diff --git a/benches/benchmarks.rs b/benches/benchmarks.rs
index dea4dcb..d20ebf4 100644
--- a/benches/benchmarks.rs
+++ b/benches/benchmarks.rs
@@ -3,37 +3,189 @@
 #![allow(missing_docs)]
 #![allow(let_underscore_drop)]
 
-use criterion::{black_box, criterion_group, criterion_main, Criterion};
+use agent_precommit::{Config, Detector, Mode};
+use criterion::{black_box, criterion_group, criterion_main, BenchmarkId, Criterion};
 
+/// Benchmark the full mode detection pipeline.
+///
+/// This tests the `Detector::detect()` method which checks:
+/// - APC_MODE environment variable
+/// - AGENT_MODE flag
+/// - Known agent environment variables (20+ vars)
+/// - Custom agent environment variables from config
+/// - CI environment variables (20+ vars)
+/// - TTY detection
 fn benchmark_mode_detection(c: &mut Criterion) {
-    c.bench_function("mode_detection", |b| {
+    let config = Config::default();
+
+    c.bench_function("detector_detect", |b| {
+        b.iter(|| {
+            let detector = Detector::new(black_box(&config));
+            black_box(detector.detect())
+        });
+    });
+}
+
+/// Benchmark mode detection with custom agent environment variables configured.
+fn benchmark_mode_detection_with_custom_vars(c: &mut Criterion) {
+    let mut config = Config::default();
+    // Add custom agent environment variables to simulate real-world config
+    config.detection.agent_env_vars = vec![
+        "MY_CUSTOM_AGENT".to_string(),
+        "INTERNAL_BOT_TOKEN".to_string(),
+        "AUTOMATION_SERVICE".to_string(),
+    ];
+
+    c.bench_function("detector_detect_with_custom_vars", |b| {
         b.iter(|| {
-            // Simple benchmark placeholder
-            // In a real benchmark, we'd test the detector
-            black_box(1 + 1)
+            let detector = Detector::new(black_box(&config));
+            black_box(detector.detect())
         });
     });
 }
 
+/// Benchmark Mode::from_str parsing for different inputs.
+fn benchmark_mode_parsing(c: &mut Criterion) {
+    let mut group = c.benchmark_group("mode_parsing");
+
+    let test_cases = ["human", "agent", "ci", "HUMAN", "Agent", "CI"];
+
+    for input in test_cases {
+        group.bench_with_input(BenchmarkId::from_parameter(input), input, |b, input| {
+            b.iter(|| {
+                let result: Result<Mode, _> = black_box(input).parse();
+                black_box(result)
+            });
+        });
+    }
+
+    group.finish();
+}
+
+/// Benchmark Mode helper methods.
+fn benchmark_mode_methods(c: &mut Criterion) {
+    let mut group = c.benchmark_group("mode_methods");
+
+    let modes = [Mode::Human, Mode::Agent, Mode::Ci];
+
+    for mode in modes {
+        group.bench_with_input(BenchmarkId::new("name", mode.name()), &mode, |b, mode| {
+            b.iter(|| black_box(mode.name()));
+        });
+
+        group.bench_with_input(
+            BenchmarkId::new("is_thorough", mode.name()),
+            &mode,
+            |b, mode| {
+                b.iter(|| black_box(mode.is_thorough()));
+            },
+        );
+
+        group.bench_with_input(
+            BenchmarkId::new("display", mode.name()),
+            &mode,
+            |b, mode| {
+                b.iter(|| black_box(mode.to_string()));
+            },
+        );
+    }
+
+    group.finish();
+}
+
+/// Benchmark config parsing with the actual Config struct.
 fn benchmark_config_parsing(c: &mut Criterion) {
     let toml_content = r#"
+[detection]
+agent_env_vars = ["MY_AGENT", "CUSTOM_BOT"]
+
+[human]
+checks = ["pre-commit"]
+timeout = "30s"
+fail_fast = true
+
+[agent]
+checks = ["pre-commit-all", "test-unit", "build-verify"]
+timeout = "15m"
+fail_fast = false
+
+[checks.custom-lint]
+run = "npm run lint"
+description = "Run ESLint"
+
+[checks.custom-lint.enabled_if]
+file_exists = "package.json"
+"#;
+
+    c.bench_function("config_parsing_full", |b| {
+        b.iter(|| {
+            let result: Config = toml::from_str(black_box(toml_content)).expect("parse config");
+            black_box(result)
+        });
+    });
+}
+
+/// Benchmark config parsing for minimal config.
+fn benchmark_config_parsing_minimal(c: &mut Criterion) {
+    let toml_content = r#"
 [human]
 checks = ["pre-commit"]
 timeout = "30s"
 
 [agent]
-checks = ["pre-commit-all", "test-unit"]
+checks = ["pre-commit-all"]
 timeout = "15m"
 "#;
 
-    c.bench_function("config_parsing", |b| {
+    c.bench_function("config_parsing_minimal", |b| {
         b.iter(|| {
-            let result: toml::Value =
-                toml::from_str(black_box(toml_content)).expect("parse config");
+            let result: Config = toml::from_str(black_box(toml_content)).expect("parse config");
             black_box(result)
         });
     });
 }
 
-criterion_group!(benches, benchmark_mode_detection, benchmark_config_parsing);
+/// Benchmark config default generation.
+fn benchmark_config_default(c: &mut Criterion) {
+    c.bench_function("config_default", |b| {
+        b.iter(|| black_box(Config::default()));
+    });
+}
+
+/// Benchmark config preset generation.
+fn benchmark_config_presets(c: &mut Criterion) {
+    let mut group = c.benchmark_group("config_presets");
+
+    let presets = ["python", "rust", "node", "go"];
+
+    for preset in presets {
+        group.bench_with_input(BenchmarkId::from_parameter(preset), preset, |b, preset| {
+            b.iter(|| black_box(Config::for_preset(black_box(preset))));
+        });
+    }
+
+    group.finish();
+}
+
+/// Benchmark config validation.
+fn benchmark_config_validation(c: &mut Criterion) {
+    let config = Config::default();
+
+    c.bench_function("config_validation", |b| {
+        b.iter(|| black_box(config.validate()));
+    });
+}
+
+criterion_group!(
+    benches,
+    benchmark_mode_detection,
+    benchmark_mode_detection_with_custom_vars,
+    benchmark_mode_parsing,
+    benchmark_mode_methods,
+    benchmark_config_parsing,
+    benchmark_config_parsing_minimal,
+    benchmark_config_default,
+    benchmark_config_presets,
+    benchmark_config_validation,
+);
 criterion_main!(benches);