diff --git a/src/arbiter/integrations/formatters.py b/src/arbiter/integrations/formatters.py
new file mode 100644
index 0000000..7a797b3
--- /dev/null
+++ b/src/arbiter/integrations/formatters.py
@@ -0,0 +1,249 @@
+"""Formatters for review comments on code hosting platforms."""
+
+from collections import defaultdict
+from typing import TYPE_CHECKING, ClassVar
+
+from arbiter.models import Finding, Severity, Verdict
+
+if TYPE_CHECKING:
+    from arbiter.deliberation import Conflict, DeliberationResult
+
+# HTML comment marker for identifying Arbiter review comments
+ARBITER_MARKER = "<!-- arbiter-review -->"
+
+
+def has_arbiter_marker(text: str) -> bool:
+    return ARBITER_MARKER in text
+
+
+class ReviewCommentFormatter:
+    """Formats DeliberationResult as Markdown for PR comments."""
+
+    # GitHub comment limit
+    MAX_COMMENT_LENGTH = 65535
+
+    # Verdict icons
+    VERDICT_ICONS: ClassVar[dict[Verdict, str]] = {
+        Verdict.APPROVE: "\u2705",  # green checkmark
+        Verdict.REQUEST_CHANGES: "\u274c",  # red X
+        Verdict.COMMENT: "\U0001f4ac",  # speech bubble
+    }
+
+    # Severity icons
+    SEVERITY_ICONS: ClassVar[dict[Severity, str]] = {
+        Severity.CRITICAL: "\U0001f6a8",  # rotating light
+        Severity.HIGH: "\U0001f534",  # red circle
+        Severity.MEDIUM: "\U0001f7e0",  # orange circle
+        Severity.LOW: "\U0001f7e1",  # yellow circle
+        Severity.INFO: "\U0001f535",  # blue circle
+    }
+
+    def __init__(self, include_cost: bool = False, max_findings: int = 50) -> None:
+        self.include_cost = include_cost
+        self.max_findings = max_findings
+
+    def format(self, result: "DeliberationResult") -> str:
+        """Format a DeliberationResult as Markdown.
+
+        Args:
+            result: Deliberation result to format.
+
+        Returns:
+            Markdown-formatted comment body.
+        """
+        sections: list[str] = []
+
+        # Header with verdict
+        sections.append(self._format_header(result))
+
+        # Summary statistics
+        sections.append(self._format_summary(result))
+
+        # Findings by severity
+        if result.findings:
+            sections.append(self._format_findings(result.findings))
+
+        # Conflicts section
+        if result.conflicts:
+            sections.append(self._format_conflicts(result.conflicts))
+
+        # Cost footer (optional)
+        if self.include_cost:
+            sections.append(self._format_cost(result))
+
+        # Footer
+        sections.append(self._format_footer())
+
+        # Add marker at the end (invisible in rendered markdown)
+        sections.append(ARBITER_MARKER)
+
+        comment = "\n\n".join(sections)
+
+        # Truncate if needed
+        if len(comment) > self.MAX_COMMENT_LENGTH:
+            comment = self._truncate(comment)
+
+        return comment
+
+    def _format_header(self, result: "DeliberationResult") -> str:
+        """Format the verdict header."""
+        icon = self.VERDICT_ICONS.get(result.verdict, "")
+        verdict_text = result.verdict.value.replace("_", " ").title()
+
+        header = f"## {icon} Arbiter Review: {verdict_text}\n\n"
+        header += f"**Confidence:** {result.verdict_confidence:.0%}\n\n"
+
+        if result.verdict_reasoning:
+            header += f"> {result.verdict_reasoning}"
+
+        return header
+
+    def _format_summary(self, result: "DeliberationResult") -> str:
+        """Format summary statistics."""
+        lines: list[str] = ["### Summary"]
+
+        # Count by severity
+        counts: dict[Severity, int] = defaultdict(int)
+        for finding in result.findings:
+            counts[finding.severity] += 1
+
+        # Format counts table
+        if result.total_findings > 0:
+            lines.append("")
+            lines.append("| Severity | Count |")
+            lines.append("|----------|-------|")
+
+            for severity in [
+                Severity.CRITICAL,
+                Severity.HIGH,
+                Severity.MEDIUM,
+                Severity.LOW,
+                Severity.INFO,
+            ]:
+                if counts[severity] > 0:
+                    icon = self.SEVERITY_ICONS.get(severity, "")
+                    lines.append(f"| {icon} {severity.value.title()} | {counts[severity]} |")
+
+            lines.append("")
+            lines.append(f"**Total findings:** {result.total_findings}")
+        else:
+            lines.append("\nNo issues found.")
+
+        if result.conflicts:
+            lines.append(f"\n**Conflicts detected:** {len(result.conflicts)}")
+
+        return "\n".join(lines)
+
+    def _format_findings(self, findings: list[Finding]) -> str:
+        """Format findings grouped by severity."""
+        lines: list[str] = ["### Findings"]
+
+        # Group by severity
+        by_severity: dict[Severity, list[Finding]] = defaultdict(list)
+        for finding in findings:
+            by_severity[finding.severity].append(finding)
+
+        # Display in severity order
+        displayed = 0
+        for severity in [
+            Severity.CRITICAL,
+            Severity.HIGH,
+            Severity.MEDIUM,
+            Severity.LOW,
+            Severity.INFO,
+        ]:
+            severity_findings = by_severity[severity]
+            if not severity_findings:
+                continue
+
+            icon = self.SEVERITY_ICONS.get(severity, "")
+            lines.append(f"\n#### {icon} {severity.value.title()}")
+
+            for finding in severity_findings:
+                if displayed >= self.max_findings:
+                    remaining = len(findings) - displayed
+                    lines.append(f"\n*...and {remaining} more findings*")
+                    return "\n".join(lines)
+
+                lines.append(self._format_finding(finding))
+                displayed += 1
+
+        return "\n".join(lines)
+
+    def _format_finding(self, finding: Finding) -> str:
+        """Format a single finding."""
+        lines: list[str] = []
+
+        # Location and title
+        location = f"`{finding.file}:{finding.line_start}"
+        if finding.line_end != finding.line_start:
+            location += f"-{finding.line_end}"
+        location += "`"
+
+        lines.append(f"\n**{finding.title}** ({location})")
+        lines.append(
+            f"\n_{finding.agent.value.title()} agent_ | Confidence: {finding.confidence:.0%}"
+        )
+
+        # Description
+        lines.append(f"\n{finding.description}")
+
+        # Suggestion (if present)
+        if finding.suggestion:
+            lines.append(f"\n> **Suggestion:** {finding.suggestion}")
+
+        # References (if present)
+        if finding.references:
+            refs = " | ".join(f"[{i + 1}]({ref})" for i, ref in enumerate(finding.references[:3]))
+            lines.append(f"\n**References:** {refs}")
+
+        return "\n".join(lines)
+
+    def _format_conflicts(self, conflicts: list["Conflict"]) -> str:
+        """Format the conflicts section."""
+        lines: list[str] = ["### Agent Conflicts"]
+        lines.append("\nThe following conflicts were detected between agents:")
+
+        for conflict in conflicts[:10]:  # Limit to 10 conflicts
+            lines.append(
+                f"\n- **{conflict.nature.value.replace('_', ' ').title()}**: {conflict.description}"
+            )
+            if conflict.resolution:
+                lines.append(f"  - *Resolution:* {conflict.resolution}")
+
+        if len(conflicts) > 10:
+            lines.append(f"\n*...and {len(conflicts) - 10} more conflicts*")
+
+        return "\n".join(lines)
+
+    def _format_cost(self, result: "DeliberationResult") -> str:
+        lines: list[str] = ["---"]
+        lines.append("<details>")
+        lines.append("<summary>Review cost</summary>")
+        lines.append("")
+        lines.append(f"- Tokens used: {result.tokens_used:,}")
+        lines.append(f"- Cost: ${result.cost_usd:.4f}")
+        lines.append("</details>")
+        return "\n".join(lines)
+
+    def _format_footer(self) -> str:
+        return (
+            "---\n"
+            "*Generated by [Arbiter](https://github.com/kschappell/arbiter) "
+            "- Multi-agent code review*"
+        )
+
+    def _truncate(self, comment: str) -> str:
+        """Truncate comment to fit within GitHub's limit."""
+        # Leave room for truncation message
+        truncation_msg = "\n\n*[Comment truncated due to length limit]*"
+        max_len = self.MAX_COMMENT_LENGTH - len(truncation_msg)
+
+        # Try to truncate at a paragraph boundary
+        truncated = comment[:max_len]
+        last_double_newline = truncated.rfind("\n\n")
+
+        if last_double_newline > max_len // 2:
+            truncated = truncated[:last_double_newline]
+
+        return truncated + truncation_msg