diff --git a/src/loops/review.go b/src/loops/review.go
index c19ef62..14a2c57 100644
--- a/src/loops/review.go
+++ b/src/loops/review.go
@@ -38,10 +38,18 @@ func RunReview(ctx context.Context, db *lib.DB, available []runners.Runner, git
 		return nil, fmt.Errorf("no diff found — nothing to review")
 	}
 
-	// Truncate diff for prompt
-	const maxDiff = 30000
+	// Truncate diff at file boundaries to preserve reviewable context.
+	// 500K chars ≈ 125K tokens — well within Claude (200K) and Gemini (1M) limits.
+	const maxDiff = 500000
 	if len(diff) > maxDiff {
-		diff = diff[:maxDiff] + "\n... (diff truncated)"
+		// Cut at last file boundary to avoid mid-hunk truncation
+		cut := diff[:maxDiff]
+		if idx := strings.LastIndex(cut, "\ndiff --git "); idx > 0 {
+			cut = cut[:idx]
+			diff = cut + "\n\n... (diff truncated at file boundary — review remaining files separately)"
+		} else {
+			diff = cut + "\n\n... (diff truncated mid-hunk — no clean file boundary found in first 500K chars)"
+		}
 	}
 
 	prompt := cfg.Prompt
diff --git a/src/runners/claude.go b/src/runners/claude.go
index cec6491..401bffb 100644
--- a/src/runners/claude.go
+++ b/src/runners/claude.go
@@ -5,8 +5,10 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"os"
 	"os/exec"
 	"strconv"
+	"strings"
 )
 
 type ClaudeRunner struct{}
@@ -25,11 +27,21 @@ type claudeResponse struct {
 func (r *ClaudeRunner) Name() string { return "claude" }
 
 func (r *ClaudeRunner) Available() bool {
-	_, err := exec.LookPath("claude")
-	return err == nil
+	if _, err := exec.LookPath("claude"); err != nil {
+		return false
+	}
+	// OAuth tokens (sk-ant-oat*) don't work for subprocess claude -p calls.
+	// Only real API keys (sk-ant-api*) or no key (keychain auth) work.
+	key := os.Getenv("ANTHROPIC_API_KEY")
+	if strings.HasPrefix(key, "sk-ant-oat") {
+		fmt.Fprintf(os.Stderr, "claude: skipping — ANTHROPIC_API_KEY is an OAuth token (sk-ant-oat*), which doesn't work for subprocess calls\n")
+		return false
+	}
+	return true
 }
 
 func (r *ClaudeRunner) Run(ctx context.Context, prompt string, opts RunOpts) (RunResult, error) {
+	// claude -p is "print mode" — the prompt is a positional argument.
 	args := []string{
 		"-p", prompt,
 		"--output-format", "json",
@@ -63,17 +75,14 @@ func (r *ClaudeRunner) Run(ctx context.Context, prompt string, opts RunOpts) (Ru
 		if exitErr, ok := err.(*exec.ExitError); ok {
 			exitCode = exitErr.ExitCode()
 		} else {
-			return RunResult{ExitCode: 1}, fmt.Errorf("claude failed to start: %w — is claude CLI installed?", err)
+			return RunResult{ExitCode: 1}, fmt.Errorf("claude failed to run: %w", err)
 		}
 	}
 
 	var resp claudeResponse
 	if err := json.Unmarshal(stdout.Bytes(), &resp); err != nil {
-		// If JSON parsing fails, return raw output
-		return RunResult{
-			Output:   stdout.String(),
-			ExitCode: exitCode,
-		}, nil
+		return RunResult{Output: stdout.String(), ExitCode: exitCode},
+			fmt.Errorf("claude returned non-JSON output: %s", TruncStr(stdout.String(), 200))
 	}
 
 	return RunResult{
diff --git a/src/runners/codex.go b/src/runners/codex.go
index 0ac2f4e..194c581 100644
--- a/src/runners/codex.go
+++ b/src/runners/codex.go
@@ -17,6 +17,8 @@ func (r *CodexRunner) Available() bool {
 }
 
 func (r *CodexRunner) Run(ctx context.Context, prompt string, opts RunOpts) (RunResult, error) {
+	// Codex reads stdin and appends to the prompt argument.
+	// Pipe the full prompt via stdin to handle large diffs safely.
 	args := []string{"exec", "--full-auto", prompt}
 
 	cmd := exec.CommandContext(ctx, "codex", args...)
@@ -34,12 +36,20 @@ func (r *CodexRunner) Run(ctx context.Context, prompt string, opts RunOpts) (Run
 		if exitErr, ok := err.(*exec.ExitError); ok {
 			exitCode = exitErr.ExitCode()
 		} else {
-			return RunResult{ExitCode: 1}, fmt.Errorf("codex failed to start: %w", err)
+			return RunResult{ExitCode: 1}, fmt.Errorf("codex failed to run: %w", err)
 		}
 	}
 
-	return RunResult{
+	result := RunResult{
 		Output:   stdout.String(),
 		ExitCode: exitCode,
-	}, nil
+	}
+	if exitCode != 0 {
+		errMsg := stderr.String()
+		if errMsg == "" {
+			errMsg = stdout.String()
+		}
+		return result, fmt.Errorf("codex exited %d: %s", exitCode, TruncStr(errMsg, 200))
+	}
+	return result, nil
 }
diff --git a/src/runners/gemini.go b/src/runners/gemini.go
index 8a26ccc..9761315 100644
--- a/src/runners/gemini.go
+++ b/src/runners/gemini.go
@@ -5,6 +5,7 @@ import (
 	"context"
 	"fmt"
 	"os/exec"
+	"strings"
 )
 
 type GeminiRunner struct{}
@@ -17,8 +18,9 @@ func (r *GeminiRunner) Available() bool {
 }
 
 func (r *GeminiRunner) Run(ctx context.Context, prompt string, opts RunOpts) (RunResult, error) {
+	// Gemini reads from stdin and -p appends to it. To avoid duplicate
+	// instructions, pipe the full prompt via stdin only — no -p flag.
 	args := []string{
-		"-p", prompt,
 		"-y",
 		"--output-format", "text",
 	}
@@ -28,6 +30,8 @@ func (r *GeminiRunner) Run(ctx context.Context, prompt string, opts RunOpts) (Ru
 		cmd.Dir = opts.WorkDir
 	}
 
+	cmd.Stdin = strings.NewReader(prompt)
+
 	var stdout, stderr bytes.Buffer
 	cmd.Stdout = &stdout
 	cmd.Stderr = &stderr
@@ -38,7 +42,7 @@ func (r *GeminiRunner) Run(ctx context.Context, prompt string, opts RunOpts) (Ru
 		if exitErr, ok := err.(*exec.ExitError); ok {
 			exitCode = exitErr.ExitCode()
 		} else {
-			return RunResult{ExitCode: 1}, fmt.Errorf("gemini failed to start: %w", err)
+			return RunResult{ExitCode: 1}, fmt.Errorf("gemini failed to run: %w", err)
 		}
 	}