agent-nuvira 1.14.0

package/dist/agents/agents/reviewer.js

@@ -0,0 +1,159 @@
+/**
+ * ReviewerAgent — Validates code changes produced by the WriterAgent.
+ *
+ * The reviewer checks for:
+ * - Syntax errors and type mismatches
+ * - Security vulnerabilities (SQL injection, XSS, etc.)
+ * - Code style and conventions
+ * - Correctness of the implementation against the task description
+ * - Missing edge cases and error handling
+ *
+ * Output is stored in the shared context bus as conversation messages.
+ */
+import { Agent } from '../agent.js';
+import { logger } from '../../utils/logger.js';
+const REVIEWER_SYSTEM_PROMPT = `You are a senior code reviewer. Review the following code changes for quality, correctness, and security.
+
+Focus on:
+1. **Correctness** — Does the code correctly implement the described task?
+2. **Security** — Any SQL injection, XSS, path traversal, or other vulnerabilities?
+3. **Error handling** — Are edge cases and invalid inputs handled?
+4. **Code quality** — Is the code clean, readable, and maintainable?
+5. **Type safety** — Are there any type mismatches or implicit any types?
+6. **Performance** — Any obvious performance issues?
+
+## Output Format
+
+ONLY list issues that actually exist. Do NOT describe absent issues.
+
+For each issue found, use this exact format:
+- CRITICAL: <description>
+  Location: <file/line>
+  Fix: <suggestion>
+
+- WARNING: <description>
+  Location: <file/line>
+  Fix: <suggestion>
+
+- SUGGESTION: <description>
+  Location: <file/line>
+
+If NO issues are found, respond ONLY with:
+✅ Review passed. No issues found.
+
+Do NOT mention potential issues that don't exist. Do NOT use the words "CRITICAL", "WARNING", or "SUGGESTION" unless you are actually flagging an issue.`;
+/** Maximum API retry attempts for transient LLM failures */
+const MAX_API_RETRIES = 2;
+/** Base delay for exponential backoff (doubles each retry: 1s, 2s) */
+const BASE_RETRY_DELAY_MS = 1000;
+/**
+ * ReviewerAgent — Validates code changes produced by WriterAgent.
+ */
+export class ReviewerAgent extends Agent {
+    name = 'Reviewer';
+    description = 'Validates code changes for correctness, security, and quality';
+    async execute(context, callLLM) {
+        let lastError;
+        if (context.fileChanges.length === 0) {
+            return {
+                success: true,
+                summary: 'No files to review',
+                details: 'The WriterAgent did not produce any file changes.',
+            };
+        }
+        for (let attempt = 0; attempt <= MAX_API_RETRIES; attempt++) {
+            try {
+                const prompt = this.buildPrompt(context);
+                const response = await callLLM(prompt, {
+                    temperature: 0.2,
+                    maxTokens: 4096,
+                });
+                // Log the review as a conversation message
+                context.conversations.push({
+                    from: 'Reviewer',
+                    to: 'Orchestrator',
+                    content: response,
+                    timestamp: Date.now(),
+                });
+                const hasCriticalIssues = this.hasCriticalIssues(response);
+                return {
+                    success: !hasCriticalIssues,
+                    summary: hasCriticalIssues
+                        ? 'Review found critical issues'
+                        : 'Review passed',
+                    details: response,
+                    error: hasCriticalIssues
+                        ? 'Critical issues found — see review details'
+                        : undefined,
+                };
+            }
+            catch (err) {
+                lastError = err instanceof Error ? err.message : String(err);
+                if (attempt < MAX_API_RETRIES) {
+                    const delayMs = BASE_RETRY_DELAY_MS * Math.pow(2, attempt);
+                    logger.warn(`Reviewer API error (attempt ${attempt + 1}/${MAX_API_RETRIES + 1}): ` +
+                        `${lastError.slice(0, 200)}. Retrying in ${delayMs}ms...`);
+                    await new Promise((resolve) => setTimeout(resolve, delayMs));
+                    continue;
+                }
+                logger.error(`Reviewer failed after ${MAX_API_RETRIES + 1} API attempts: ${lastError}`);
+                return {
+                    success: false,
+                    summary: 'Review failed',
+                    error: lastError,
+                };
+            }
+        }
+        // Unreachable
+        return {
+            success: false,
+            summary: 'Review failed',
+            error: lastError || 'Unknown error',
+        };
+    }
+    /**
+     * Build the review prompt from the task plan, relevant context, and changes.
+     */
+    buildPrompt(context) {
+        // Format the task description
+        const taskDescriptions = context.taskPlan
+            .filter((s) => s.agentType === 'writer')
+            .map((s) => `  - ${s.description}`)
+            .join('\n');
+        // Format the file changes as diffs
+        const diffs = context.fileChanges
+            .map((change) => {
+            const header = `--- a/${change.path}\n+++ b/${change.path}`;
+            if (change.originalContent && change.newContent) {
+                // Simple diff: show old and new
+                return `${header}\n@@ ... @@\n${change.originalContent}\n---\n${change.newContent}`;
+            }
+            if (change.status === 'created') {
+                return `${header}\n@@ -0,0 +1 @@\n+ (new file)\n${change.newContent}`;
+            }
+            return header;
+        })
+            .join('\n\n');
+        return `${REVIEWER_SYSTEM_PROMPT}\n\n## Task Description\n${taskDescriptions || context.goal}\n\n## Changes to Review\n${diffs || '(No changes provided)'}\n\n## Instructions\nReview the above changes. Identify any issues and provide feedback.`;
+    }
+    /**
+     * Check if the review response contains any critical issues.
+     * Looks for "CRITICAL:" (with colon) at the start of a bullet or line.
+     * This avoids false positives from LLMs that say things like
+     * "No critical issues found" (which lacks the colon prefix).
+     */
+    hasCriticalIssues(review) {
+        // Match "CRITICAL:" (with colon) on its own line, as a bullet, or in flow text.
+        // The colon prefix is critical — it's how the prompt tells the LLM to format issues.
+        // "No critical issues found" won't match; "CRITICAL: SQL injection" will.
+        const criticalPrefix = /(?:^|\n|[-*]\s*)CRITICAL\s*:/im;
+        // Also match the old patterns for backwards compatibility
+        const oldPatterns = [
+            /🔴/,
+            /\bBlocking\b/,
+            /\bSecurity\s*vulnerability\b/i,
+        ];
+        return criticalPrefix.test(review) || oldPatterns.some((p) => p.test(review));
+    }
+}
+//# sourceMappingURL=reviewer.js.map

package/dist/agents/agents/reviewer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"reviewer.js","sourceRoot":"","sources":["../../../src/agents/agents/reviewer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,KAAK,EAAuC,MAAM,aAAa,CAAC;AAEzE,OAAO,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAE/C,MAAM,sBAAsB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yJA6B0H,CAAC;AAE1J,4DAA4D;AAC5D,MAAM,eAAe,GAAG,CAAC,CAAC;AAE1B,sEAAsE;AACtE,MAAM,mBAAmB,GAAG,IAAI,CAAC;AAEjC;;GAEG;AACH,MAAM,OAAO,aAAc,SAAQ,KAAK;IAC7B,IAAI,GAAG,UAAU,CAAC;IAClB,WAAW,GAAG,+DAA+D,CAAC;IAEvF,KAAK,CAAC,OAAO,CAAC,OAAqB,EAAE,OAAkB;QACrD,IAAI,SAA6B,CAAC;QAElC,IAAI,OAAO,CAAC,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACrC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE,oBAAoB;gBAC7B,OAAO,EAAE,mDAAmD;aAC7D,CAAC;QACJ,CAAC;QAED,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,eAAe,EAAE,OAAO,EAAE,EAAE,CAAC;YAC5D,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;gBACzC,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,MAAM,EAAE;oBACrC,WAAW,EAAE,GAAG;oBAChB,SAAS,EAAE,IAAI;iBAChB,CAAC,CAAC;gBAEH,2CAA2C;gBAC3C,OAAO,CAAC,aAAa,CAAC,IAAI,CAAC;oBACzB,IAAI,EAAE,UAAU;oBAChB,EAAE,EAAE,cAAc;oBAClB,OAAO,EAAE,QAAQ;oBACjB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;iBACtB,CAAC,CAAC;gBAEH,MAAM,iBAAiB,GAAG,IAAI,CAAC,iBAAiB,CAAC,QAAQ,CAAC,CAAC;gBAE3D,OAAO;oBACL,OAAO,EAAE,CAAC,iBAAiB;oBAC3B,OAAO,EAAE,iBAAiB;wBACxB,CAAC,CAAC,8BAA8B;wBAChC,CAAC,CAAC,eAAe;oBACnB,OAAO,EAAE,QAAQ;oBACjB,KAAK,EAAE,iBAAiB;wBACtB,CAAC,CAAC,4CAA4C;wBAC9C,CAAC,CAAC,SAAS;iBACd,CAAC;YACJ,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,SAAS,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAE7D,IAAI,OAAO,GAAG,eAAe,EAAE,CAAC;oBAC9B,MAAM,OAAO,GAAG,mBAAmB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;oBAC3D,MAAM,CAAC,IAAI,CACT,+BAA+B,OAAO,GAAG,CAAC,IAAI,eAAe,GAAG,CAAC,KAAK;wBACtE,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,iBAAiB,OAAO,OAAO,CAC1D,CAAC;oBACF,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;oBAC7D,SAAS;gBACX,CAAC;gBAED,MAAM,CAAC,KAAK,CAAC,yBAAyB,eAAe,GAAG,CAAC,kBAAkB,SAAS,EAAE,CAAC,CAAC;gBACxF,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,OAAO,EAAE,eAAe;oBACxB,KAAK,EAAE,SAAS;iBACjB,CAAC;YACJ,CAAC;QACH,CAAC;QAED,cAAc;QACd,OAAO;YACL,OAAO,EAAE,KAAK;YACd,OAAO,EAAE,eAAe;YACxB,KAAK,EAAE,SAAS,IAAI,eAAe;SACpC,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,WAAW,CAAC,OAAqB;QACvC,8BAA8B;QAC9B,MAAM,gBAAgB,GAAG,OAAO,CAAC,QAAQ;aACtC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,QAAQ,CAAC;aACvC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,WAAW,EAAE,CAAC;aAClC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,mCAAmC;QACnC,MAAM,KAAK,GAAG,OAAO,CAAC,WAAW;aAC9B,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE;YACd,MAAM,MAAM,GAAG,SAAS,MAAM,CAAC,IAAI,WAAW,MAAM,CAAC,IAAI,EAAE,CAAC;YAC5D,IAAI,MAAM,CAAC,eAAe,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;gBAChD,gCAAgC;gBAChC,OAAO,GAAG,MAAM,gBAAgB,MAAM,CAAC,eAAe,UAAU,MAAM,CAAC,UAAU,EAAE,CAAC;YACtF,CAAC;YACD,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;gBAChC,OAAO,GAAG,MAAM,kCAAkC,MAAM,CAAC,UAAU,EAAE,CAAC;YACxE,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC,CAAC;aACD,IAAI,CAAC,MAAM,CAAC,CAAC;QAEhB,OAAO,GAAG,sBAAsB,4BAA4B,gBAAgB,IAAI,OAAO,CAAC,IAAI,6BAA6B,KAAK,IAAI,uBAAuB,0FAA0F,CAAC;IACtP,CAAC;IAED;;;;;OAKG;IACK,iBAAiB,CAAC,MAAc;QACtC,gFAAgF;QAChF,qFAAqF;QACrF,0EAA0E;QAC1E,MAAM,cAAc,GAAG,gCAAgC,CAAC;QAExD,0DAA0D;QAC1D,MAAM,WAAW,GAAG;YAClB,IAAI;YACJ,cAAc;YACd,+BAA+B;SAChC,CAAC;QAEF,OAAO,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC;IAChF,CAAC;CACF"}

package/dist/agents/agents/runner.d.ts

@@ -0,0 +1,73 @@
+/**
+ * RunnerAgent — Executes shell commands in the project directory and captures output.
+ *
+ * This is the agent that makes agent-nuvira capable of *running* the programs
+ * it creates. Without this, the system can write files but can never execute
+ * them or show the user what happened.
+ *
+ * Usage in task plans:
+ * ```json
+ * { "id": "step-03-run", "description": "Run: python hello.py", "agentType": "runner", "dependsOn": ["step-02-write"] }
+ * ```
+ *
+ * The command to run is determined by:
+ * 1. The task description — if it contains a command wrapped in backticks
+ *    (e.g., "Run `python hello.py`"), that command is extracted and executed.
+ * 2. The "Run:" prefix — if the description starts with "Run:", the rest is
+ *    treated as the command (e.g., "Run: python hello.py").
+ * 3. The LLM fallback — if no explicit command is found, the LLM is asked
+ *    what command to run based on the current context (files created, project type).
+ *
+ * Output is stored in context metadata as `runResult` and returned in the summary.
+ */
+import { Agent, type AgentContext, type AgentResult } from '../agent.js';
+import type { LLMCallFn } from '../agent.js';
+/**
+ * Result of running a command, stored in context.metadata.runResult.
+ */
+export interface RunResult {
+    /** Whether the command exited with code 0 */
+    success: boolean;
+    /** The exact command that was executed */
+    command: string;
+    /** Process exit code */
+    exitCode: number;
+    /** Standard output */
+    stdout: string;
+    /** Standard error */
+    stderr: string;
+    /** Duration in milliseconds */
+    duration: number;
+    /** Error message if execSync threw */
+    error?: string;
+}
+/**
+ * RunnerAgent — Executes shell commands and captures output.
+ */
+export declare class RunnerAgent extends Agent {
+    readonly name = "Runner";
+    readonly description = "Executes shell commands and captures output";
+    execute(context: AgentContext, callLLM: LLMCallFn): Promise<AgentResult>;
+    /**
+     * Determine the command to run.
+     *
+     * Priority order:
+     * 1. Parse from task description (backtick-wrapped command or "Run:" prefix)
+     * 2. Ask the LLM what command to run based on the files that were created
+     */
+    private determineCommand;
+    /**
+     * Execute a command inside a Docker sandbox container.
+     * Falls back to host execution if Docker is not available.
+     */
+    private executeWithDocker;
+    /**
+     * Execute a command directly on the host machine.
+     */
+    private executeOnHost;
+    /**
+     * Fallback: ask the LLM what command to run based on the project context.
+     */
+    private askLLMForCommand;
+}
+//# sourceMappingURL=runner.d.ts.map

package/dist/agents/agents/runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../../src/agents/agents/runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAKH,OAAO,EAAE,KAAK,EAAE,KAAK,YAAY,EAAE,KAAK,WAAW,EAAE,MAAM,aAAa,CAAC;AACzE,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAY7C;;GAEG;AACH,MAAM,WAAW,SAAS;IACxB,6CAA6C;IAC7C,OAAO,EAAE,OAAO,CAAC;IACjB,0CAA0C;IAC1C,OAAO,EAAE,MAAM,CAAC;IAChB,wBAAwB;IACxB,QAAQ,EAAE,MAAM,CAAC;IACjB,sBAAsB;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,qBAAqB;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,+BAA+B;IAC/B,QAAQ,EAAE,MAAM,CAAC;IACjB,sCAAsC;IACtC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED;;GAEG;AACH,qBAAa,WAAY,SAAQ,KAAK;IACpC,QAAQ,CAAC,IAAI,YAAY;IACzB,QAAQ,CAAC,WAAW,iDAAiD;IAE/D,OAAO,CAAC,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,CAAC;IAiC9E;;;;;;OAMG;YACW,gBAAgB;IAyB9B;;;OAGG;YACW,iBAAiB;IAmG/B;;OAEG;YACW,aAAa;IAiF3B;;OAEG;YACW,gBAAgB;CA8C/B"}

package/dist/agents/agents/runner.js

@@ -0,0 +1,299 @@
+/**
+ * RunnerAgent — Executes shell commands in the project directory and captures output.
+ *
+ * This is the agent that makes agent-nuvira capable of *running* the programs
+ * it creates. Without this, the system can write files but can never execute
+ * them or show the user what happened.
+ *
+ * Usage in task plans:
+ * ```json
+ * { "id": "step-03-run", "description": "Run: python hello.py", "agentType": "runner", "dependsOn": ["step-02-write"] }
+ * ```
+ *
+ * The command to run is determined by:
+ * 1. The task description — if it contains a command wrapped in backticks
+ *    (e.g., "Run `python hello.py`"), that command is extracted and executed.
+ * 2. The "Run:" prefix — if the description starts with "Run:", the rest is
+ *    treated as the command (e.g., "Run: python hello.py").
+ * 3. The LLM fallback — if no explicit command is found, the LLM is asked
+ *    what command to run based on the current context (files created, project type).
+ *
+ * Output is stored in context metadata as `runResult` and returned in the summary.
+ */
+import { execSync } from 'node:child_process';
+import { platform } from 'node:os';
+import { Agent } from '../agent.js';
+import { logger } from '../../utils/logger.js';
+import { SandboxManager } from '../../sandbox/manager.js';
+import { detectProjectImage } from '../../sandbox/images.js';
+import { getSandboxConfig } from '../../sandbox/types.js';
+/** Maximum stdout/stderr length to store in context metadata */
+const MAX_OUTPUT_LENGTH = 10_000;
+/** Timeout per command in milliseconds (default: 2 minutes) */
+const DEFAULT_TIMEOUT_MS = 120_000;
+/**
+ * RunnerAgent — Executes shell commands and captures output.
+ */
+export class RunnerAgent extends Agent {
+    name = 'Runner';
+    description = 'Executes shell commands and captures output';
+    async execute(context, callLLM) {
+        try {
+            // 1. Determine which command to run
+            const command = await this.determineCommand(context, callLLM);
+            if (!command) {
+                return {
+                    success: false,
+                    summary: 'No command to run',
+                    error: 'Could not determine which command to execute from the task description or context.',
+                };
+            }
+            // Check if we should run inside a Docker sandbox
+            const useDocker = context.metadata.useDockerSandbox === true ||
+                getSandboxConfig().enabled === true;
+            if (useDocker) {
+                return await this.executeWithDocker(context, command);
+            }
+            // 2. Execute the command on the host via shared method
+            return await this.executeOnHost(context, command);
+        }
+        catch (err) {
+            const msg = err instanceof Error ? err.message : String(err);
+            return {
+                success: false,
+                summary: 'Runner failed',
+                error: msg,
+            };
+        }
+    }
+    /**
+     * Determine the command to run.
+     *
+     * Priority order:
+     * 1. Parse from task description (backtick-wrapped command or "Run:" prefix)
+     * 2. Ask the LLM what command to run based on the files that were created
+     */
+    async determineCommand(context, callLLM) {
+        // Find the current 'runner' task in the plan
+        const runnerTask = context.taskPlan.find((s) => s.agentType === 'runner' && s.status === 'running');
+        const description = runnerTask?.description || context.goal;
+        // Strategy 1: Extract command from backticks in the description
+        // e.g., "Run `python hello.py` and verify output"
+        const backtickMatch = description.match(/`([^`]+)`/);
+        if (backtickMatch) {
+            return backtickMatch[1].trim();
+        }
+        // Strategy 2: Extract from "Run:" prefix
+        // e.g., "Run: python hello.py"
+        const runPrefixMatch = description.match(/^Run:\s*(.+)/i);
+        if (runPrefixMatch) {
+            return runPrefixMatch[1].trim();
+        }
+        // Strategy 3: Ask the LLM what command to run
+        return await this.askLLMForCommand(context, callLLM);
+    }
+    /**
+     * Execute a command inside a Docker sandbox container.
+     * Falls back to host execution if Docker is not available.
+     */
+    async executeWithDocker(context, command) {
+        const sandboxManager = new SandboxManager();
+        let containerId = '';
+        try {
+            // Check Docker availability
+            const dockerAvailable = await sandboxManager.isDockerAvailable();
+            if (!dockerAvailable) {
+                // Fall back to host execution
+                return this.executeOnHost(context, command);
+            }
+            // Detect the right image for the project
+            const image = detectProjectImage(context.workingDirectory);
+            // Allow timeout override via context.metadata.runnerTimeout
+            const timeoutMs = (typeof context.metadata.runnerTimeout === 'number')
+                ? context.metadata.runnerTimeout
+                : DEFAULT_TIMEOUT_MS;
+            // Create a Docker container (use default /workspace as workdir)
+            containerId = await sandboxManager.createContainer(image.image, {
+                memoryLimit: '512m',
+                cpuLimit: 0.5,
+                timeoutMs,
+                networkAccess: false,
+            });
+            // Copy project files to the container's workspace
+            await sandboxManager.copyProjectToContainer(containerId, context.workingDirectory);
+            // Run the command inside the container
+            if (context.metadata.verboseLogging) {
+                logger.info(`     Running (Docker): ${command}`);
+            }
+            const result = await sandboxManager.runCommand(containerId, command, timeoutMs);
+            // Build run result from sandbox result
+            const runResult = {
+                success: result.success,
+                command,
+                exitCode: result.exitCode,
+                stdout: result.stdout.slice(0, MAX_OUTPUT_LENGTH),
+                stderr: result.stderr.slice(0, MAX_OUTPUT_LENGTH),
+                duration: result.durationMs,
+                error: result.error,
+            };
+            context.metadata['runResult'] = runResult;
+            // Build summary
+            const lines = [];
+            lines.push(`Command: ${command} (Docker)`);
+            lines.push(`Exit code: ${result.exitCode}`);
+            lines.push(`Duration: ${result.durationMs}ms`);
+            if (result.stdout) {
+                const truncated = result.stdout.length > 500;
+                lines.push(`stdout:${truncated ? ' (first 500 chars)' : ''}`);
+                lines.push(result.stdout.slice(0, 500));
+                if (truncated)
+                    lines.push(`... (${result.stdout.length - 500} more chars)`);
+            }
+            if (result.stderr && result.exitCode !== 0) {
+                const truncated = result.stderr.length > 500;
+                lines.push(`stderr:${truncated ? ' (first 500 chars)' : ''}`);
+                lines.push(result.stderr.slice(0, 500));
+                if (truncated)
+                    lines.push(`... (${result.stderr.length - 500} more chars)`);
+            }
+            // Clean up
+            await sandboxManager.destroyContainer(containerId).catch(() => { });
+            return {
+                success: result.exitCode === 0,
+                summary: result.exitCode === 0
+                    ? `✅ Command succeeded (Docker): ${command}`
+                    : `❌ Command failed (exit ${result.exitCode}): ${command}`,
+                details: lines.join('\n'),
+                error: result.error || undefined,
+            };
+        }
+        catch (err) {
+            if (containerId) {
+                await sandboxManager.destroyContainer(containerId).catch(() => { });
+            }
+            const msg = err instanceof Error ? err.message : String(err);
+            return {
+                success: false,
+                summary: 'Docker sandbox execution failed',
+                error: msg,
+            };
+        }
+    }
+    /**
+     * Execute a command directly on the host machine.
+     */
+    async executeOnHost(context, command) {
+        if (context.metadata.verboseLogging) {
+            logger.info(`     Running: ${command}`);
+        }
+        const timeoutMs = (typeof context.metadata.runnerTimeout === 'number')
+            ? context.metadata.runnerTimeout
+            : DEFAULT_TIMEOUT_MS;
+        const startTime = Date.now();
+        let exitCode = 0;
+        let stdout = '';
+        let stderr = '';
+        let execError;
+        try {
+            const output = execSync(command, {
+                cwd: context.workingDirectory,
+                timeout: timeoutMs,
+                stdio: 'pipe',
+                encoding: 'utf-8',
+                shell: platform() === 'win32' ? (process.env.COMSPEC || 'cmd.exe') : '/bin/sh',
+                maxBuffer: 1024 * 1024,
+            });
+            stdout = output.trim();
+        }
+        catch (err) {
+            const error = err;
+            exitCode = error.status ?? 1;
+            stdout = (typeof error.stdout === 'string' ? error.stdout : String(error.stdout || '')).trim();
+            stderr = (typeof error.stderr === 'string' ? error.stderr : String(error.stderr || '')).trim();
+            execError = error.message;
+        }
+        const duration = Date.now() - startTime;
+        const runResult = {
+            success: exitCode === 0,
+            command,
+            exitCode,
+            stdout: stdout.slice(0, MAX_OUTPUT_LENGTH),
+            stderr: stderr.slice(0, MAX_OUTPUT_LENGTH),
+            duration,
+            error: execError,
+        };
+        context.metadata['runResult'] = runResult;
+        const lines = [];
+        lines.push(`Command: ${command}`);
+        lines.push(`Exit code: ${exitCode}`);
+        lines.push(`Duration: ${duration}ms`);
+        if (stdout) {
+            const truncated = stdout.length > 500;
+            lines.push(`stdout:${truncated ? ' (first 500 chars)' : ''}`);
+            lines.push(stdout.slice(0, 500));
+            if (truncated)
+                lines.push(`... (${stdout.length - 500} more chars)`);
+        }
+        if (stderr && exitCode !== 0) {
+            const truncated = stderr.length > 500;
+            lines.push(`stderr:${truncated ? ' (first 500 chars)' : ''}`);
+            lines.push(stderr.slice(0, 500));
+            if (truncated)
+                lines.push(`... (${stderr.length - 500} more chars)`);
+        }
+        return {
+            success: exitCode === 0,
+            summary: exitCode === 0
+                ? `✅ Command succeeded: ${command}`
+                : `❌ Command failed (exit ${exitCode}): ${command}`,
+            details: lines.join('\n'),
+            error: execError && exitCode !== 0 ? execError : undefined,
+        };
+    }
+    /**
+     * Fallback: ask the LLM what command to run based on the project context.
+     */
+    async askLLMForCommand(context, callLLM) {
+        const fileList = context.fileChanges
+            .map((c) => `  - ${c.path} (${c.status})`)
+            .join('\n');
+        const artifactList = context.artifacts
+            .slice(0, 5)
+            .map((a) => `  - ${a.path}`)
+            .join('\n');
+        const prompt = [
+            'You are a build-and-run expert. Given the context below, what single shell command should be executed',
+            'to verify the work that was done? Return ONLY the command, no explanation, no markdown.',
+            '',
+            `Goal: ${context.goal}`,
+            '',
+            'Files changed:',
+            fileList || '  (no files changed)',
+            '',
+            'Relevant project files:',
+            artifactList || '  (empty project)',
+            '',
+            'Return ONLY the command to run. Example: "python hello.py" or "node index.js" or "npm test" or "go run main.go".',
+            'Rules:',
+            '- Return a single line command only',
+            '- No backticks, no explanation, no $ prefix',
+            '- Use absolute or working-directory-relative paths',
+            '- If unsure, suggest the most appropriate verification command',
+        ].join('\n');
+        try {
+            const response = await callLLM(prompt, {
+                temperature: 0.1,
+                maxTokens: 256,
+            });
+            const command = response.trim().replace(/^```(?:bash|sh)?\s*|\s*```$/g, '').trim();
+            if (command && !command.includes('\n') && command.length < 500) {
+                return command;
+            }
+        }
+        catch {
+            // LLM fallback failed — return null
+        }
+        return null;
+    }
+}
+//# sourceMappingURL=runner.js.map

package/dist/agents/agents/runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../../src/agents/agents/runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAEnC,OAAO,EAAE,KAAK,EAAuC,MAAM,aAAa,CAAC;AAEzE,OAAO,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAC1D,OAAO,EAAE,kBAAkB,EAAE,MAAM,yBAAyB,CAAC;AAC7D,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAE1D,gEAAgE;AAChE,MAAM,iBAAiB,GAAG,MAAM,CAAC;AAEjC,+DAA+D;AAC/D,MAAM,kBAAkB,GAAG,OAAO,CAAC;AAsBnC;;GAEG;AACH,MAAM,OAAO,WAAY,SAAQ,KAAK;IAC3B,IAAI,GAAG,QAAQ,CAAC;IAChB,WAAW,GAAG,6CAA6C,CAAC;IAErE,KAAK,CAAC,OAAO,CAAC,OAAqB,EAAE,OAAkB;QACrD,IAAI,CAAC;YACH,oCAAoC;YACpC,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9D,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,OAAO,EAAE,mBAAmB;oBAC5B,KAAK,EAAE,oFAAoF;iBAC5F,CAAC;YACJ,CAAC;YAED,iDAAiD;YACjD,MAAM,SAAS,GAAG,OAAO,CAAC,QAAQ,CAAC,gBAAgB,KAAK,IAAI;gBAC1D,gBAAgB,EAAE,CAAC,OAAO,KAAK,IAAI,CAAC;YAEtC,IAAI,SAAS,EAAE,CAAC;gBACd,OAAO,MAAM,IAAI,CAAC,iBAAiB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YACxD,CAAC;YAED,uDAAuD;YACvD,OAAO,MAAM,IAAI,CAAC,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAEpD,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7D,OAAO;gBACL,OAAO,EAAE,KAAK;gBACd,OAAO,EAAE,eAAe;gBACxB,KAAK,EAAE,GAAG;aACX,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;;;;;OAMG;IACK,KAAK,CAAC,gBAAgB,CAAC,OAAqB,EAAE,OAAkB;QACtE,6CAA6C;QAC7C,MAAM,UAAU,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,CACtC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,CAC1D,CAAC;QACF,MAAM,WAAW,GAAG,UAAU,EAAE,WAAW,IAAI,OAAO,CAAC,IAAI,CAAC;QAE5D,gEAAgE;QAChE,kDAAkD;QAClD,MAAM,aAAa,GAAG,WAAW,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACrD,IAAI,aAAa,EAAE,CAAC;YAClB,OAAO,aAAa,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAED,yCAAyC;QACzC,+BAA+B;QAC/B,MAAM,cAAc,GAAG,WAAW,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC;QAC1D,IAAI,cAAc,EAAE,CAAC;YACnB,OAAO,cAAc,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAClC,CAAC;QAED,8CAA8C;QAC9C,OAAO,MAAM,IAAI,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IACvD,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,iBAAiB,CAAC,OAAqB,EAAE,OAAe;QACpE,MAAM,cAAc,GAAG,IAAI,cAAc,EAAE,CAAC;QAC5C,IAAI,WAAW,GAAG,EAAE,CAAC;QAErB,IAAI,CAAC;YACH,4BAA4B;YAC5B,MAAM,eAAe,GAAG,MAAM,cAAc,CAAC,iBAAiB,EAAE,CAAC;YACjE,IAAI,CAAC,eAAe,EAAE,CAAC;gBACrB,8BAA8B;gBAC9B,OAAO,IAAI,CAAC,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,CAAC;YAED,yCAAyC;YACzC,MAAM,KAAK,GAAG,kBAAkB,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;YAE3D,4DAA4D;YAC5D,MAAM,SAAS,GAAG,CAAC,OAAO,OAAO,CAAC,QAAQ,CAAC,aAAa,KAAK,QAAQ,CAAC;gBACpE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,aAAa;gBAChC,CAAC,CAAC,kBAAkB,CAAC;YAEvB,gEAAgE;YAChE,WAAW,GAAG,MAAM,cAAc,CAAC,eAAe,CAChD,KAAK,CAAC,KAAK,EACX;gBACE,WAAW,EAAE,MAAM;gBACnB,QAAQ,EAAE,GAAG;gBACb,SAAS;gBACT,aAAa,EAAE,KAAK;aACrB,CACF,CAAC;YAEF,kDAAkD;YAClD,MAAM,cAAc,CAAC,sBAAsB,CAAC,WAAW,EAAE,OAAO,CAAC,gBAAgB,CAAC,CAAC;YAEnF,uCAAuC;YACvC,IAAI,OAAO,CAAC,QAAQ,CAAC,cAAc,EAAE,CAAC;gBACpC,MAAM,CAAC,IAAI,CAAC,0BAA0B,OAAO,EAAE,CAAC,CAAC;YACnD,CAAC;YAED,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,UAAU,CAAC,WAAW,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;YAEhF,uCAAuC;YACvC,MAAM,SAAS,GAAc;gBAC3B,OAAO,EAAE,MAAM,CAAC,OAAO;gBACvB,OAAO;gBACP,QAAQ,EAAE,MAAM,CAAC,QAAQ;gBACzB,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,iBAAiB,CAAC;gBACjD,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,iBAAiB,CAAC;gBACjD,QAAQ,EAAE,MAAM,CAAC,UAAU;gBAC3B,KAAK,EAAE,MAAM,CAAC,KAAK;aACpB,CAAC;YAEF,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC,GAAG,SAAS,CAAC;YAE1C,gBAAgB;YAChB,MAAM,KAAK,GAAa,EAAE,CAAC;YAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,OAAO,WAAW,CAAC,CAAC;YAC3C,KAAK,CAAC,IAAI,CAAC,cAAc,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;YAC5C,KAAK,CAAC,IAAI,CAAC,aAAa,MAAM,CAAC,UAAU,IAAI,CAAC,CAAC;YAE/C,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;gBAClB,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC;gBAC7C,KAAK,CAAC,IAAI,CAAC,UAAU,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBAC9D,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;gBACxC,IAAI,SAAS;oBAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,GAAG,cAAc,CAAC,CAAC;YAC9E,CAAC;YAED,IAAI,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;gBAC3C,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC;gBAC7C,KAAK,CAAC,IAAI,CAAC,UAAU,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBAC9D,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;gBACxC,IAAI,SAAS;oBAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,GAAG,cAAc,CAAC,CAAC;YAC9E,CAAC;YAED,WAAW;YACX,MAAM,cAAc,CAAC,gBAAgB,CAAC,WAAW,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;YAEnE,OAAO;gBACL,OAAO,EAAE,MAAM,CAAC,QAAQ,KAAK,CAAC;gBAC9B,OAAO,EAAE,MAAM,CAAC,QAAQ,KAAK,CAAC;oBAC5B,CAAC,CAAC,iCAAiC,OAAO,EAAE;oBAC5C,CAAC,CAAC,0BAA0B,MAAM,CAAC,QAAQ,MAAM,OAAO,EAAE;gBAC5D,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;gBACzB,KAAK,EAAE,MAAM,CAAC,KAAK,IAAI,SAAS;aACjC,CAAC;QACJ,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAI,WAAW,EAAE,CAAC;gBAChB,MAAM,cAAc,CAAC,gBAAgB,CAAC,WAAW,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7D,OAAO;gBACL,OAAO,EAAE,KAAK;gBACd,OAAO,EAAE,iCAAiC;gBAC1C,KAAK,EAAE,GAAG;aACX,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,aAAa,CAAC,OAAqB,EAAE,OAAe;QAChE,IAAI,OAAO,CAAC,QAAQ,CAAC,cAAc,EAAE,CAAC;YACpC,MAAM,CAAC,IAAI,CAAC,iBAAiB,OAAO,EAAE,CAAC,CAAC;QAC1C,CAAC;QAED,MAAM,SAAS,GAAG,CAAC,OAAO,OAAO,CAAC,QAAQ,CAAC,aAAa,KAAK,QAAQ,CAAC;YACpE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,aAAa;YAChC,CAAC,CAAC,kBAAkB,CAAC;QAEvB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,QAAQ,GAAG,CAAC,CAAC;QACjB,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,IAAI,SAA6B,CAAC;QAElC,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,QAAQ,CAAC,OAAO,EAAE;gBAC/B,GAAG,EAAE,OAAO,CAAC,gBAAgB;gBAC7B,OAAO,EAAE,SAAS;gBAClB,KAAK,EAAE,MAAM;gBACb,QAAQ,EAAE,OAAO;gBACjB,KAAK,EAAE,QAAQ,EAAE,KAAK,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,IAAI,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS;gBAC9E,SAAS,EAAE,IAAI,GAAG,IAAI;aACvB,CAAC,CAAC;YACH,MAAM,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;QACzB,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,KAAK,GAAG,GAKb,CAAC;YACF,QAAQ,GAAG,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC;YAC7B,MAAM,GAAG,CAAC,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;YAC/F,MAAM,GAAG,CAAC,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;YAC/F,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC;QAC5B,CAAC;QAED,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QAExC,MAAM,SAAS,GAAc;YAC3B,OAAO,EAAE,QAAQ,KAAK,CAAC;YACvB,OAAO;YACP,QAAQ;YACR,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,iBAAiB,CAAC;YAC1C,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,iBAAiB,CAAC;YAC1C,QAAQ;YACR,KAAK,EAAE,SAAS;SACjB,CAAC;QAEF,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC,GAAG,SAAS,CAAC;QAE1C,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,OAAO,EAAE,CAAC,CAAC;QAClC,KAAK,CAAC,IAAI,CAAC,cAAc,QAAQ,EAAE,CAAC,CAAC;QACrC,KAAK,CAAC,IAAI,CAAC,aAAa,QAAQ,IAAI,CAAC,CAAC;QAEtC,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC;YACtC,KAAK,CAAC,IAAI,CAAC,UAAU,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAC9D,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;YACjC,IAAI,SAAS;gBAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,MAAM,CAAC,MAAM,GAAG,GAAG,cAAc,CAAC,CAAC;QACvE,CAAC;QAED,IAAI,MAAM,IAAI,QAAQ,KAAK,CAAC,EAAE,CAAC;YAC7B,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC;YACtC,KAAK,CAAC,IAAI,CAAC,UAAU,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAC9D,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;YACjC,IAAI,SAAS;gBAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,MAAM,CAAC,MAAM,GAAG,GAAG,cAAc,CAAC,CAAC;QACvE,CAAC;QAED,OAAO;YACL,OAAO,EAAE,QAAQ,KAAK,CAAC;YACvB,OAAO,EAAE,QAAQ,KAAK,CAAC;gBACrB,CAAC,CAAC,wBAAwB,OAAO,EAAE;gBACnC,CAAC,CAAC,0BAA0B,QAAQ,MAAM,OAAO,EAAE;YACrD,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;YACzB,KAAK,EAAE,SAAS,IAAI,QAAQ,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;SAC3D,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,gBAAgB,CAAC,OAAqB,EAAE,OAAkB;QACtE,MAAM,QAAQ,GAAG,OAAO,CAAC,WAAW;aACjC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,MAAM,GAAG,CAAC;aACzC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,MAAM,YAAY,GAAG,OAAO,CAAC,SAAS;aACnC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;aACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;aAC3B,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,MAAM,MAAM,GAAG;YACb,uGAAuG;YACvG,yFAAyF;YACzF,EAAE;YACF,SAAS,OAAO,CAAC,IAAI,EAAE;YACvB,EAAE;YACF,gBAAgB;YAChB,QAAQ,IAAI,sBAAsB;YAClC,EAAE;YACF,yBAAyB;YACzB,YAAY,IAAI,mBAAmB;YACnC,EAAE;YACF,kHAAkH;YAClH,QAAQ;YACR,qCAAqC;YACrC,6CAA6C;YAC7C,oDAAoD;YACpD,gEAAgE;SACjE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEb,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,MAAM,EAAE;gBACrC,WAAW,EAAE,GAAG;gBAChB,SAAS,EAAE,GAAG;aACf,CAAC,CAAC;YAEH,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,8BAA8B,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;YACnF,IAAI,OAAO,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,OAAO,CAAC,MAAM,GAAG,GAAG,EAAE,CAAC;gBAC/D,OAAO,OAAO,CAAC;YACjB,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,oCAAoC;QACtC,CAAC;QAED,OAAO,IAAI,CAAC;IACd,CAAC;CACF"}

package/dist/agents/agents/security-agent.d.ts

@@ -0,0 +1,33 @@
+/**
+ * SecurityAgent — Detects security issues in agent prompts and generated code.
+ *
+ * Capabilities:
+ * - Scan LLM prompts for injection/jailbreak attempts before sending
+ * - Scan generated code changes for PII, dangerous patterns, vulnerabilities
+ * - Validate file paths for directory traversal attempts
+ * - Integrated with the orchestrator pipeline to block unsafe content
+ *
+ * Usage in task plans:
+ * ```json
+ * { "id": "step-security", "description": "Scan code for security issues", "agentType": "security", "dependsOn": ["step-write"] }
+ * ```
+ */
+import { Agent, type AgentContext, type AgentResult } from '../agent.js';
+import type { LLMCallFn } from '../agent.js';
+/**
+ * SecurityAgent — Scans prompts and generated content for security issues.
+ */
+export declare class SecurityAgent extends Agent {
+    readonly name = "Security";
+    readonly description = "Scans for PII, prompt injection, and dangerous code patterns";
+    execute(context: AgentContext, callLLM: LLMCallFn): Promise<AgentResult>;
+    /**
+     * Parse which scan types to run based on the task description.
+     */
+    private parseScanTypes;
+    /**
+     * Check if a file path is safe (no directory traversal, no protected paths).
+     */
+    private checkPathSafety;
+}
+//# sourceMappingURL=security-agent.d.ts.map

package/dist/agents/agents/security-agent.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"security-agent.d.ts","sourceRoot":"","sources":["../../../src/agents/agents/security-agent.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAIH,OAAO,EAAE,KAAK,EAAE,KAAK,YAAY,EAAE,KAAK,WAAW,EAAE,MAAM,aAAa,CAAC;AACzE,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAgC7C;;GAEG;AACH,qBAAa,aAAc,SAAQ,KAAK;IACtC,QAAQ,CAAC,IAAI,cAAc;IAC3B,QAAQ,CAAC,WAAW,kEAAkE;IAEhF,OAAO,CAAC,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,CAAC;IAwG9E;;OAEG;IACH,OAAO,CAAC,cAAc;IA4BtB;;OAEG;IACH,OAAO,CAAC,eAAe;CA8BxB"}