diff --git a/internal/report/markdown.go b/internal/report/markdown.go
index 44b1bb1..6dab2c2 100644
--- a/internal/report/markdown.go
+++ b/internal/report/markdown.go
@@ -40,7 +40,7 @@ func Markdown(results []Result) []byte {
 	for i := 0; i < len(order); i++ {
 		target := order[i]
 		b.WriteString("## ")
-		b.WriteString(target)
+		b.WriteString(sanitizeHeading(target))
 		b.WriteString("\n\n")
 
 		mods := byTarget[target]
@@ -49,7 +49,7 @@ func Markdown(results []Result) []byte {
 
 		for j := 0; j < len(mods); j++ {
 			b.WriteString("### ")
-			b.WriteString(mods[j].Module)
+			b.WriteString(sanitizeHeading(mods[j].Module))
 			b.WriteString("\n\n")
 			b.WriteString("```json\n")
 			b.WriteString(prettyJSON(mods[j].Data))
@@ -60,6 +60,17 @@ func Markdown(results []Result) []byte {
 	return []byte(b.String())
 }
 
+// sanitizeHeading strips CR/LF from operator-supplied text (targets, module
+// ids) before it's written into a markdown heading line. both target and
+// module id can come from scan input, so an embedded newline followed by
+// "## ..." would otherwise render as a standalone, injected heading.
+func sanitizeHeading(s string) string {
+	s = strings.ReplaceAll(s, "\r\n", " ")
+	s = strings.ReplaceAll(s, "\r", " ")
+	s = strings.ReplaceAll(s, "\n", " ")
+	return s
+}
+
 // prettyJSON re-indents the raw finding for readability; if it doesn't parse as
 // json (shouldn't happen, but never trust it) the raw bytes are returned as-is.
 func prettyJSON(raw json.RawMessage) string {
diff --git a/internal/report/report_test.go b/internal/report/report_test.go
index c0a636f..4406a7a 100644
--- a/internal/report/report_test.go
+++ b/internal/report/report_test.go
@@ -14,6 +14,8 @@ package report
 
 import (
 	"encoding/json"
+	"reflect"
+	"sort"
 	"strings"
 	"testing"
 )
@@ -75,6 +77,46 @@ func TestSARIF_ValidAndContainsFindings(t *testing.T) {
 	}
 }
 
+func TestSARIF_RulesAreSorted(t *testing.T) {
+	// rules are collected from a map internally, so without an explicit sort
+	// the emitted order is whatever the map iteration happened to produce.
+	// several distinct module ids make an accidentally-sorted iteration
+	// vanishingly unlikely, so this pins the fix rather than relying on luck.
+	results := []Result{
+		{Target: "https://t", Module: "zeta", Data: json.RawMessage(`{}`)},
+		{Target: "https://t", Module: "mike", Data: json.RawMessage(`{}`)},
+		{Target: "https://t", Module: "alpha", Data: json.RawMessage(`{}`)},
+		{Target: "https://t", Module: "yankee", Data: json.RawMessage(`{}`)},
+		{Target: "https://t", Module: "bravo", Data: json.RawMessage(`{}`)},
+		{Target: "https://t", Module: "delta", Data: json.RawMessage(`{}`)},
+		{Target: "https://t", Module: "kilo", Data: json.RawMessage(`{}`)},
+	}
+
+	var firstIDs []string
+	for run := 0; run < 5; run++ {
+		out, err := SARIF(results)
+		if err != nil {
+			t.Fatalf("SARIF: %v", err)
+		}
+		var doc sarifLog
+		if err := json.Unmarshal(out, &doc); err != nil {
+			t.Fatalf("invalid json: %v", err)
+		}
+		ids := make([]string, len(doc.Runs[0].Tool.Driver.Rules))
+		for i, r := range doc.Runs[0].Tool.Driver.Rules {
+			ids[i] = r.ID
+		}
+		if !sort.StringsAreSorted(ids) {
+			t.Fatalf("run %d: driver.rules not sorted: %v", run, ids)
+		}
+		if run == 0 {
+			firstIDs = ids
+		} else if !reflect.DeepEqual(firstIDs, ids) {
+			t.Fatalf("driver.rules order changed across runs: %v vs %v", firstIDs, ids)
+		}
+	}
+}
+
 func TestSARIF_DedupesRulesAcrossTargets(t *testing.T) {
 	// the same module on two targets must yield one rule but two results.
 	results := []Result{
@@ -193,6 +235,24 @@ func TestMarkdown_GroupsByTarget(t *testing.T) {
 	}
 }
 
+func TestMarkdown_StripsNewlinesFromTargetHeader(t *testing.T) {
+	// same newline-injection guard as sanitizeHeading (markdown.go); this exercises the malicious-input case.
+	results := []Result{
+		{Target: "https://evil.example.com\n## injected", Module: "probe", Data: json.RawMessage(`{"status_code":200}`)},
+	}
+	out := string(Markdown(results))
+
+	lines := strings.Split(out, "\n")
+	for _, line := range lines {
+		if line == "## injected" {
+			t.Errorf("target newline produced a standalone injected heading:\n%s", out)
+		}
+	}
+	if strings.Contains(out, "\r") {
+		t.Errorf("markdown output must not contain carriage returns:\n%q", out)
+	}
+}
+
 // sarifHasResult reports whether any result carries the given rule id and target
 // uri, the pairing that proves a finding survived serialization.
 func sarifHasResult(results []sarifResult, ruleID, target string) bool {
diff --git a/internal/report/sarif.go b/internal/report/sarif.go
index 2452e1a..76f657c 100644
--- a/internal/report/sarif.go
+++ b/internal/report/sarif.go
@@ -15,6 +15,7 @@ package report
 import (
 	"encoding/json"
 	"fmt"
+	"sort"
 )
 
 // sarif format/version constants pinned to the 2.1.0 schema so the output is
@@ -121,9 +122,18 @@ func SARIF(results []Result) ([]byte, error) {
 	}
 
 	// rules must list each id exactly once; build it from the set so duplicate
-	// modules across targets don't duplicate the rule.
-	rules := make([]sarifRule, 0, len(ruleSet))
+	// modules across targets don't duplicate the rule. the set itself ranges
+	// in random map order, so sort the ids first: otherwise driver.rules
+	// would come out in a different order on every run, making the sarif
+	// output byte-unstable across identical scans.
+	ruleIDs := make([]string, 0, len(ruleSet))
 	for id := range ruleSet {
+		ruleIDs = append(ruleIDs, id)
+	}
+	sort.Strings(ruleIDs)
+
+	rules := make([]sarifRule, 0, len(ruleIDs))
+	for _, id := range ruleIDs {
 		rules = append(rules, sarifRule{ID: id})
 	}