diff --git a/cmd/roborev/analyze.go b/cmd/roborev/analyze.go
index 908a0d1c4..8eba8c731 100644
--- a/cmd/roborev/analyze.go
+++ b/cmd/roborev/analyze.go
@@ -170,7 +170,7 @@ To fix an existing analysis job, use: roborev fix <job_id>
 
 	cmd.Flags().StringVar(&agentName, "agent", "", "agent to use for analysis (default: from config)")
 	cmd.Flags().StringVar(&model, "model", "", "model for analysis agent")
-	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: fast, standard, or thorough")
+	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: fast, standard, medium, thorough, or maximum")
 	cmd.Flags().BoolVar(&wait, "wait", false, "wait for job to complete and show result")
 	cmd.Flags().BoolVarP(&quiet, "quiet", "q", false, "suppress output (just enqueue)")
 	cmd.Flags().BoolVar(&listTypes, "list", false, "list available analysis types")
diff --git a/cmd/roborev/ci.go b/cmd/roborev/ci.go
index 40a9210b1..6cf81d6f1 100644
--- a/cmd/roborev/ci.go
+++ b/cmd/roborev/ci.go
@@ -83,7 +83,7 @@ Flags override config values. When run inside GitHub ` +
 	cmd.Flags().StringVar(&reviewTypes, "review-types", "",
 		"comma-separated review types (overrides config)")
 	cmd.Flags().StringVar(&reasoning, "reasoning", "",
-		"reasoning level: thorough, standard, fast")
+		"reasoning level: fast, standard, medium, thorough, or maximum")
 	cmd.Flags().StringVar(&minSeverity, "min-severity", "",
 		"minimum severity filter: critical, high, medium, low")
 	cmd.Flags().StringVar(&synthesisAgent, "synthesis-agent", "",
diff --git a/cmd/roborev/compact.go b/cmd/roborev/compact.go
index e89cddc36..3f8f9cf71 100644
--- a/cmd/roborev/compact.go
+++ b/cmd/roborev/compact.go
@@ -89,7 +89,7 @@ Examples:
 
 	cmd.Flags().StringVar(&agentName, "agent", "", "agent to use for verification (defaults to fix agent from config)")
 	cmd.Flags().StringVar(&model, "model", "", "model to use")
-	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level (fast/standard/thorough)")
+	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level (fast/standard/medium/thorough/maximum)")
 	cmd.Flags().BoolVarP(&quiet, "quiet", "q", false, "suppress progress output")
 	cmd.Flags().BoolVar(&allBranches, "all-branches", false, "include all branches")
 	cmd.Flags().StringVar(&branch, "branch", "", "filter by branch (default: current branch)")
diff --git a/cmd/roborev/fix.go b/cmd/roborev/fix.go
index 5aa83d650..5bbabb988 100644
--- a/cmd/roborev/fix.go
+++ b/cmd/roborev/fix.go
@@ -208,7 +208,7 @@ Examples:
 
 	cmd.Flags().StringVar(&agentName, "agent", "", "agent to use for fixes (default: from config)")
 	cmd.Flags().StringVar(&model, "model", "", "model for agent")
-	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: fast, standard, or thorough")
+	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: fast, standard, medium, thorough, or maximum")
 	cmd.Flags().StringVar(&minSeverity, "min-severity", "", "minimum finding severity to address: critical, high, medium, or low")
 	cmd.Flags().BoolVarP(&quiet, "quiet", "q", false, "suppress progress output")
 	cmd.Flags().BoolVar(&open, "open", false, "fix all open completed jobs for the current repo")
diff --git a/cmd/roborev/local_review_integration_test.go b/cmd/roborev/local_review_integration_test.go
index f20c8668f..44e0f57c1 100644
--- a/cmd/roborev/local_review_integration_test.go
+++ b/cmd/roborev/local_review_integration_test.go
@@ -17,7 +17,9 @@ func TestLocalReviewReasoningLevels(t *testing.T) {
 	}{
 		{name: "Fast", reasoning: string(agent.ReasoningFast), expected: "reasoning: " + string(agent.ReasoningFast)},
 		{name: "Standard", reasoning: string(agent.ReasoningStandard), expected: "reasoning: " + string(agent.ReasoningStandard)},
+		{name: "Medium", reasoning: string(agent.ReasoningMedium), expected: "reasoning: " + string(agent.ReasoningMedium)},
 		{name: "Thorough", reasoning: string(agent.ReasoningThorough), expected: "reasoning: " + string(agent.ReasoningThorough)},
+		{name: "Maximum", reasoning: string(agent.ReasoningMaximum), expected: "reasoning: " + string(agent.ReasoningMaximum)},
 		{name: "Default", reasoning: "", expected: "reasoning: " + string(agent.ReasoningThorough)}, // default (agent defaults)
 	}
 
diff --git a/cmd/roborev/refine.go b/cmd/roborev/refine.go
index bba3bda73..a6057cd87 100644
--- a/cmd/roborev/refine.go
+++ b/cmd/roborev/refine.go
@@ -132,7 +132,7 @@ Use --all-branches to discover and refine all branches with failed reviews.`,
 
 	cmd.Flags().StringVar(&opts.agentName, "agent", "", "agent to use for addressing findings (default: from config)")
 	cmd.Flags().StringVar(&opts.model, "model", "", "model for agent (format varies: opencode uses provider/model, others use model name)")
-	cmd.Flags().StringVar(&opts.reasoning, "reasoning", "", "reasoning level: fast, standard (default), or thorough")
+	cmd.Flags().StringVar(&opts.reasoning, "reasoning", "", "reasoning level: fast, standard (default), medium, thorough, or maximum")
 	cmd.Flags().StringVar(&opts.minSeverity, "min-severity", "", "minimum finding severity to address: critical, high, medium, or low")
 	cmd.Flags().BoolVar(&fast, "fast", false, "shorthand for --reasoning fast")
 	cmd.Flags().IntVar(&opts.maxIterations, "max-iterations", 10, "maximum refinement iterations")
diff --git a/cmd/roborev/review.go b/cmd/roborev/review.go
index 825cdd95f..fdef2464f 100644
--- a/cmd/roborev/review.go
+++ b/cmd/roborev/review.go
@@ -344,7 +344,7 @@ Examples:
 	cmd.Flags().StringVar(&sha, "sha", "HEAD", "commit SHA to review (used when no positional args)")
 	cmd.Flags().StringVar(&agent, "agent", "", "agent to use (codex, claude-code, gemini, copilot, opencode, cursor, kiro, kilo, pi)")
 	cmd.Flags().StringVar(&model, "model", "", "model for agent (format varies: opencode uses provider/model, others use model name)")
-	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: thorough (default), standard, or fast")
+	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: fast, standard, medium, thorough (default), or maximum")
 	cmd.Flags().BoolVar(&fast, "fast", false, "shorthand for --reasoning fast")
 	cmd.Flags().BoolVarP(&quiet, "quiet", "q", false, "suppress output (for use in hooks)")
 	cmd.Flags().BoolVar(&dirty, "dirty", false, "review uncommitted changes instead of a commit")
diff --git a/cmd/roborev/run.go b/cmd/roborev/run.go
index 1e17f0140..c2fa64d42 100644
--- a/cmd/roborev/run.go
+++ b/cmd/roborev/run.go
@@ -68,7 +68,7 @@ Examples:
 
 	cmd.Flags().StringVar(&agentName, "agent", "", "agent to use (default: from config)")
 	cmd.Flags().StringVar(&model, "model", "", "model for agent (format varies: opencode uses provider/model, others use model name)")
-	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: fast, standard, or thorough (default)")
+	cmd.Flags().StringVar(&reasoning, "reasoning", "", "reasoning level: fast, standard, medium, thorough (default), or maximum")
 	cmd.Flags().BoolVar(&wait, "wait", false, "wait for job to complete and show result")
 	cmd.Flags().BoolVarP(&quiet, "quiet", "q", false, "suppress output (just enqueue)")
 	cmd.Flags().BoolVar(&noContext, "no-context", false, "don't include repository context in prompt")
diff --git a/internal/agent/agent.go b/internal/agent/agent.go
index 3443f2eec..f0baa36c3 100644
--- a/internal/agent/agent.go
+++ b/internal/agent/agent.go
@@ -15,9 +15,13 @@ import (
 type ReasoningLevel string
 
 const (
-	// ReasoningThorough uses maximum reasoning for deep analysis (slower)
+	// ReasoningMaximum uses the highest available reasoning (e.g., codex xhigh, claude max)
+	ReasoningMaximum ReasoningLevel = "maximum"
+	// ReasoningThorough uses deep reasoning for thorough analysis (slower)
 	ReasoningThorough ReasoningLevel = "thorough"
-	// ReasoningStandard uses balanced reasoning (default)
+	// ReasoningMedium uses moderate reasoning (e.g., claude --effort medium)
+	ReasoningMedium ReasoningLevel = "medium"
+	// ReasoningStandard uses the model's default reasoning (no effort override)
 	ReasoningStandard ReasoningLevel = "standard"
 	// ReasoningFast uses minimal reasoning for quick responses
 	ReasoningFast ReasoningLevel = "fast"
@@ -25,17 +29,21 @@ const (
 
 // ReasoningLevels returns the canonical reasoning level names.
 func ReasoningLevels() []string {
-	return []string{string(ReasoningFast), string(ReasoningStandard), string(ReasoningThorough)}
+	return []string{string(ReasoningFast), string(ReasoningStandard), string(ReasoningMedium), string(ReasoningThorough), string(ReasoningMaximum)}
 }
 
 // ParseReasoningLevel converts a string to ReasoningLevel, defaulting to standard
 func ParseReasoningLevel(s string) ReasoningLevel {
 	switch s {
+	case "maximum", "max", "xhigh":
+		return ReasoningMaximum
 	case "thorough", "high":
 		return ReasoningThorough
+	case "medium":
+		return ReasoningMedium
 	case "fast", "low":
 		return ReasoningFast
-	case "standard", "medium", "":
+	case "standard", "":
 		return ReasoningStandard
 	default:
 		return ReasoningStandard
diff --git a/internal/agent/agent_test.go b/internal/agent/agent_test.go
index 71734081c..379165106 100644
--- a/internal/agent/agent_test.go
+++ b/internal/agent/agent_test.go
@@ -135,12 +135,15 @@ func TestParseReasoningLevel(t *testing.T) {
 		input string
 		want  ReasoningLevel
 	}{
+		{"maximum", ReasoningMaximum},
+		{"max", ReasoningMaximum},
+		{"xhigh", ReasoningMaximum},
 		{"thorough", ReasoningThorough},
 		{"high", ReasoningThorough},
 		{"fast", ReasoningFast},
 		{"low", ReasoningFast},
+		{"medium", ReasoningMedium},
 		{"standard", ReasoningStandard},
-		{"medium", ReasoningStandard},
 		{"", ReasoningStandard},
 		{"unknown", ReasoningStandard},
 	}
@@ -155,6 +158,7 @@ func TestCodexReasoningEffortMapping(t *testing.T) {
 		level ReasoningLevel
 		want  string
 	}{
+		{ReasoningMaximum, "xhigh"},
 		{ReasoningThorough, "high"},
 		{ReasoningFast, "low"},
 		{ReasoningStandard, ""},
@@ -168,6 +172,26 @@ func TestCodexReasoningEffortMapping(t *testing.T) {
 	}
 }
 
+func TestClaudeEffortMapping(t *testing.T) {
+	tests := []struct {
+		level ReasoningLevel
+		want  string
+	}{
+		{ReasoningMaximum, "max"},
+		{ReasoningThorough, "high"},
+		{ReasoningMedium, "medium"},
+		{ReasoningFast, "low"},
+		{ReasoningStandard, ""},
+	}
+
+	for _, tt := range tests {
+		a := NewClaudeAgent("").WithReasoning(tt.level)
+		claude, ok := a.(*ClaudeAgent)
+		require.True(t, ok, "expected ClaudeAgent, got %T", a)
+		assert.Equal(t, tt.want, claude.claudeEffort(), "claudeEffort(%q)", tt.level)
+	}
+}
+
 type agentTestDef struct {
 	name                  string
 	factory               func(string) Agent
@@ -247,6 +271,21 @@ func TestWithModelEmptyPreservesDefault(t *testing.T) {
 	}
 }
 
+func TestClaudeBuildArgsEffort(t *testing.T) {
+	a := NewClaudeAgent("").WithModel("opus").WithReasoning(ReasoningMaximum)
+	cmdLine := a.CommandLine()
+
+	assert.Contains(t, cmdLine, "--effort max")
+	assert.Contains(t, cmdLine, "--model opus")
+}
+
+func TestClaudeBuildArgsNoEffortForStandard(t *testing.T) {
+	a := NewClaudeAgent("").WithReasoning(ReasoningStandard)
+	cmdLine := a.CommandLine()
+
+	assert.NotContains(t, cmdLine, "--effort")
+}
+
 func TestAgentBuildArgsWithModel(t *testing.T) {
 	for _, tt := range agentFixtures {
 		t.Run(tt.name+" with explicit model", func(t *testing.T) {
diff --git a/internal/agent/claude.go b/internal/agent/claude.go
index f5474d194..4c374a12d 100644
--- a/internal/agent/claude.go
+++ b/internal/agent/claude.go
@@ -18,14 +18,16 @@ import (
 type ClaudeAgent struct {
 	Command   string         // The claude command to run (default: "claude")
 	Model     string         // Model to use (e.g., "opus", "sonnet", or full name)
-	Reasoning ReasoningLevel // Reasoning level (for future extended thinking support)
+	Reasoning ReasoningLevel // Reasoning level mapped to --effort flag
 	Agentic   bool           // Whether agentic mode is enabled (allow file edits)
 	SessionID string         // Existing session ID to resume
 }
 
 const claudeDangerousFlag = "--dangerously-skip-permissions"
+const claudeEffortFlag = "--effort"
 
 var claudeDangerousSupport sync.Map
+var claudeEffortSupport sync.Map
 
 // NewClaudeAgent creates a new Claude Code agent
 func NewClaudeAgent(command string) *ClaudeAgent {
@@ -35,7 +37,7 @@ func NewClaudeAgent(command string) *ClaudeAgent {
 	return &ClaudeAgent{Command: command, Reasoning: ReasoningStandard}
 }
 
-// WithReasoning returns a copy of the agent with the model preserved (reasoning not yet supported).
+// WithReasoning returns a copy of the agent with the specified reasoning level.
 func (a *ClaudeAgent) WithReasoning(level ReasoningLevel) Agent {
 	return &ClaudeAgent{
 		Command:   a.Command,
@@ -82,6 +84,22 @@ func (a *ClaudeAgent) WithSessionID(sessionID string) Agent {
 	}
 }
 
+// claudeEffort maps ReasoningLevel to Claude Code's --effort flag values
+func (a *ClaudeAgent) claudeEffort() string {
+	switch a.Reasoning {
+	case ReasoningMaximum:
+		return "max"
+	case ReasoningThorough:
+		return "high"
+	case ReasoningMedium:
+		return "medium"
+	case ReasoningFast:
+		return "low"
+	default:
+		return "" // use claude default (standard = no override)
+	}
+}
+
 func (a *ClaudeAgent) Name() string {
 	return "claude-code"
 }
@@ -92,11 +110,11 @@ func (a *ClaudeAgent) CommandName() string {
 
 func (a *ClaudeAgent) CommandLine() string {
 	agenticMode := a.Agentic || AllowUnsafeAgents()
-	args := a.buildArgs(agenticMode)
+	args := a.buildArgs(agenticMode, true)
 	return a.Command + " " + strings.Join(args, " ")
 }
 
-func (a *ClaudeAgent) buildArgs(agenticMode bool) []string {
+func (a *ClaudeAgent) buildArgs(agenticMode, includeEffort bool) []string {
 	sessionID := sanitizedResumeSessionID(a.SessionID)
 	// Always use stdin piping + stream-json for non-interactive execution
 	// (following claude-code-action pattern from Anthropic)
@@ -109,6 +127,12 @@ func (a *ClaudeAgent) buildArgs(agenticMode bool) []string {
 		args = append(args, "--resume", sessionID)
 	}
 
+	if includeEffort {
+		if effort := a.claudeEffort(); effort != "" {
+			args = append(args, claudeEffortFlag, effort)
+		}
+	}
+
 	if agenticMode {
 		// Agentic mode: Claude can use tools and make file changes
 		args = append(args, claudeDangerousFlag)
@@ -134,6 +158,17 @@ func claudeSupportsDangerousFlag(ctx context.Context, command string) (bool, err
 	return supported, nil
 }
 
+func claudeSupportsEffortFlag(ctx context.Context, command string) bool {
+	if cached, ok := claudeEffortSupport.Load(command); ok {
+		return cached.(bool)
+	}
+	cmd := exec.CommandContext(ctx, command, "--help")
+	output, _ := cmd.CombinedOutput()
+	supported := strings.Contains(string(output), claudeEffortFlag)
+	claudeEffortSupport.Store(command, supported)
+	return supported
+}
+
 func (a *ClaudeAgent) Review(ctx context.Context, repoPath, commitSHA, prompt string, output io.Writer) (string, error) {
 	// Use agentic mode if either per-job setting or global setting enables it
 	agenticMode := a.Agentic || AllowUnsafeAgents()
@@ -148,8 +183,11 @@ func (a *ClaudeAgent) Review(ctx context.Context, repoPath, commitSHA, prompt st
 		}
 	}
 
+	// Only pass --effort if the installed Claude Code supports it
+	includeEffort := a.claudeEffort() != "" && claudeSupportsEffortFlag(ctx, a.Command)
+
 	// Build args - always uses stdin piping + stream-json for non-interactive execution
-	args := a.buildArgs(agenticMode)
+	args := a.buildArgs(agenticMode, includeEffort)
 
 	cmd := exec.CommandContext(ctx, a.Command, args...)
 	cmd.Dir = repoPath
diff --git a/internal/agent/claude_test.go b/internal/agent/claude_test.go
index 4e02931f2..ba71be139 100644
--- a/internal/agent/claude_test.go
+++ b/internal/agent/claude_test.go
@@ -48,7 +48,7 @@ func TestClaudeBuildArgs(t *testing.T) {
 
 	t.Run("ReviewMode", func(t *testing.T) {
 		// Non-agentic mode (review only): read-only tools, no dangerous flag
-		args := a.buildArgs(false)
+		args := a.buildArgs(false, false)
 		for _, req := range []string{"--output-format", "stream-json", "--verbose", "-p", "--allowedTools"} {
 			assertContainsArg(t, args, req)
 		}
@@ -61,7 +61,7 @@ func TestClaudeBuildArgs(t *testing.T) {
 
 	t.Run("AgenticMode", func(t *testing.T) {
 		// Agentic mode: write tools + dangerous flag
-		args := a.buildArgs(true)
+		args := a.buildArgs(true, false)
 		assertContainsArg(t, args, claudeDangerousFlag)
 		assertContainsArg(t, args, "--allowedTools")
 
@@ -71,13 +71,13 @@ func TestClaudeBuildArgs(t *testing.T) {
 	})
 
 	t.Run("ResumeSession", func(t *testing.T) {
-		args := a.WithSessionID("session-123").(*ClaudeAgent).buildArgs(false)
+		args := a.WithSessionID("session-123").(*ClaudeAgent).buildArgs(false, false)
 		assertContainsArg(t, args, "--resume")
 		assertContainsArg(t, args, "session-123")
 	})
 
 	t.Run("RejectInvalidResumeSession", func(t *testing.T) {
-		args := a.WithSessionID("-bad-session").(*ClaudeAgent).buildArgs(false)
+		args := a.WithSessionID("-bad-session").(*ClaudeAgent).buildArgs(false, false)
 		assertNotContainsArg(t, args, "--resume")
 		assertNotContainsArg(t, args, "-bad-session")
 	})
diff --git a/internal/agent/codex.go b/internal/agent/codex.go
index 9582d765d..639d69199 100644
--- a/internal/agent/codex.go
+++ b/internal/agent/codex.go
@@ -86,6 +86,8 @@ func (a *CodexAgent) WithSessionID(sessionID string) Agent {
 // codexReasoningEffort maps ReasoningLevel to codex-specific effort values
 func (a *CodexAgent) codexReasoningEffort() string {
 	switch a.Reasoning {
+	case ReasoningMaximum:
+		return "xhigh"
 	case ReasoningThorough:
 		return "high"
 	case ReasoningFast:
diff --git a/internal/agent/droid.go b/internal/agent/droid.go
index 285d3a1d0..aeb75ff0e 100644
--- a/internal/agent/droid.go
+++ b/internal/agent/droid.go
@@ -46,7 +46,7 @@ func (a *DroidAgent) WithModel(model string) Agent {
 // droidReasoningEffort maps ReasoningLevel to droid-specific effort values
 func (a *DroidAgent) droidReasoningEffort() string {
 	switch a.Reasoning {
-	case ReasoningThorough:
+	case ReasoningMaximum, ReasoningThorough:
 		return "high"
 	case ReasoningFast:
 		return "low"
diff --git a/internal/agent/kilo.go b/internal/agent/kilo.go
index bf4a38cd6..c2b074fa5 100644
--- a/internal/agent/kilo.go
+++ b/internal/agent/kilo.go
@@ -84,7 +84,7 @@ func (a *KiloAgent) CommandName() string {
 // kiloVariant maps ReasoningLevel to kilo's --variant flag values
 func (a *KiloAgent) kiloVariant() string {
 	switch a.Reasoning {
-	case ReasoningThorough:
+	case ReasoningMaximum, ReasoningThorough:
 		return "high"
 	case ReasoningFast:
 		return "minimal"
diff --git a/internal/agent/pi.go b/internal/agent/pi.go
index 2a0325b52..88f34c780 100644
--- a/internal/agent/pi.go
+++ b/internal/agent/pi.go
@@ -134,7 +134,7 @@ func (a *PiAgent) buildArgs(repoPath string) []string {
 
 func (a *PiAgent) thinkingLevel() string {
 	switch a.Reasoning {
-	case ReasoningThorough:
+	case ReasoningMaximum, ReasoningThorough:
 		return "high"
 	case ReasoningFast:
 		return "low"
diff --git a/internal/config/config.go b/internal/config/config.go
index 83503f4af..9c6d03164 100644
--- a/internal/config/config.go
+++ b/internal/config/config.go
@@ -65,43 +65,63 @@ type Config struct {
 	ReviewAgent           string `toml:"review_agent"`
 	ReviewAgentFast       string `toml:"review_agent_fast"`
 	ReviewAgentStandard   string `toml:"review_agent_standard"`
+	ReviewAgentMedium     string `toml:"review_agent_medium"`
 	ReviewAgentThorough   string `toml:"review_agent_thorough"`
+	ReviewAgentMaximum    string `toml:"review_agent_maximum"`
 	RefineAgent           string `toml:"refine_agent"`
 	RefineAgentFast       string `toml:"refine_agent_fast"`
 	RefineAgentStandard   string `toml:"refine_agent_standard"`
+	RefineAgentMedium     string `toml:"refine_agent_medium"`
 	RefineAgentThorough   string `toml:"refine_agent_thorough"`
+	RefineAgentMaximum    string `toml:"refine_agent_maximum"`
 	ReviewModel           string `toml:"review_model"`
 	ReviewModelFast       string `toml:"review_model_fast"`
 	ReviewModelStandard   string `toml:"review_model_standard"`
+	ReviewModelMedium     string `toml:"review_model_medium"`
 	ReviewModelThorough   string `toml:"review_model_thorough"`
+	ReviewModelMaximum    string `toml:"review_model_maximum"`
 	RefineModel           string `toml:"refine_model"`
 	RefineModelFast       string `toml:"refine_model_fast"`
 	RefineModelStandard   string `toml:"refine_model_standard"`
+	RefineModelMedium     string `toml:"refine_model_medium"`
 	RefineModelThorough   string `toml:"refine_model_thorough"`
+	RefineModelMaximum    string `toml:"refine_model_maximum"`
 	FixAgent              string `toml:"fix_agent"`
 	FixAgentFast          string `toml:"fix_agent_fast"`
 	FixAgentStandard      string `toml:"fix_agent_standard"`
+	FixAgentMedium        string `toml:"fix_agent_medium"`
 	FixAgentThorough      string `toml:"fix_agent_thorough"`
+	FixAgentMaximum       string `toml:"fix_agent_maximum"`
 	FixModel              string `toml:"fix_model"`
 	FixModelFast          string `toml:"fix_model_fast"`
 	FixModelStandard      string `toml:"fix_model_standard"`
+	FixModelMedium        string `toml:"fix_model_medium"`
 	FixModelThorough      string `toml:"fix_model_thorough"`
+	FixModelMaximum       string `toml:"fix_model_maximum"`
 	SecurityAgent         string `toml:"security_agent"`
 	SecurityAgentFast     string `toml:"security_agent_fast"`
 	SecurityAgentStandard string `toml:"security_agent_standard"`
+	SecurityAgentMedium   string `toml:"security_agent_medium"`
 	SecurityAgentThorough string `toml:"security_agent_thorough"`
+	SecurityAgentMaximum  string `toml:"security_agent_maximum"`
 	SecurityModel         string `toml:"security_model"`
 	SecurityModelFast     string `toml:"security_model_fast"`
 	SecurityModelStandard string `toml:"security_model_standard"`
+	SecurityModelMedium   string `toml:"security_model_medium"`
 	SecurityModelThorough string `toml:"security_model_thorough"`
+	SecurityModelMaximum  string `toml:"security_model_maximum"`
 	DesignAgent           string `toml:"design_agent"`
 	DesignAgentFast       string `toml:"design_agent_fast"`
 	DesignAgentStandard   string `toml:"design_agent_standard"`
+	DesignAgentMedium     string `toml:"design_agent_medium"`
 	DesignAgentThorough   string `toml:"design_agent_thorough"`
+	DesignAgentMaximum    string `toml:"design_agent_maximum"`
 	DesignModel           string `toml:"design_model"`
 	DesignModelFast       string `toml:"design_model_fast"`
 	DesignModelStandard   string `toml:"design_model_standard"`
+	DesignModelMedium     string `toml:"design_model_medium"`
 	DesignModelThorough   string `toml:"design_model_thorough"`
+	DesignModelMaximum    string `toml:"design_model_maximum"`
 
 	// Backup agents for failover
 	ReviewBackupAgent   string `toml:"review_backup_agent"`
@@ -540,8 +560,8 @@ type RepoCIConfig struct {
 	// the ReviewTypes x Agents cross-product for this repo.
 	Reviews map[string][]string `toml:"reviews" comment:"Explicit CI review matrix for this repo: agent name to review types."`
 
-	// Reasoning overrides the reasoning level for CI reviews (thorough, standard, fast).
-	Reasoning string `toml:"reasoning" comment:"Override the CI reasoning level for this repo: fast, standard, or thorough."`
+	// Reasoning overrides the reasoning level for CI reviews.
+	Reasoning string `toml:"reasoning" comment:"Override the CI reasoning level for this repo: fast, standard, medium, thorough, or maximum."`
 
 	// MinSeverity overrides the minimum severity filter for CI synthesis.
 	MinSeverity string `toml:"min_severity" comment:"Override the minimum CI severity included in synthesized output."`
@@ -563,9 +583,9 @@ type RepoConfig struct {
 	ExcludedBranches           []string `toml:"excluded_branches" comment:"Branches that should be skipped for automatic review in this repo."`
 	ExcludedCommitPatterns     []string `toml:"excluded_commit_patterns" comment:"Commit message substrings that should skip review for this repo."`
 	DisplayName                string   `toml:"display_name" comment:"Display name shown for this repo in the TUI and output."`
-	ReviewReasoning            string   `toml:"review_reasoning" comment:"Reasoning level for reviews in this repo: fast, standard, or thorough."`    // Reasoning level for reviews: thorough, standard, fast
-	RefineReasoning            string   `toml:"refine_reasoning" comment:"Reasoning level for refine in this repo: fast, standard, or thorough."`     // Reasoning level for refine: thorough, standard, fast
-	FixReasoning               string   `toml:"fix_reasoning" comment:"Reasoning level for fix in this repo: fast, standard, or thorough."`           // Reasoning level for fix: thorough, standard, fast
+	ReviewReasoning            string   `toml:"review_reasoning" comment:"Reasoning level for reviews in this repo: fast, standard, medium, thorough, or maximum."`
+	RefineReasoning            string   `toml:"refine_reasoning" comment:"Reasoning level for refine in this repo: fast, standard, medium, thorough, or maximum."`
+	FixReasoning               string   `toml:"fix_reasoning" comment:"Reasoning level for fix in this repo: fast, standard, medium, thorough, or maximum."`
 	FixMinSeverity             string   `toml:"fix_min_severity" comment:"Minimum severity for fix in this repo: critical, high, medium, or low."`    // Minimum severity for fix: critical, high, medium, low
 	RefineMinSeverity          string   `toml:"refine_min_severity" comment:"Minimum severity for refine in this repo: critical, high, medium, low."` // Minimum severity for refine: critical, high, medium, low
 	ExcludePatterns            []string `toml:"exclude_patterns" comment:"Filenames or glob patterns to exclude from review diffs for this repo."`
@@ -580,43 +600,63 @@ type RepoConfig struct {
 	ReviewAgent           string `toml:"review_agent" comment:"Agent override for standard review in this repo."`
 	ReviewAgentFast       string `toml:"review_agent_fast" comment:"Agent override for fast review in this repo."`
 	ReviewAgentStandard   string `toml:"review_agent_standard" comment:"Agent override for standard review in this repo."`
+	ReviewAgentMedium     string `toml:"review_agent_medium" comment:"Agent override for medium review in this repo."`
 	ReviewAgentThorough   string `toml:"review_agent_thorough" comment:"Agent override for thorough review in this repo."`
+	ReviewAgentMaximum    string `toml:"review_agent_maximum" comment:"Agent override for maximum review in this repo."`
 	RefineAgent           string `toml:"refine_agent" comment:"Agent override for refine in this repo."`
 	RefineAgentFast       string `toml:"refine_agent_fast" comment:"Agent override for fast refine in this repo."`
 	RefineAgentStandard   string `toml:"refine_agent_standard" comment:"Agent override for standard refine in this repo."`
+	RefineAgentMedium     string `toml:"refine_agent_medium" comment:"Agent override for medium refine in this repo."`
 	RefineAgentThorough   string `toml:"refine_agent_thorough" comment:"Agent override for thorough refine in this repo."`
+	RefineAgentMaximum    string `toml:"refine_agent_maximum" comment:"Agent override for maximum refine in this repo."`
 	ReviewModel           string `toml:"review_model" comment:"Model override for standard review in this repo."`
 	ReviewModelFast       string `toml:"review_model_fast" comment:"Model override for fast review in this repo."`
 	ReviewModelStandard   string `toml:"review_model_standard" comment:"Model override for standard review in this repo."`
+	ReviewModelMedium     string `toml:"review_model_medium" comment:"Model override for medium review in this repo."`
 	ReviewModelThorough   string `toml:"review_model_thorough" comment:"Model override for thorough review in this repo."`
+	ReviewModelMaximum    string `toml:"review_model_maximum" comment:"Model override for maximum review in this repo."`
 	RefineModel           string `toml:"refine_model" comment:"Model override for standard refine in this repo."`
 	RefineModelFast       string `toml:"refine_model_fast" comment:"Model override for fast refine in this repo."`
 	RefineModelStandard   string `toml:"refine_model_standard" comment:"Model override for standard refine in this repo."`
+	RefineModelMedium     string `toml:"refine_model_medium" comment:"Model override for medium refine in this repo."`
 	RefineModelThorough   string `toml:"refine_model_thorough" comment:"Model override for thorough refine in this repo."`
+	RefineModelMaximum    string `toml:"refine_model_maximum" comment:"Model override for maximum refine in this repo."`
 	FixAgent              string `toml:"fix_agent" comment:"Agent override for fix in this repo."`
 	FixAgentFast          string `toml:"fix_agent_fast" comment:"Agent override for fast fix in this repo."`
 	FixAgentStandard      string `toml:"fix_agent_standard" comment:"Agent override for standard fix in this repo."`
+	FixAgentMedium        string `toml:"fix_agent_medium" comment:"Agent override for medium fix in this repo."`
 	FixAgentThorough      string `toml:"fix_agent_thorough" comment:"Agent override for thorough fix in this repo."`
+	FixAgentMaximum       string `toml:"fix_agent_maximum" comment:"Agent override for maximum fix in this repo."`
 	FixModel              string `toml:"fix_model" comment:"Model override for standard fix in this repo."`
 	FixModelFast          string `toml:"fix_model_fast" comment:"Model override for fast fix in this repo."`
 	FixModelStandard      string `toml:"fix_model_standard" comment:"Model override for standard fix in this repo."`
+	FixModelMedium        string `toml:"fix_model_medium" comment:"Model override for medium fix in this repo."`
 	FixModelThorough      string `toml:"fix_model_thorough" comment:"Model override for thorough fix in this repo."`
+	FixModelMaximum       string `toml:"fix_model_maximum" comment:"Model override for maximum fix in this repo."`
 	SecurityAgent         string `toml:"security_agent" comment:"Agent override for security review in this repo."`
 	SecurityAgentFast     string `toml:"security_agent_fast" comment:"Agent override for fast security review in this repo."`
 	SecurityAgentStandard string `toml:"security_agent_standard" comment:"Agent override for standard security review in this repo."`
+	SecurityAgentMedium   string `toml:"security_agent_medium" comment:"Agent override for medium security review in this repo."`
 	SecurityAgentThorough string `toml:"security_agent_thorough" comment:"Agent override for thorough security review in this repo."`
+	SecurityAgentMaximum  string `toml:"security_agent_maximum" comment:"Agent override for maximum security review in this repo."`
 	SecurityModel         string `toml:"security_model" comment:"Model override for standard security review in this repo."`
 	SecurityModelFast     string `toml:"security_model_fast" comment:"Model override for fast security review in this repo."`
 	SecurityModelStandard string `toml:"security_model_standard" comment:"Model override for standard security review in this repo."`
+	SecurityModelMedium   string `toml:"security_model_medium" comment:"Model override for medium security review in this repo."`
 	SecurityModelThorough string `toml:"security_model_thorough" comment:"Model override for thorough security review in this repo."`
+	SecurityModelMaximum  string `toml:"security_model_maximum" comment:"Model override for maximum security review in this repo."`
 	DesignAgent           string `toml:"design_agent" comment:"Agent override for design review in this repo."`
 	DesignAgentFast       string `toml:"design_agent_fast" comment:"Agent override for fast design review in this repo."`
 	DesignAgentStandard   string `toml:"design_agent_standard" comment:"Agent override for standard design review in this repo."`
+	DesignAgentMedium     string `toml:"design_agent_medium" comment:"Agent override for medium design review in this repo."`
 	DesignAgentThorough   string `toml:"design_agent_thorough" comment:"Agent override for thorough design review in this repo."`
+	DesignAgentMaximum    string `toml:"design_agent_maximum" comment:"Agent override for maximum design review in this repo."`
 	DesignModel           string `toml:"design_model" comment:"Model override for standard design review in this repo."`
 	DesignModelFast       string `toml:"design_model_fast" comment:"Model override for fast design review in this repo."`
 	DesignModelStandard   string `toml:"design_model_standard" comment:"Model override for standard design review in this repo."`
+	DesignModelMedium     string `toml:"design_model_medium" comment:"Model override for medium design review in this repo."`
 	DesignModelThorough   string `toml:"design_model_thorough" comment:"Model override for thorough design review in this repo."`
+	DesignModelMaximum    string `toml:"design_model_maximum" comment:"Model override for maximum design review in this repo."`
 
 	// Backup agents for failover
 	ReviewBackupAgent   string `toml:"review_backup_agent" comment:"Backup agent for review in this repo."`
@@ -1074,7 +1114,7 @@ func ValidateReviewTypes(types []string) ([]string, error) {
 }
 
 // NormalizeReasoning validates and normalizes a reasoning level string.
-// Returns the canonical form (thorough, standard, fast) or an error if invalid.
+// Returns the canonical form (maximum, thorough, medium, standard, fast) or an error if invalid.
 // Returns empty string (no error) for empty input.
 func NormalizeReasoning(value string) (string, error) {
 	normalized := strings.ToLower(strings.TrimSpace(value))
@@ -1083,9 +1123,13 @@ func NormalizeReasoning(value string) (string, error) {
 	}
 
 	switch normalized {
+	case "maximum", "max", "xhigh":
+		return "maximum", nil
 	case "thorough", "high":
 		return "thorough", nil
-	case "standard", "medium":
+	case "medium":
+		return "medium", nil
+	case "standard":
 		return "standard", nil
 	case "fast", "low":
 		return "fast", nil
diff --git a/internal/config/config_test.go b/internal/config/config_test.go
index 55a0b83fd..5ddcbc88d 100644
--- a/internal/config/config_test.go
+++ b/internal/config/config_test.go
@@ -1449,6 +1449,19 @@ func TestResolveAgentForWorkflow(t *testing.T) {
 		{"design global level-specific", "", nil, &Config{DesignAgent: "codex", DesignAgentThorough: "claude"}, "design", "thorough", "claude"},
 		{"design isolated from review", "", M{"review_agent": "claude"}, &Config{DefaultAgent: "codex"}, "design", "fast", "codex"},
 		{"design isolated from security", "", M{"security_agent": "claude"}, &Config{DefaultAgent: "codex"}, "design", "fast", "codex"},
+
+		// Maximum level
+		{"maximum repo level-specific", "", M{"review_agent_maximum": "claude"}, nil, "review", "maximum", "claude"},
+		{"maximum global level-specific", "", nil, &Config{ReviewAgentMaximum: "claude"}, "review", "maximum", "claude"},
+		{"maximum falls back to workflow", "", M{"review_agent": "claude"}, nil, "review", "maximum", "claude"},
+		{"maximum falls back to generic", "", M{"agent": "claude"}, nil, "review", "maximum", "claude"},
+		{"fix maximum level-specific", "", M{"fix_agent_maximum": "claude"}, nil, "fix", "maximum", "claude"},
+
+		// Medium level
+		{"medium repo level-specific", "", M{"review_agent_medium": "claude"}, nil, "review", "medium", "claude"},
+		{"medium global level-specific", "", nil, &Config{ReviewAgentMedium: "claude"}, "review", "medium", "claude"},
+		{"medium falls back to workflow", "", M{"review_agent": "claude"}, nil, "review", "medium", "claude"},
+		{"medium falls back to generic", "", M{"agent": "claude"}, nil, "review", "medium", "claude"},
 	}
 
 	for _, tt := range tests {