@oddessentials/odd-ai-reviewers 1.12.0 → 1.13.0

package/dist/cli/commands/local-review.js

@@ -32,6 +32,7 @@ import { setupSignalHandlers, setPartialResultsContext, clearPartialResultsConte
 import { countBySeverity } from '../../report/formats.js';
 import { existsSync } from 'fs';
 import { join, resolve } from 'path';
+import { format as formatConsoleArgs } from 'util';
 import { checkDependenciesForPasses, displayDependencyErrors, displaySkippedPassWarnings, getDependenciesForAgent, } from '../dependencies/index.js';
 // =============================================================================
 // Default Dependencies
@@ -44,23 +45,6 @@ export function createDefaultDependencies() {
         env: process.env,
         exitHandler: (code) => {
             process.exitCode = code;
-            const exit = () => process.exit(code);
-            const drainingStreams = [process.stdout, process.stderr].filter((stream) => stream.writableNeedDrain);
-            if (drainingStreams.length === 0) {
-                exit();
-                return;
-            }
-            let remaining = drainingStreams.length;
-            for (const stream of drainingStreams) {
-                stream.once('drain', () => {
-                    remaining -= 1;
-                    if (remaining === 0) {
-                        exit();
-                    }
-                });
-            }
-            // Safety: avoid hanging if drain never fires
-            setTimeout(exit, 100);
         },
         stdout: process.stdout,
         stderr: process.stderr,
@@ -88,6 +72,28 @@ export const ExitCode = {
     /** Incomplete review (partial results available) */
     INCOMPLETE: 3,
 };
+async function withConsoleRedirectToStderr(enabled, stderr, fn) {
+    if (!enabled) {
+        return fn();
+    }
+    const originalLog = console.log;
+    const originalWarn = console.warn;
+    const originalInfo = console.info;
+    const redirect = (...args) => {
+        stderr.write(formatConsoleArgs(...args) + '\n');
+    };
+    console.log = redirect;
+    console.warn = redirect;
+    console.info = redirect;
+    try {
+        return await fn();
+    }
+    finally {
+        console.log = originalLog;
+        console.warn = originalWarn;
+        console.info = originalInfo;
+    }
+}
 // =============================================================================
 // Helper Functions
 // =============================================================================
@@ -536,46 +542,21 @@ export async function runLocalReview(rawOptions, deps = createDefaultDependencie
         };
     }
     const { options, warnings } = parseResult.value;
-    // Print warnings
-    for (const warning of warnings) {
-        stderr.write(c.yellow(`Warning: ${warning}`) + '\n');
-    }
-    // 2. Infer git context
-    const inferFn = deps.inferGitContext ?? inferGitContext;
-    const gitResult = inferFn(resolve(options.path));
-    if (!isOk(gitResult)) {
-        const error = gitResult.error.code === GitContextErrorCode.GIT_NOT_FOUND
-            ? new GitNotFoundError(gitResult.error.message)
-            : gitResult.error.code === GitContextErrorCode.INVALID_PATH
-                ? new InvalidPathError(gitResult.error.message, gitResult.error.path)
-                : new NotAGitRepoError(gitResult.error.path ?? options.path);
-        stderr.write(formatCLIError(error, colored) + '\n');
-        return {
-            exitCode: ExitCode.INVALID_ARGS,
-            findingsCount: 0,
-            partialFindingsCount: 0,
-            error: error.message,
-        };
-    }
-    const gitContext = gitResult.value;
-    // Apply defaults from git context
-    const resolvedOptions = applyOptionDefaults(options, gitContext);
-    const diffRange = resolveDiffRange(resolvedOptions, gitContext);
-    const baseRef = diffRange.baseRef;
-    // 3. Load configuration (with zero-config fallback)
-    let config;
-    let configSource;
-    let configPath;
-    let zeroConfigResult;
-    try {
-        const configResult = await loadConfigWithFallback(gitContext.repoRoot, env, deps, resolvedOptions.config);
-        config = configResult.config;
-        configSource = configResult.source;
-        configPath = configResult.path;
-        zeroConfigResult = configResult.zeroConfigResult;
-    }
-    catch (error) {
-        if (error instanceof NoCredentialsError) {
+    const isMachineReadableFormat = options.format === 'json' || options.format === 'sarif';
+    return withConsoleRedirectToStderr(isMachineReadableFormat, stderr, async () => {
+        // Print warnings
+        for (const warning of warnings) {
+            stderr.write(c.yellow(`Warning: ${warning}`) + '\n');
+        }
+        // 2. Infer git context
+        const inferFn = deps.inferGitContext ?? inferGitContext;
+        const gitResult = inferFn(resolve(options.path));
+        if (!isOk(gitResult)) {
+            const error = gitResult.error.code === GitContextErrorCode.GIT_NOT_FOUND
+                ? new GitNotFoundError(gitResult.error.message)
+                : gitResult.error.code === GitContextErrorCode.INVALID_PATH
+                    ? new InvalidPathError(gitResult.error.message, gitResult.error.path)
+                    : new NotAGitRepoError(gitResult.error.path ?? options.path);
             stderr.write(formatCLIError(error, colored) + '\n');
             return {
                 exitCode: ExitCode.INVALID_ARGS,
@@ -584,327 +565,359 @@ export async function runLocalReview(rawOptions, deps = createDefaultDependencie
                 error: error.message,
             };
         }
-        const errorMsg = error instanceof Error ? error.message : String(error);
-        const configError = new InvalidConfigError(resolvedOptions.config ?? '.ai-review.yml', [
-            errorMsg,
-        ]);
-        stderr.write(formatCLIError(configError, colored) + '\n');
-        return {
-            exitCode: ExitCode.INVALID_ARGS,
-            findingsCount: 0,
-            partialFindingsCount: 0,
-            error: errorMsg,
-        };
-    }
-    // Show zero-config message if applicable
-    if (configSource === 'zero-config' && zeroConfigResult && !resolvedOptions.quiet) {
-        stdout.write(c.yellow(`\nUsing ${zeroConfigResult.provider} (${zeroConfigResult.keySource} found)\n`));
-        if (zeroConfigResult.ignoredProviders.length > 0) {
-            for (const ignored of zeroConfigResult.ignoredProviders) {
-                stdout.write(c.gray(`Note: ${ignored.keySource} also set but ignored due to priority order\n`));
+        const gitContext = gitResult.value;
+        // Apply defaults from git context
+        const resolvedOptions = applyOptionDefaults(options, gitContext);
+        const diffRange = resolveDiffRange(resolvedOptions, gitContext);
+        const baseRef = diffRange.baseRef;
+        // 3. Load configuration (with zero-config fallback)
+        let config;
+        let configSource;
+        let configPath;
+        let zeroConfigResult;
+        try {
+            const configResult = await loadConfigWithFallback(gitContext.repoRoot, env, deps, resolvedOptions.config);
+            config = configResult.config;
+            configSource = configResult.source;
+            configPath = configResult.path;
+            zeroConfigResult = configResult.zeroConfigResult;
+        }
+        catch (error) {
+            if (error instanceof NoCredentialsError) {
+                stderr.write(formatCLIError(error, colored) + '\n');
+                return {
+                    exitCode: ExitCode.INVALID_ARGS,
+                    findingsCount: 0,
+                    partialFindingsCount: 0,
+                    error: error.message,
+                };
             }
+            const errorMsg = error instanceof Error ? error.message : String(error);
+            const configError = new InvalidConfigError(resolvedOptions.config ?? '.ai-review.yml', [
+                errorMsg,
+            ]);
+            stderr.write(formatCLIError(configError, colored) + '\n');
+            return {
+                exitCode: ExitCode.INVALID_ARGS,
+                findingsCount: 0,
+                partialFindingsCount: 0,
+                error: errorMsg,
+            };
         }
-        stdout.write(c.gray('Tip: Create .ai-review.yml to customize settings\n\n'));
-    }
-    // 4. Build execution plan (FR-001 through FR-007)
-    // The plan is the single source of truth for all downstream code paths.
-    // No downstream consumer may read raw CLI flags — they operate on the plan.
-    let executionPlan;
-    try {
-        const planMode = resolvedOptions.dryRun
-            ? 'dry-run'
-            : resolvedOptions.costOnly
-                ? 'cost-only'
-                : 'execute';
-        executionPlan = buildExecutionPlan({
-            config,
-            mode: planMode,
-            passFilter: resolvedOptions.pass,
-            agentFilter: resolvedOptions.agent,
-            provider: zeroConfigResult?.provider ?? config.provider ?? null,
-            model: config.models?.default ?? null,
-            configSource: configSource === 'zero-config' ? 'zero-config' : (configPath ?? '.ai-review.yml'),
-        });
-    }
-    catch (error) {
-        if (error instanceof ConfigError) {
-            stderr.write(formatCLIError(new InvalidConfigError(resolvedOptions.config ?? '.ai-review.yml', [error.message]), colored) + '\n');
+        // Show zero-config guidance only for human-readable terminal output.
+        // Machine-readable modes must reserve stdout exclusively for the payload.
+        if (configSource === 'zero-config' &&
+            zeroConfigResult &&
+            !resolvedOptions.quiet &&
+            !isMachineReadableFormat) {
+            stdout.write(c.yellow(`\nUsing ${zeroConfigResult.provider} (${zeroConfigResult.keySource} found)\n`));
+            if (zeroConfigResult.ignoredProviders.length > 0) {
+                for (const ignored of zeroConfigResult.ignoredProviders) {
+                    stdout.write(c.gray(`Note: ${ignored.keySource} also set but ignored due to priority order\n`));
+                }
+            }
+            stdout.write(c.gray('Tip: Create .ai-review.yml to customize settings\n\n'));
+        }
+        // 4. Build execution plan (FR-001 through FR-007)
+        // The plan is the single source of truth for all downstream code paths.
+        // No downstream consumer may read raw CLI flags — they operate on the plan.
+        let executionPlan;
+        try {
+            const planMode = resolvedOptions.dryRun
+                ? 'dry-run'
+                : resolvedOptions.costOnly
+                    ? 'cost-only'
+                    : 'execute';
+            executionPlan = buildExecutionPlan({
+                config,
+                mode: planMode,
+                passFilter: resolvedOptions.pass,
+                agentFilter: resolvedOptions.agent,
+                provider: zeroConfigResult?.provider ?? config.provider ?? null,
+                model: config.models?.default ?? null,
+                configSource: configSource === 'zero-config' ? 'zero-config' : (configPath ?? '.ai-review.yml'),
+            });
+        }
+        catch (error) {
+            if (error instanceof ConfigError) {
+                stderr.write(formatCLIError(new InvalidConfigError(resolvedOptions.config ?? '.ai-review.yml', [error.message]), colored) + '\n');
+                return {
+                    exitCode: exitCodeFromStatus('config_error'),
+                    status: 'config_error',
+                    findingsCount: 0,
+                    partialFindingsCount: 0,
+                    error: error.message,
+                };
+            }
+            throw error;
+        }
+        // Emit plan in verbose mode for debugging
+        if (resolvedOptions.verbose && !resolvedOptions.quiet) {
+            stderr.write(c.gray('Execution plan:\n'));
+            stderr.write(c.gray(serializeExecutionPlan(executionPlan)));
+            stderr.write('\n');
+        }
+        // Build a config-like passes array from the execution plan for downstream compatibility
+        const planPasses = executionPlan.passes.map((p) => ({
+            name: p.name,
+            agents: [...p.agents],
+            enabled: true,
+            required: p.required,
+        }));
+        // 5. Handle special modes (dry-run and cost-only) - before dependency check
+        // These modes don't execute agents, so dependencies are not required
+        if (resolvedOptions.dryRun) {
+            const dryRunResult = await executeDryRun(resolvedOptions, gitContext, config, configSource, configPath, deps);
+            // Override agents to match the execution plan (FR-002: dry-run shows only plan agents)
+            dryRunResult.agents = [...new Set(executionPlan.passes.flatMap((p) => [...p.agents]))];
+            const output = formatDryRunOutput(dryRunResult, colored, resolvedOptions.format);
+            stdout.write(output);
+            return {
+                exitCode: ExitCode.SUCCESS,
+                status: 'complete',
+                findingsCount: 0,
+                partialFindingsCount: 0,
+            };
+        }
+        if (resolvedOptions.costOnly) {
+            // FR-003: Cost estimation MUST be scoped to the execution plan's passes only
+            const planFilteredConfig = {
+                ...config,
+                passes: planPasses.map((p) => ({
+                    name: p.name,
+                    agents: [...p.agents],
+                    enabled: true,
+                    required: p.required,
+                })),
+            };
+            const costResult = await executeCostOnly(resolvedOptions, gitContext, planFilteredConfig, deps);
+            const output = formatCostOutput(costResult, colored);
+            stdout.write(output);
+            return {
+                exitCode: ExitCode.SUCCESS,
+                status: 'complete',
+                findingsCount: 0,
+                partialFindingsCount: 0,
+            };
+        }
+        // 6. Dependency preflight check — scoped to execution plan's passes only (FR-001)
+        const checkDepsFn = deps.checkDependenciesForPasses ?? checkDependenciesForPasses;
+        const depSummary = checkDepsFn(planPasses);
+        if (depSummary.hasBlockingIssues) {
+            displayDependencyErrors(depSummary, stderr);
             return {
                 exitCode: exitCodeFromStatus('config_error'),
                 status: 'config_error',
                 findingsCount: 0,
                 partialFindingsCount: 0,
-                error: error.message,
+                error: 'Missing required dependencies',
             };
         }
-        throw error;
-    }
-    // Emit plan in verbose mode for debugging
-    if (resolvedOptions.verbose && !resolvedOptions.quiet) {
-        stderr.write(c.gray('Execution plan:\n'));
-        stderr.write(c.gray(serializeExecutionPlan(executionPlan)));
-        stderr.write('\n');
-    }
-    // Build a config-like passes array from the execution plan for downstream compatibility
-    const planPasses = executionPlan.passes.map((p) => ({
-        name: p.name,
-        agents: [...p.agents],
-        enabled: true,
-        required: p.required,
-    }));
-    // 5. Handle special modes (dry-run and cost-only) - before dependency check
-    // These modes don't execute agents, so dependencies are not required
-    if (resolvedOptions.dryRun) {
-        const dryRunResult = await executeDryRun(resolvedOptions, gitContext, config, configSource, configPath, deps);
-        // Override agents to match the execution plan (FR-002: dry-run shows only plan agents)
-        dryRunResult.agents = [...new Set(executionPlan.passes.flatMap((p) => [...p.agents]))];
-        const output = formatDryRunOutput(dryRunResult, colored, resolvedOptions.format);
-        stdout.write(output);
-        return {
-            exitCode: ExitCode.SUCCESS,
-            status: 'complete',
-            findingsCount: 0,
-            partialFindingsCount: 0,
-        };
-    }
-    if (resolvedOptions.costOnly) {
-        // FR-003: Cost estimation MUST be scoped to the execution plan's passes only
-        const planFilteredConfig = {
-            ...config,
-            passes: planPasses.map((p) => ({
-                name: p.name,
-                agents: [...p.agents],
-                enabled: true,
-                required: p.required,
-            })),
-        };
-        const costResult = await executeCostOnly(resolvedOptions, gitContext, planFilteredConfig, deps);
-        const output = formatCostOutput(costResult, colored);
-        stdout.write(output);
-        return {
-            exitCode: ExitCode.SUCCESS,
-            status: 'complete',
-            findingsCount: 0,
-            partialFindingsCount: 0,
-        };
-    }
-    // 6. Dependency preflight check — scoped to execution plan's passes only (FR-001)
-    const checkDepsFn = deps.checkDependenciesForPasses ?? checkDependenciesForPasses;
-    const depSummary = checkDepsFn(planPasses);
-    if (depSummary.hasBlockingIssues) {
-        displayDependencyErrors(depSummary, stderr);
-        return {
-            exitCode: exitCodeFromStatus('config_error'),
-            status: 'config_error',
-            findingsCount: 0,
-            partialFindingsCount: 0,
-            error: 'Missing required dependencies',
-        };
-    }
-    // Build skipped pass info for warnings (using plan passes, not raw config)
-    const skippedPassInfo = buildSkippedPassInfo(planPasses, depSummary);
-    // Show warnings for skipped passes (graceful degradation)
-    if (skippedPassInfo.length > 0 && !resolvedOptions.quiet) {
-        displaySkippedPassWarnings(skippedPassInfo, stderr);
-    }
-    // Show execution plan's skipped passes
-    if (executionPlan.skippedPasses.length > 0 && !resolvedOptions.quiet) {
-        for (const sp of executionPlan.skippedPasses) {
-            stderr.write(c.yellow(`Note: ${sp.reason}\n`));
+        // Build skipped pass info for warnings (using plan passes, not raw config)
+        const skippedPassInfo = buildSkippedPassInfo(planPasses, depSummary);
+        // Show warnings for skipped passes (graceful degradation)
+        if (skippedPassInfo.length > 0 && !resolvedOptions.quiet) {
+            displaySkippedPassWarnings(skippedPassInfo, stderr);
         }
-    }
-    // Filter config to only include runnable passes from the execution plan
-    const runnableConfig = filterToRunnablePasses({ ...config, passes: planPasses }, depSummary.runnablePasses);
-    // Check if all passes were filtered out (all optional, all missing dependencies)
-    const enabledRunnablePasses = runnableConfig.passes.filter((p) => p.enabled);
-    if (enabledRunnablePasses.length === 0) {
-        const warnMsg = colored
-            ? `${c.yellow('⚠')} All review passes were skipped due to missing dependencies.\n` +
-                `  Run 'ai-review check' to see what's missing.\n` +
-                `  No review will be performed.\n`
-            : `Warning: All review passes were skipped due to missing dependencies.\n` +
-                `  Run 'ai-review check' to see what's missing.\n` +
-                `  No review will be performed.\n`;
-        stderr.write(warnMsg);
-        return {
-            exitCode: ExitCode.SUCCESS,
-            status: 'complete',
-            findingsCount: 0,
-            partialFindingsCount: 0,
-        };
-    }
-    // Show warnings for other optional missing dependencies
-    if (depSummary.hasWarnings && !resolvedOptions.quiet && skippedPassInfo.length === 0) {
-        displayDependencyErrors(depSummary, stderr);
-    }
-    // 7. Load .reviewignore patterns (before diff to filter early)
-    const reviewIgnoreResult = await loadReviewIgnore(gitContext.repoRoot);
-    // 7. Generate diff once with reviewignore filtering applied
-    const getDiffFn = deps.getLocalDiff ?? getLocalDiff;
-    const diff = getDiffFn(gitContext.repoRoot, {
-        baseRef,
-        headRef: diffRange.headRef,
-        rangeOperator: diffRange.rangeOperator,
-        stagedOnly: resolvedOptions.staged,
-        uncommitted: resolvedOptions.uncommitted,
-        pathFilter: reviewIgnoreResult.patterns.length > 0
-            ? { reviewIgnorePatterns: reviewIgnoreResult.patterns }
-            : undefined,
-    });
-    // 8. Check for changes (using already-generated diff)
-    if (diff.files.length === 0) {
-        // No changes to review - could be no changes or all filtered by .reviewignore
-        const headLabel = resolvedOptions.staged
-            ? 'STAGED'
-            : resolvedOptions.uncommitted
-                ? 'WORKTREE'
-                : diffRange.headRef;
-        const output = colored
-            ? `${c.green('✓')} No changes to review\n\n  Base: ${baseRef}\n  Head: ${headLabel}\n\n  No uncommitted or staged changes found.\n`
-            : `No changes to review\n\n  Base: ${baseRef}\n  Head: ${headLabel}\n\n  No uncommitted or staged changes found.\n`;
-        stdout.write(output);
-        return {
-            exitCode: ExitCode.SUCCESS,
-            findingsCount: 0,
-            partialFindingsCount: 0,
-            status: 'complete',
-        };
-    }
-    // 9. Build agent context (using runnableConfig with filtered passes)
-    const routerEnv = buildRouterEnv(env);
-    const agentContext = buildAgentContext(gitContext.repoRoot, diff, runnableConfig, routerEnv);
-    const configHash = hashConfig(config); // Use original config for consistent cache key
-    // 10. Check budget
-    const diffContent = buildCombinedDiff(diff.files, config.limits?.max_diff_lines ?? 5000);
-    const estimatedTokensCount = estimateTokens(diffContent);
-    const budgetContext = {
-        fileCount: diff.files.length,
-        diffLines: diff.totalAdditions + diff.totalDeletions,
-        estimatedTokens: estimatedTokensCount,
-    };
-    const budgetCheck = checkBudget(budgetContext, config.limits ?? {
-        max_files: 50,
-        max_diff_lines: 2000,
-        max_tokens_per_pr: 50000,
-        max_usd_per_pr: 0.1,
-        monthly_budget_usd: 10,
-    });
-    // 11. Setup signal handlers for graceful shutdown
-    // exitOnSignal defaults to true - first Ctrl+C stops execution immediately
-    // This is the correct behavior for CLI tools to avoid runaway costs
-    setupSignalHandlers({
-        // Cleanup must be SYNCHRONOUS to guarantee completion before process.exit()
-        // Only uses stdout.write() which is sync - no async I/O allowed here
-        cleanup: () => {
-            // Log partial results context if available
-            const ctx = getPartialResultsContext();
-            if (ctx && ctx.completedAgents > 0) {
-                const lines = formatPartialResultsMessage(ctx);
-                for (const line of lines) {
-                    stdout.write(line + '\n');
-                }
+        // Show execution plan's skipped passes
+        if (executionPlan.skippedPasses.length > 0 && !resolvedOptions.quiet) {
+            for (const sp of executionPlan.skippedPasses) {
+                stderr.write(c.yellow(`Note: ${sp.reason}\n`));
             }
-        },
-        showPartialResultsMessage: true,
-        // exitOnSignal: true (default) - process.exit() called on first signal
-        exitOnSignal: true,
-        logger: {
-            log: (msg) => stdout.write(msg + '\n'),
-            warn: (msg) => stderr.write(msg + '\n'),
-        },
-    });
-    // 12. Execute agent passes
-    let executeResult;
-    try {
-        // Set up partial results tracking (using runnableConfig which excludes skipped passes)
-        const totalAgents = runnableConfig.passes.reduce((sum, p) => (p.enabled ? sum + p.agents.length : sum), 0);
-        setPartialResultsContext({
-            totalAgents,
-            completedAgents: 0,
-            completedAgentNames: [],
-            currentAgent: undefined,
+        }
+        // Filter config to only include runnable passes from the execution plan
+        const runnableConfig = filterToRunnablePasses({ ...config, passes: planPasses }, depSummary.runnablePasses);
+        // Check if all passes were filtered out (all optional, all missing dependencies)
+        const enabledRunnablePasses = runnableConfig.passes.filter((p) => p.enabled);
+        if (enabledRunnablePasses.length === 0) {
+            const warnMsg = colored
+                ? `${c.yellow('⚠')} All review passes were skipped due to missing dependencies.\n` +
+                    `  Run 'ai-review check' to see what's missing.\n` +
+                    `  No review will be performed.\n`
+                : `Warning: All review passes were skipped due to missing dependencies.\n` +
+                    `  Run 'ai-review check' to see what's missing.\n` +
+                    `  No review will be performed.\n`;
+            stderr.write(warnMsg);
+            return {
+                exitCode: ExitCode.SUCCESS,
+                status: 'complete',
+                findingsCount: 0,
+                partialFindingsCount: 0,
+            };
+        }
+        // Show warnings for other optional missing dependencies
+        if (depSummary.hasWarnings && !resolvedOptions.quiet && skippedPassInfo.length === 0) {
+            displayDependencyErrors(depSummary, stderr);
+        }
+        // 7. Load .reviewignore patterns (before diff to filter early)
+        const reviewIgnoreResult = await loadReviewIgnore(gitContext.repoRoot);
+        // 7. Generate diff once with reviewignore filtering applied
+        const getDiffFn = deps.getLocalDiff ?? getLocalDiff;
+        const diff = getDiffFn(gitContext.repoRoot, {
+            baseRef,
+            headRef: diffRange.headRef,
+            rangeOperator: diffRange.rangeOperator,
+            stagedOnly: resolvedOptions.staged,
+            uncommitted: resolvedOptions.uncommitted,
+            pathFilter: reviewIgnoreResult.patterns.length > 0
+                ? { reviewIgnorePatterns: reviewIgnoreResult.patterns }
+                : undefined,
         });
-        const executeFn = deps.executeAllPasses ?? executeAllPasses;
-        executeResult = await executeFn(runnableConfig, agentContext, routerEnv, budgetCheck, {
-            configHash,
-            head: diff.headSha,
+        // 8. Check for changes (using already-generated diff)
+        if (diff.files.length === 0) {
+            // No changes to review - could be no changes or all filtered by .reviewignore
+            const headLabel = resolvedOptions.staged
+                ? 'STAGED'
+                : resolvedOptions.uncommitted
+                    ? 'WORKTREE'
+                    : diffRange.headRef;
+            const output = colored
+                ? `${c.green('✓')} No changes to review\n\n  Base: ${baseRef}\n  Head: ${headLabel}\n\n  No uncommitted or staged changes found.\n`
+                : `No changes to review\n\n  Base: ${baseRef}\n  Head: ${headLabel}\n\n  No uncommitted or staged changes found.\n`;
+            stdout.write(output);
+            return {
+                exitCode: ExitCode.SUCCESS,
+                findingsCount: 0,
+                partialFindingsCount: 0,
+                status: 'complete',
+            };
+        }
+        // 9. Build agent context (using runnableConfig with filtered passes)
+        const routerEnv = buildRouterEnv(env);
+        const agentContext = buildAgentContext(gitContext.repoRoot, diff, runnableConfig, routerEnv);
+        const configHash = hashConfig(config); // Use original config for consistent cache key
+        // 10. Check budget
+        const diffContent = buildCombinedDiff(diff.files, config.limits?.max_diff_lines ?? 5000);
+        const estimatedTokensCount = estimateTokens(diffContent);
+        const budgetContext = {
+            fileCount: diff.files.length,
+            diffLines: diff.totalAdditions + diff.totalDeletions,
+            estimatedTokens: estimatedTokensCount,
+        };
+        const budgetCheck = checkBudget(budgetContext, config.limits ?? {
+            max_files: 50,
+            max_diff_lines: 2000,
+            max_tokens_per_pr: 50000,
+            max_usd_per_pr: 0.1,
+            monthly_budget_usd: 10,
         });
-    }
-    catch (error) {
-        clearPartialResultsContext();
-        const errorMsg = error instanceof Error ? error.message : String(error);
-        // FR-021: Extract partial results from FatalExecutionError when available.
-        // If agents completed before the failure, report their findings in degraded mode.
-        if (error instanceof FatalExecutionError && error.partialResults) {
-            stderr.write(c.yellow(`\n⚠ Incomplete review: ${errorMsg}\n`));
-            stderr.write(c.yellow(`  ${error.partialResults.completeFindings.length} findings from completed agents preserved\n`));
-            // Process partial results through the standard pipeline
-            const executionTimeMs = Date.now() - startTime;
-            const estimatedCostUsd = error.partialResults.allResults.reduce((sum, r) => {
-                if ('metrics' in r && r.metrics?.estimatedCostUsd) {
-                    return sum + r.metrics.estimatedCostUsd;
+        // 11. Setup signal handlers for graceful shutdown
+        // exitOnSignal defaults to true - first Ctrl+C stops execution immediately
+        // This is the correct behavior for CLI tools to avoid runaway costs
+        setupSignalHandlers({
+            // Cleanup must be SYNCHRONOUS to guarantee completion before process.exit()
+            // Only uses stdout.write() which is sync - no async I/O allowed here
+            cleanup: () => {
+                // Log partial results context if available
+                const ctx = getPartialResultsContext();
+                if (ctx && ctx.completedAgents > 0) {
+                    const lines = formatPartialResultsMessage(ctx);
+                    for (const line of lines) {
+                        stdout.write(line + '\n');
+                    }
                 }
-                return sum;
-            }, 0);
-            const terminalContext = buildTerminalContext(resolvedOptions, gitContext, configSource, configPath, getPackageVersion());
-            terminalContext.executionTimeMs = executionTimeMs;
-            terminalContext.estimatedCostUsd = Math.max(0, estimatedCostUsd);
-            const processed = processLocalReportFindings(error.partialResults.completeFindings, error.partialResults.partialFindings, diff.files, config, 'local');
-            const incompleteStatus = 'incomplete';
-            const reportFn = deps.reportToTerminal ?? reportToTerminal;
-            await reportFn(processed.complete, processed.partial, terminalContext, config, diff.files, {
-                status: incompleteStatus,
-                suppressionSummary: processed.suppressionSummary,
+            },
+            showPartialResultsMessage: true,
+            // exitOnSignal: true (default) - process.exit() called on first signal
+            exitOnSignal: true,
+            logger: {
+                log: (msg) => isMachineReadableFormat ? stderr.write(msg + '\n') : stdout.write(msg + '\n'),
+                warn: (msg) => stderr.write(msg + '\n'),
+            },
+        });
+        // 12. Execute agent passes
+        let executeResult;
+        try {
+            // Set up partial results tracking (using runnableConfig which excludes skipped passes)
+            const totalAgents = runnableConfig.passes.reduce((sum, p) => (p.enabled ? sum + p.agents.length : sum), 0);
+            setPartialResultsContext({
+                totalAgents,
+                completedAgents: 0,
+                completedAgentNames: [],
+                currentAgent: undefined,
             });
-            // FR-021: Exit code 3 for incomplete review — partial results are NOT used for gating
+            const executeFn = deps.executeAllPasses ?? executeAllPasses;
+            executeResult = await executeFn(runnableConfig, agentContext, routerEnv, budgetCheck, {
+                configHash,
+                head: diff.headSha,
+            });
+        }
+        catch (error) {
+            clearPartialResultsContext();
+            const errorMsg = error instanceof Error ? error.message : String(error);
+            // FR-021: Extract partial results from FatalExecutionError when available.
+            // If agents completed before the failure, report their findings in degraded mode.
+            if (error instanceof FatalExecutionError && error.partialResults) {
+                stderr.write(c.yellow(`\n⚠ Incomplete review: ${errorMsg}\n`));
+                stderr.write(c.yellow(`  ${error.partialResults.completeFindings.length} findings from completed agents preserved\n`));
+                // Process partial results through the standard pipeline
+                const executionTimeMs = Date.now() - startTime;
+                const estimatedCostUsd = error.partialResults.allResults.reduce((sum, r) => {
+                    if ('metrics' in r && r.metrics?.estimatedCostUsd) {
+                        return sum + r.metrics.estimatedCostUsd;
+                    }
+                    return sum;
+                }, 0);
+                const terminalContext = buildTerminalContext(resolvedOptions, gitContext, configSource, configPath, getPackageVersion());
+                terminalContext.executionTimeMs = executionTimeMs;
+                terminalContext.estimatedCostUsd = Math.max(0, estimatedCostUsd);
+                const processed = processLocalReportFindings(error.partialResults.completeFindings, error.partialResults.partialFindings, diff.files, config, 'local');
+                const incompleteStatus = 'incomplete';
+                const reportFn = deps.reportToTerminal ?? reportToTerminal;
+                await reportFn(processed.complete, processed.partial, terminalContext, config, diff.files, {
+                    status: incompleteStatus,
+                    suppressionSummary: processed.suppressionSummary,
+                });
+                // FR-021: Exit code 3 for incomplete review — partial results are NOT used for gating
+                return {
+                    exitCode: exitCodeFromStatus(incompleteStatus),
+                    status: incompleteStatus,
+                    findingsCount: processed.complete.length,
+                    partialFindingsCount: processed.partial.length,
+                    error: errorMsg,
+                };
+            }
+            // Fatal failure with no preserved findings — hard crash, not a degraded run.
+            // Use config_error (exit 2) to distinguish from incomplete (exit 3, which has findings).
+            stderr.write(c.red(`\nExecution error: ${errorMsg}\n`));
             return {
-                exitCode: exitCodeFromStatus(incompleteStatus),
-                status: incompleteStatus,
-                findingsCount: processed.complete.length,
-                partialFindingsCount: processed.partial.length,
+                exitCode: exitCodeFromStatus('config_error'),
+                status: 'config_error',
+                findingsCount: 0,
+                partialFindingsCount: 0,
                 error: errorMsg,
             };
         }
-        // Fatal failure with no preserved findings — hard crash, not a degraded run.
-        // Use config_error (exit 2) to distinguish from incomplete (exit 3, which has findings).
-        stderr.write(c.red(`\nExecution error: ${errorMsg}\n`));
+        clearPartialResultsContext();
+        // 13. Calculate execution time and cost
+        const executionTimeMs = Date.now() - startTime;
+        const estimatedCostUsd = executeResult.allResults.reduce((sum, r) => {
+            if ('metrics' in r && r.metrics?.estimatedCostUsd) {
+                return sum + r.metrics.estimatedCostUsd;
+            }
+            return sum;
+        }, 0);
+        // 14. Build terminal context with execution info
+        const terminalContext = buildTerminalContext(resolvedOptions, gitContext, configSource, configPath, getPackageVersion());
+        terminalContext.executionTimeMs = executionTimeMs;
+        terminalContext.estimatedCostUsd = Math.max(0, estimatedCostUsd); // Clamp to non-negative (FR-REL-002)
+        // 15. Post-process findings (FR-018: CLI parity — sanitize → semantic → framework → diff-bound)
+        // FR-022: Local mode uses working tree config for suppressions (developer trusted)
+        const processed = processLocalReportFindings(executeResult.completeFindings, executeResult.partialFindings, diff.files, config, 'local');
+        const runStatus = determineRunStatus(processed.complete, config);
+        const exitCode = exitCodeFromStatus(runStatus);
+        // 16. Report findings to terminal
+        const reportFn = deps.reportToTerminal ?? reportToTerminal;
+        const reportResult = await reportFn(processed.complete, processed.partial, terminalContext, config, diff.files, {
+            status: runStatus,
+            suppressionSummary: processed.suppressionSummary,
+        });
         return {
-            exitCode: exitCodeFromStatus('config_error'),
-            status: 'config_error',
-            findingsCount: 0,
-            partialFindingsCount: 0,
-            error: errorMsg,
+            exitCode,
+            status: runStatus,
+            findingsCount: reportResult.findingsCount,
+            partialFindingsCount: reportResult.partialFindingsCount,
         };
-    }
-    clearPartialResultsContext();
-    // 13. Calculate execution time and cost
-    const executionTimeMs = Date.now() - startTime;
-    const estimatedCostUsd = executeResult.allResults.reduce((sum, r) => {
-        if ('metrics' in r && r.metrics?.estimatedCostUsd) {
-            return sum + r.metrics.estimatedCostUsd;
-        }
-        return sum;
-    }, 0);
-    // 14. Build terminal context with execution info
-    const terminalContext = buildTerminalContext(resolvedOptions, gitContext, configSource, configPath, getPackageVersion());
-    terminalContext.executionTimeMs = executionTimeMs;
-    terminalContext.estimatedCostUsd = Math.max(0, estimatedCostUsd); // Clamp to non-negative (FR-REL-002)
-    // 15. Post-process findings (FR-018: CLI parity — sanitize → semantic → framework → diff-bound)
-    // FR-022: Local mode uses working tree config for suppressions (developer trusted)
-    const processed = processLocalReportFindings(executeResult.completeFindings, executeResult.partialFindings, diff.files, config, 'local');
-    const runStatus = determineRunStatus(processed.complete, config);
-    const exitCode = exitCodeFromStatus(runStatus);
-    // 16. Report findings to terminal
-    const reportFn = deps.reportToTerminal ?? reportToTerminal;
-    const reportResult = await reportFn(processed.complete, processed.partial, terminalContext, config, diff.files, {
-        status: runStatus,
-        suppressionSummary: processed.suppressionSummary,
     });
-    return {
-        exitCode,
-        status: runStatus,
-        findingsCount: reportResult.findingsCount,
-        partialFindingsCount: reportResult.partialFindingsCount,
-    };
 }
 //# sourceMappingURL=local-review.js.map