),
+		url.PathEscape(repo),
+		number)
+
+	payload := struct {
+		Reviewers []string `json:"reviewers"`
+	}{Reviewers: []string{reviewer}}
+	data, err := json.Marshal(payload)
+	if err != nil {
+		return fmt.Errorf("marshal reviewer request: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, reqURL, bytes.NewReader(data))
+	if err != nil {
+		return fmt.Errorf("create reviewer request: %w", err)
+	}
+	req.Header.Set("Authorization", "token "+c.token)
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := c.http.Do(req)
+	if err != nil {
+		return fmt.Errorf("request reviewer: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK && resp.StatusCode != http.StatusCreated && resp.StatusCode != http.StatusNoContent {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 256))
+		return fmt.Errorf("request reviewer failed (status %d): %s", resp.StatusCode, body)
+	}
+	return nil
+}
+
+// ListReviewComments returns the inline comments attached to a specific review.
+// Paginates through all pages.
+func (c *Client) ListReviewComments(ctx context.Context, owner, repo string, prNumber int, reviewID int64) ([]ReviewComment, error) {
+	const pageSize = 50
+	var all []ReviewComment
+	for page := 1; ; page++ {
+		reqURL := fmt.Sprintf("%s/api/v1/repos/%s/%s/pulls/%d/reviews/%d/comments?limit=%d&page=%d",
+			c.baseURL,
+			url.PathEscape(owner),
+			url.PathEscape(repo),
+			prNumber,
+			reviewID,
+			pageSize,
+			page)
+		body, err := c.doGet(ctx, reqURL)
+		if err != nil {
+			return nil, fmt.Errorf("list review comments (page %d): %w", page, err)
+		}
+		var batch []ReviewComment
+		if err := json.Unmarshal(body, &batch); err != nil {
+			return nil, fmt.Errorf("parse review comments (page %d): %w", page, err)
+		}
+		all = append(all, batch...)
+		if len(batch) < pageSize {
+			break
+		}
+	}
+	return all, nil
+}
+
+// ResolveComment marks an inline review comment as resolved.
+func (c *Client) ResolveComment(ctx context.Context, owner, repo string, commentID int64) error {
+	reqURL := fmt.Sprintf("%s/api/v1/repos/%s/%s/pulls/comments/%d/resolve",
+		c.baseURL,
+		url.PathEscape(owner),
+		url.PathEscape(repo),
+		commentID)
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, reqURL, nil)
+	if err != nil {
+		return fmt.Errorf("create resolve request: %w", err)
+	}
+	req.Header.Set("Authorization", "token "+c.token)
+
+	resp, err := c.http.Do(req)
+	if err != nil {
+		return fmt.Errorf("resolve comment: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK && resp.StatusCode != http.StatusCreated && resp.StatusCode != http.StatusNoContent {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 256))
+		return fmt.Errorf("resolve comment failed (status %d): %s", resp.StatusCode, body)
+	}
+	return nil
+}

gitea/client_test.go

@@ -5,8 +5,10 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
 	"net/http"
 	"net/http/httptest"
+	"strings"
 	"testing"
 )
 
@@ -602,3 +604,142 @@ func TestIsNotFound(t *testing.T) {
 		})
 	}
 }
+
+func TestGetAuthenticatedUser(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/api/v1/user" {
+			t.Errorf("unexpected path: %s", r.URL.Path)
+			http.NotFound(w, r)
+			return
+		}
+		if r.Header.Get("Authorization") != "token test-token" {
+			t.Error("missing or wrong auth header")
+		}
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprint(w, `{"login":"my-bot","id":42}`)
+	}))
+	defer server.Close()
+
+	client := NewClient(server.URL, "test-token")
+	login, err := client.GetAuthenticatedUser(context.Background())
+	if err != nil {
+		t.Fatalf("GetAuthenticatedUser() error = %v", err)
+	}
+	if login != "my-bot" {
+		t.Errorf("login = %q, want %q", login, "my-bot")
+	}
+}
+
+func TestRequestReviewer(t *testing.T) {
+	var gotBody []byte
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodPost {
+			t.Errorf("expected POST, got %s", r.Method)
+		}
+		expected := "/api/v1/repos/owner/repo/pulls/7/requested_reviewers"
+		if r.URL.Path != expected {
+			t.Errorf("path = %q, want %q", r.URL.Path, expected)
+		}
+		gotBody, _ = io.ReadAll(r.Body)
+		w.WriteHeader(http.StatusCreated)
+	}))
+	defer server.Close()
+
+	client := NewClient(server.URL, "test-token")
+	err := client.RequestReviewer(context.Background(), "owner", "repo", 7, "bot-user")
+	if err != nil {
+		t.Fatalf("RequestReviewer() error = %v", err)
+	}
+	if !strings.Contains(string(gotBody), `"bot-user"`) {
+		t.Errorf("body = %s, want to contain bot-user", gotBody)
+	}
+}
+
+func TestRequestReviewer_204(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNoContent)
+	}))
+	defer server.Close()
+
+	client := NewClient(server.URL, "test-token")
+	err := client.RequestReviewer(context.Background(), "owner", "repo", 1, "user")
+	if err != nil {
+		t.Fatalf("RequestReviewer() should accept 204, got error = %v", err)
+	}
+}
+
+func TestRequestReviewer_Error(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusForbidden)
+		fmt.Fprint(w, "no permission")
+	}))
+	defer server.Close()
+
+	client := NewClient(server.URL, "test-token")
+	err := client.RequestReviewer(context.Background(), "owner", "repo", 1, "user")
+	if err == nil {
+		t.Fatal("expected error for 403 response")
+	}
+	if !strings.Contains(err.Error(), "403") {
+		t.Errorf("error should mention status code: %v", err)
+	}
+}
+
+func TestListReviewComments(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if !strings.Contains(r.URL.Path, "/pulls/1/reviews/42/comments") {
+			t.Errorf("unexpected path: %s", r.URL.Path)
+		}
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprint(w, `[{"id":100,"path":"main.go","new_position":5,"body":"finding"},{"id":101,"path":"lib.go","new_position":10,"body":"another"}]`)
+	}))
+	defer server.Close()
+
+	client := NewClient(server.URL, "test-token")
+	comments, err := client.ListReviewComments(context.Background(), "owner", "repo", 1, 42)
+	if err != nil {
+		t.Fatalf("ListReviewComments() error = %v", err)
+	}
+	if len(comments) != 2 {
+		t.Fatalf("got %d comments, want 2", len(comments))
+	}
+	if comments[0].ID != 100 {
+		t.Errorf("comments[0].ID = %d, want 100", comments[0].ID)
+	}
+	if comments[1].Path != "lib.go" {
+		t.Errorf("comments[1].Path = %q, want %q", comments[1].Path, "lib.go")
+	}
+}
+
+func TestResolveComment(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodPost {
+			t.Errorf("expected POST, got %s", r.Method)
+		}
+		if !strings.Contains(r.URL.Path, "/pulls/comments/99/resolve") {
+			t.Errorf("unexpected path: %s", r.URL.Path)
+		}
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer server.Close()
+
+	client := NewClient(server.URL, "test-token")
+	err := client.ResolveComment(context.Background(), "owner", "repo", 99)
+	if err != nil {
+		t.Fatalf("ResolveComment() error = %v", err)
+	}
+}
+
+func TestResolveComment_Error(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+		fmt.Fprint(w, "not found")
+	}))
+	defer server.Close()
+
+	client := NewClient(server.URL, "test-token")
+	err := client.ResolveComment(context.Background(), "owner", "repo", 99)
+	if err == nil {
+		t.Fatal("expected error for 404 response")
+	}
+}

review/parser.go

@@ -29,7 +29,12 @@ func ParseResponse(response string) (*ReviewResult, error) {
 
 	var result ReviewResult
 	if err := json.Unmarshal([]byte(cleaned), &result); err != nil {
-		return nil, fmt.Errorf("parse LLM response as JSON: %w\nRaw response: %s", err, response)
+		// LLMs sometimes produce JSON with unescaped quotes inside string values.
+		// Try to repair before giving up.
+		repaired := repairJSON(cleaned)
+		if err2 := json.Unmarshal([]byte(repaired), &result); err2 != nil {
+			return nil, fmt.Errorf("parse LLM response as JSON: %w\nRaw response: %s", err, response)
+		}
 	}
 
 	// Validate verdict
@@ -74,3 +79,230 @@ func extractJSON(s string) string {
 	s = strings.TrimSpace(s)
 	return s
 }
+
+// repairJSON attempts to fix common LLM JSON issues:
+// - Unescaped double quotes inside string values
+//
+// Strategy: walk the JSON structurally. Object keys are parsed normally (LLMs
+// get those right). For string VALUES, we find all candidate closing quotes and
+// pick the LAST one that leaves valid JSON structure afterward — maximizing
+// string content, which is the correct bias for the "LLM put unescaped quotes
+// in a string value" failure mode.
+func repairJSON(s string) string {
+	runes := []rune(s)
+	var out strings.Builder
+	out.Grow(len(s) + 64)
+
+	i := 0
+	for i < len(runes) {
+		c := runes[i]
+
+		if c != '"' {
+			out.WriteRune(c)
+			i++
+			continue
+		}
+
+		// We hit an opening quote. Determine if this is a key or a value.
+		// Keys: the standard JSON parser in LLMs gets keys right, so we parse
+		// them normally (first unescaped quote closes).
+		// Values: may contain unescaped quotes — use the repair heuristic.
+		isValue := isValuePosition(runes, i)
+
+		if !isValue {
+			// Parse key/simple string normally
+			out.WriteRune('"')
+			i++
+			for i < len(runes) {
+				ch := runes[i]
+				if ch == '\\' && i+1 < len(runes) {
+					out.WriteRune(ch)
+					i++
+					out.WriteRune(runes[i])
+					i++
+					continue
+				}
+				if ch == '"' {
+					out.WriteRune('"')
+					i++
+					break
+				}
+				out.WriteRune(ch)
+				i++
+			}
+			continue
+		}
+
+		// Value string — find the correct close using last-valid-candidate heuristic
+		out.WriteRune('"')
+		i++
+
+		closeIdx := findClosingQuote(runes, i)
+
+		// Write everything between open and close, escaping interior quotes
+		for j := i; j < closeIdx; j++ {
+			ch := runes[j]
+			if ch == '\\' && j+1 < closeIdx {
+				// Already-escaped sequence — pass through
+				out.WriteRune(ch)
+				j++
+				out.WriteRune(runes[j])
+			} else if ch == '"' {
+				out.WriteRune('\\')
+				out.WriteRune('"')
+			} else {
+				out.WriteRune(ch)
+			}
+		}
+
+		// Write the closing quote
+		out.WriteRune('"')
+		i = closeIdx + 1
+	}
+
+	return out.String()
+}
+
+// isValuePosition determines if the quote at position i is opening a JSON value
+// string (as opposed to an object key). We only apply repair to values that
+// follow ':' since those are the free-text fields where LLMs produce unescaped
+// quotes. Array elements and keys are left alone (parsed normally).
+func isValuePosition(runes []rune, i int) bool {
+	// Look backward, skipping whitespace, for the preceding structural char
+	j := i - 1
+	for j >= 0 && (runes[j] == ' ' || runes[j] == '\t' || runes[j] == '\n' || runes[j] == '\r') {
+		j--
+	}
+	if j < 0 {
+		return false
+	}
+	// After ':' → definitely a value
+	return runes[j] == ':'
+}
+
+// findClosingQuote finds the index of the true closing quote for a JSON string
+// value starting at position start (the character after the opening quote).
+// It collects all unescaped quote candidates and returns the FIRST one that
+// produces valid JSON continuation (deeper lookahead verifies the next token).
+func findClosingQuote(runes []rune, start int) int {
+	// Collect all candidate positions for the closing quote.
+	var candidates []int
+	for j := start; j < len(runes); j++ {
+		if runes[j] == '\\' {
+			j++ // skip escaped character
+			continue
+		}
+		if runes[j] == '"' {
+			candidates = append(candidates, j)
+		}
+	}
+
+	if len(candidates) == 0 {
+		return len(runes)
+	}
+
+	if len(candidates) == 1 {
+		return candidates[0]
+	}
+
+	// Try candidates from FIRST to LAST. The correct closing quote is the
+	// earliest one that produces valid JSON structure after it (verified by
+	// deeper lookahead that checks the next token is a valid JSON start).
+	for _, idx := range candidates {
+		if isValidJSONAfterClose(runes, idx+1) {
+			return idx
+		}
+	}
+
+	// Fallback: return the last candidate
+	return candidates[len(candidates)-1]
+}
+
+// isValidJSONAfterClose checks whether the runes after a candidate closing quote
+// look like valid JSON continuation for a VALUE string. Since we only use this
+// for value positions, ':' is NOT a valid continuation (values are never keys).
+// Checks deeper structure to avoid being fooled by JSON-like content in strings.
+func isValidJSONAfterClose(runes []rune, pos int) bool {
+	j := pos
+	for j < len(runes) && (runes[j] == ' ' || runes[j] == '\t' || runes[j] == '\n' || runes[j] == '\r') {
+		j++
+	}
+
+	if j >= len(runes) {
+		return true
+	}
+
+	next := runes[j]
+	if next == '}' || next == ']' {
+		// Closing a container. Verify what follows the close is also valid:
+		// another structural char, comma, or EOF.
+		return isValidAfterContainerClose(runes, j+1)
+	}
+	if next == ',' {
+		// After comma, must be followed by a valid JSON token
+		j++
+		for j < len(runes) && (runes[j] == ' ' || runes[j] == '\t' || runes[j] == '\n' || runes[j] == '\r') {
+			j++
+		}
+		if j >= len(runes) {
+			return false // trailing comma with nothing after — invalid
+		}
+		return isJSONTokenStart(runes, j)
+	}
+	// ':' is NOT valid here — we're in a value position, not a key.
+	// Any other character is also invalid.
+	return false
+}
+
+// isValidAfterContainerClose checks that after a } or ], the continuation is
+// structurally valid: more closes, comma+token, or EOF.
+func isValidAfterContainerClose(runes []rune, pos int) bool {
+	j := pos
+	for j < len(runes) && (runes[j] == ' ' || runes[j] == '\t' || runes[j] == '\n' || runes[j] == '\r') {
+		j++
+	}
+	if j >= len(runes) {
+		return true
+	}
+	next := runes[j]
+	if next == '}' || next == ']' {
+		return isValidAfterContainerClose(runes, j+1)
+	}
+	if next == ',' {
+		j++
+		for j < len(runes) && (runes[j] == ' ' || runes[j] == '\t' || runes[j] == '\n' || runes[j] == '\r') {
+			j++
+		}
+		if j >= len(runes) {
+			return false
+		}
+		return isJSONTokenStart(runes, j)
+	}
+	return false
+}
+
+// isJSONTokenStart returns true if the rune could begin a JSON value or key.
+// For keywords (true/false/null), verifies the full keyword is present.
+func isJSONTokenStart(runes []rune, pos int) bool {
+	if pos >= len(runes) {
+		return false
+	}
+	r := runes[pos]
+	switch {
+	case r == '"': // string
+		return true
+	case r == '{' || r == '[': // object or array
+		return true
+	case r == 't': // true
+		return pos+4 <= len(runes) && string(runes[pos:pos+4]) == "true"
+	case r == 'f': // false
+		return pos+5 <= len(runes) && string(runes[pos:pos+5]) == "false"
+	case r == 'n': // null
+		return pos+4 <= len(runes) && string(runes[pos:pos+4]) == "null"
+	case r >= '0' && r <= '9': // number
+		return true
+	case r == '-': // negative number
+		return true
+	}
+	return false
+}

review/parser_test.go

@@ -1,6 +1,7 @@
 package review
 
 import (
+	"encoding/json"
 	"testing"
 )
 
@@ -112,3 +113,112 @@ func TestParseResponse_MarkdownFencesNoLang(t *testing.T) {
 		t.Errorf("expected APPROVE, got %q", result.Verdict)
 	}
 }
+
+func TestParseResponse_UnescapedQuotesInStrings(t *testing.T) {
+	// Real failure from CI: Sonnet puts unescaped quotes like (e.g. "28") in findings
+	input := `{"verdict": "APPROVE", "summary": "Clean PR", "findings": [{"severity": "NIT", "file": "ci/Dockerfile", "line": 14, "finding": "The comment says OTP_VERSION is the major version (e.g. \"28\") but it actually contains unescaped quotes like (e.g. "28") which breaks JSON"}], "recommendation": "Ship it"}`
+
+	result, err := ParseResponse(input)
+	if err != nil {
+		t.Fatalf("expected repair to handle unescaped quotes, got error: %v", err)
+	}
+	if result.Verdict != "APPROVE" {
+		t.Errorf("expected APPROVE, got %q", result.Verdict)
+	}
+	if len(result.Findings) != 1 {
+		t.Fatalf("expected 1 finding, got %d", len(result.Findings))
+	}
+}
+
+func TestRepairJSON_NoOpOnValid(t *testing.T) {
+	valid := `{"key": "value", "num": 42}`
+	result := repairJSON(valid)
+	if result != valid {
+		t.Errorf("repairJSON should not modify valid JSON\n  got:  %s\n  want: %s", result, valid)
+	}
+}
+
+func TestRepairJSON_FixesUnescapedQuotes(t *testing.T) {
+	// Interior quote followed by non-structural character
+	input := `{"msg": "use "foo" here"}`
+	result := repairJSON(input)
+
+	// Should be parseable now
+	var m map[string]interface{}
+	if err := json.Unmarshal([]byte(result), &m); err != nil {
+		t.Fatalf("repaired JSON should parse, got: %v\nrepaired: %s", err, result)
+	}
+}
+
+func TestRepairJSON_InteriorQuoteBeforeComma(t *testing.T) {
+	// Bug reported by reviewer: interior quoted word immediately before a comma
+	input := `{"msg": "say "yes", and go"}`
+	result := repairJSON(input)
+
+	var m map[string]interface{}
+	if err := json.Unmarshal([]byte(result), &m); err != nil {
+		t.Fatalf("repaired JSON should parse, got: %v\nrepaired: %s", err, result)
+	}
+	// The full string content should be preserved
+	msg, ok := m["msg"].(string)
+	if !ok {
+		t.Fatal("msg field missing or not a string")
+	}
+	if msg != `say "yes", and go` {
+		t.Errorf("unexpected msg content: %q", msg)
+	}
+}
+
+func TestRepairJSON_InteriorQuoteBeforeCloseBrace(t *testing.T) {
+	// Bug reported by reviewer: JSON-shaped syntax inside string values
+	input := `{"msg": "input map {"key": "val"} caused error"}`
+	result := repairJSON(input)
+
+	var m map[string]interface{}
+	if err := json.Unmarshal([]byte(result), &m); err != nil {
+		t.Fatalf("repaired JSON should parse, got: %v\nrepaired: %s", err, result)
+	}
+}
+
+func TestRepairJSON_MultipleFields(t *testing.T) {
+	// Multiple string fields with unescaped quotes in different positions
+	input := `{"a": "hello "world"", "b": "foo"}`
+	result := repairJSON(input)
+
+	var m map[string]interface{}
+	if err := json.Unmarshal([]byte(result), &m); err != nil {
+		t.Fatalf("repaired JSON should parse, got: %v\nrepaired: %s", err, result)
+	}
+	if _, ok := m["b"]; !ok {
+		t.Error("expected 'b' field to be preserved")
+	}
+}
+
+func TestRepairJSON_PreservesEscapedQuotes(t *testing.T) {
+	// Already-escaped quotes should not be double-escaped
+	input := `{"msg": "already \"escaped\" here"}`
+	result := repairJSON(input)
+
+	if result != input {
+		t.Errorf("repairJSON should not modify already-escaped quotes\n  got:  %s\n  want: %s", result, input)
+	}
+
+	var m map[string]interface{}
+	if err := json.Unmarshal([]byte(result), &m); err != nil {
+		t.Fatalf("repaired JSON should parse, got: %v\nrepaired: %s", err, result)
+	}
+}
+
+func TestRepairJSON_ComplexNestedContent(t *testing.T) {
+	// Combines both reviewer bugs: quoted words before commas AND JSON-like content
+	input := `{"verdict": "APPROVE", "findings": [{"finding": "The map {"key": "val"} and (e.g. "28") and say "yes", then stop"}]}`
+	result := repairJSON(input)
+
+	var parsed map[string]interface{}
+	if err := json.Unmarshal([]byte(result), &parsed); err != nil {
+		t.Fatalf("repaired JSON should parse, got: %v\nrepaired: %s", err, result)
+	}
+	if parsed["verdict"] != "APPROVE" {
+		t.Errorf("expected verdict APPROVE, got %v", parsed["verdict"])
+	}
+}