@build-astron-co/nimbus 0.4.2 → 0.4.3

package/src/__tests__/subagents.test.ts

@@ -1,176 +0,0 @@
-/**
- * Subagent Tests
- *
- * Validates subagent creation, configuration, tool restrictions, and
- * the @agent mention parser.
- */
-
-import { describe, test, expect } from 'vitest';
-import {
-  createSubagent,
-  parseAgentMention,
-  type SubagentType,
-  exploreConfig,
-  infraConfig,
-  securityConfig,
-  costConfig,
-  generalConfig,
-  Subagent,
-} from '../agent/subagents/index';
-
-// ===========================================================================
-// createSubagent
-// ===========================================================================
-
-describe('createSubagent', () => {
-  test('createSubagent("explore") returns Subagent with correct config', () => {
-    const agent = createSubagent('explore');
-    expect(agent).toBeInstanceOf(Subagent);
-    expect(agent.config.name).toBe('explore');
-    expect(agent.config.model).toBe('anthropic/claude-haiku-4-5');
-    expect(agent.config.maxTurns).toBe(15);
-  });
-
-  test('createSubagent("infra") returns Subagent with correct tools', () => {
-    const agent = createSubagent('infra');
-    expect(agent.config.name).toBe('infra');
-    const toolNames = agent.config.tools.map(t => t.name);
-    expect(toolNames).toContain('read_file');
-    expect(toolNames).toContain('cloud_discover');
-    expect(toolNames).toContain('cost_estimate');
-    expect(toolNames).toContain('drift_detect');
-  });
-
-  test('createSubagent("security") returns Subagent with security prompt', () => {
-    const agent = createSubagent('security');
-    expect(agent.config.name).toBe('security');
-    expect(agent.config.systemPrompt).toContain('security');
-    expect(agent.config.systemPrompt).toContain('CRITICAL');
-  });
-
-  test('createSubagent("cost") uses haiku model', () => {
-    const agent = createSubagent('cost');
-    expect(agent.config.model).toContain('haiku');
-  });
-
-  test('createSubagent("general") has bash and webfetch tools', () => {
-    const agent = createSubagent('general');
-    const toolNames = agent.config.tools.map(t => t.name);
-    expect(toolNames).toContain('bash');
-    expect(toolNames).toContain('webfetch');
-  });
-
-  test('all subagent types are valid and produce Subagent instances', () => {
-    const types: SubagentType[] = ['explore', 'infra', 'security', 'cost', 'general'];
-    for (const type of types) {
-      const agent = createSubagent(type);
-      expect(agent).toBeInstanceOf(Subagent);
-      expect(agent.config.name).toBe(type);
-    }
-  });
-});
-
-// ===========================================================================
-// parseAgentMention
-// ===========================================================================
-
-describe('parseAgentMention', () => {
-  test('parses "@explore find TODOs" correctly', () => {
-    const result = parseAgentMention('@explore find TODOs');
-    expect(result).not.toBeNull();
-    expect(result!.agent).toBe('explore');
-    expect(result!.prompt).toBe('find TODOs');
-  });
-
-  test('parses "@infra check EKS" correctly', () => {
-    const result = parseAgentMention('@infra check EKS');
-    expect(result).not.toBeNull();
-    expect(result!.agent).toBe('infra');
-    expect(result!.prompt).toBe('check EKS');
-  });
-
-  test('parses "@security scan for secrets" correctly', () => {
-    const result = parseAgentMention('@security scan for secrets');
-    expect(result).not.toBeNull();
-    expect(result!.agent).toBe('security');
-    expect(result!.prompt).toBe('scan for secrets');
-  });
-
-  test('parses "@cost estimate monthly spend" correctly', () => {
-    const result = parseAgentMention('@cost estimate monthly spend');
-    expect(result).not.toBeNull();
-    expect(result!.agent).toBe('cost');
-  });
-
-  test('parses "@general research topic" correctly', () => {
-    const result = parseAgentMention('@general research topic');
-    expect(result).not.toBeNull();
-    expect(result!.agent).toBe('general');
-  });
-
-  test('returns null for normal messages', () => {
-    expect(parseAgentMention('normal message without @mention')).toBeNull();
-    expect(parseAgentMention('fix the bug')).toBeNull();
-  });
-
-  test('returns null for empty string', () => {
-    expect(parseAgentMention('')).toBeNull();
-  });
-
-  test('returns null for unknown @agent prefix', () => {
-    expect(parseAgentMention('@unknown do something')).toBeNull();
-    expect(parseAgentMention('@deploy run it')).toBeNull();
-  });
-
-  test('returns null when @agent has no prompt', () => {
-    // The regex requires at least one character after the agent name
-    expect(parseAgentMention('@explore')).toBeNull();
-  });
-});
-
-// ===========================================================================
-// Subagent Tool Restrictions
-// ===========================================================================
-
-describe('Subagent tool restrictions', () => {
-  test('all subagent configs exclude the "task" tool (no nesting)', () => {
-    const configs = [exploreConfig, infraConfig, securityConfig, costConfig, generalConfig];
-    for (const config of configs) {
-      const hasTask = config.tools.some(t => t.name === 'task');
-      expect(hasTask).toBe(false);
-    }
-  });
-
-  test('explore subagent only has read-only tools', () => {
-    const toolNames = exploreConfig.tools.map(t => t.name);
-    expect(toolNames).toEqual(['read_file', 'glob', 'grep', 'list_dir']);
-    // None of these are destructive
-    for (const tool of exploreConfig.tools) {
-      expect(tool.isDestructive).toBeFalsy();
-    }
-  });
-
-  test('security subagent only has read-only tools', () => {
-    const toolNames = securityConfig.tools.map(t => t.name);
-    expect(toolNames).toEqual(['read_file', 'glob', 'grep', 'list_dir']);
-    for (const tool of securityConfig.tools) {
-      expect(tool.isDestructive).toBeFalsy();
-    }
-  });
-
-  test('infra subagent has cloud discovery tools', () => {
-    const toolNames = infraConfig.tools.map(t => t.name);
-    expect(toolNames).toContain('cloud_discover');
-    expect(toolNames).toContain('cost_estimate');
-    expect(toolNames).toContain('drift_detect');
-  });
-
-  test('cost subagent has cost_estimate and cloud_discover', () => {
-    const toolNames = costConfig.tools.map(t => t.name);
-    expect(toolNames).toContain('cost_estimate');
-    expect(toolNames).toContain('cloud_discover');
-    // But should not have destructive tools
-    expect(toolNames).not.toContain('terraform');
-    expect(toolNames).not.toContain('kubectl');
-  });
-});

package/src/__tests__/system-prompt.test.ts

@@ -1,248 +0,0 @@
-/**
- * System Prompt Tests
- *
- * Validates that buildSystemPrompt assembles the correct prompt sections
- * based on mode, tools, NIMBUS.md, subagent state, and environment context.
- */
-
-import { describe, test, expect, beforeEach, afterEach } from 'vitest';
-import * as fs from 'node:fs';
-import * as path from 'node:path';
-import * as os from 'node:os';
-import { buildSystemPrompt, loadNimbusMd, extractForbiddenRules } from '../agent/system-prompt';
-import type { ToolDefinition } from '../tools/schemas/types';
-import { z } from 'zod';
-
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-
-/** Create a minimal ToolDefinition for prompt tests. */
-function makeTool(name: string): ToolDefinition {
-  return {
-    name,
-    description: `Description of ${name}`,
-    inputSchema: z.object({}),
-    execute: async () => ({ output: 'ok', isError: false }),
-    permissionTier: 'auto_allow',
-    category: 'standard',
-  };
-}
-
-// ---------------------------------------------------------------------------
-// Temp directory for NIMBUS.md tests
-// ---------------------------------------------------------------------------
-
-let tmpDir: string;
-
-beforeEach(() => {
-  tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'nimbus-prompt-test-'));
-});
-
-afterEach(() => {
-  fs.rmSync(tmpDir, { recursive: true, force: true });
-});
-
-// ===========================================================================
-// buildSystemPrompt
-// ===========================================================================
-
-describe('buildSystemPrompt', () => {
-  test('includes base identity', () => {
-    const prompt = buildSystemPrompt({ mode: 'build', tools: [] });
-    expect(prompt).toContain('You are Nimbus');
-    // C2: new DevOps-operator-first framing
-    expect(prompt).toContain('autonomous DevOps operator');
-    expect(prompt).toContain('RUN commands and query live state');
-  });
-
-  test('includes mode-specific instructions for "plan"', () => {
-    const prompt = buildSystemPrompt({ mode: 'plan', tools: [] });
-    expect(prompt).toContain('Mode: PLAN');
-    expect(prompt).toContain('NOT allowed');
-  });
-
-  test('includes mode-specific instructions for "build"', () => {
-    const prompt = buildSystemPrompt({ mode: 'build', tools: [] });
-    expect(prompt).toContain('Mode: BUILD');
-    expect(prompt).toContain('Edit and create files');
-  });
-
-  test('includes mode-specific instructions for "deploy"', () => {
-    const prompt = buildSystemPrompt({ mode: 'deploy', tools: [] });
-    expect(prompt).toContain('Mode: DEPLOY');
-    expect(prompt).toContain('terraform apply');
-  });
-
-  test('includes tool-use guidelines', () => {
-    const prompt = buildSystemPrompt({ mode: 'build', tools: [] });
-    expect(prompt).toContain('Tool-Use Guidelines');
-    expect(prompt).toContain('read_file');
-  });
-
-  test('includes tools summary with correct count', () => {
-    const tools = [makeTool('alpha'), makeTool('beta'), makeTool('gamma')];
-    const prompt = buildSystemPrompt({ mode: 'build', tools });
-    expect(prompt).toContain('Available Tools (3)');
-    expect(prompt).toContain('**alpha**');
-    expect(prompt).toContain('**beta**');
-    expect(prompt).toContain('**gamma**');
-  });
-
-  test('includes NIMBUS.md content when provided', () => {
-    const prompt = buildSystemPrompt({
-      mode: 'build',
-      tools: [],
-      nimbusInstructions: 'Always use TypeScript strict mode.',
-    });
-    expect(prompt).toContain('Project Instructions (NIMBUS.md)');
-    expect(prompt).toContain('Always use TypeScript strict mode.');
-  });
-
-  test('includes environment context', () => {
-    const prompt = buildSystemPrompt({
-      mode: 'build',
-      tools: [],
-      cwd: tmpDir,
-    });
-    expect(prompt).toContain('# Environment');
-    expect(prompt).toContain(`Working directory: ${tmpDir}`);
-    expect(prompt).toContain(`Platform: ${process.platform}`);
-  });
-
-  test('includes subagent instructions when activeSubagent set', () => {
-    const prompt = buildSystemPrompt({
-      mode: 'build',
-      tools: [],
-      activeSubagent: 'explore',
-    });
-    expect(prompt).toContain('Subagent Mode: explore');
-    expect(prompt).toContain('Do NOT spawn further subagents');
-  });
-
-  test('does not include subagent section when activeSubagent is not set', () => {
-    const prompt = buildSystemPrompt({ mode: 'build', tools: [] });
-    expect(prompt).not.toContain('Subagent Mode');
-  });
-
-  test('includes date in environment context', () => {
-    const prompt = buildSystemPrompt({ mode: 'build', tools: [] });
-    // Should contain a date-like string YYYY-MM-DD
-    expect(prompt).toMatch(/Date: \d{4}-\d{2}-\d{2}/);
-  });
-});
-
-// ===========================================================================
-// loadNimbusMd
-// ===========================================================================
-
-describe('loadNimbusMd', () => {
-  test('returns null when no file exists', () => {
-    const result = loadNimbusMd(tmpDir);
-    expect(result).toBeNull();
-  });
-
-  test('loads NIMBUS.md from cwd', () => {
-    const content = '# Custom Instructions\nDo the thing.';
-    fs.writeFileSync(path.join(tmpDir, 'NIMBUS.md'), content, 'utf-8');
-    const result = loadNimbusMd(tmpDir);
-    expect(result).toBe(content);
-  });
-
-  test('loads NIMBUS.md from .nimbus subdirectory', () => {
-    const nimbusDir = path.join(tmpDir, '.nimbus');
-    fs.mkdirSync(nimbusDir, { recursive: true });
-    const content = 'Sub-dir instructions';
-    fs.writeFileSync(path.join(nimbusDir, 'NIMBUS.md'), content, 'utf-8');
-    const result = loadNimbusMd(tmpDir);
-    expect(result).toBe(content);
-  });
-
-  test('prefers cwd NIMBUS.md over .nimbus subdirectory', () => {
-    // Write both
-    fs.writeFileSync(path.join(tmpDir, 'NIMBUS.md'), 'root-level', 'utf-8');
-    const nimbusDir = path.join(tmpDir, '.nimbus');
-    fs.mkdirSync(nimbusDir, { recursive: true });
-    fs.writeFileSync(path.join(nimbusDir, 'NIMBUS.md'), 'sub-level', 'utf-8');
-
-    const result = loadNimbusMd(tmpDir);
-    expect(result).toBe('root-level');
-  });
-});
-
-// ===========================================================================
-// G14: extractForbiddenRules
-// ===========================================================================
-
-describe('extractForbiddenRules (G14)', () => {
-  test('extracts bullet items from ## Forbidden section', () => {
-    const content = `
-## Safety Rules
-
-- Do not break prod
-
-## Forbidden
-
-- Never destroy the production database
-- Never delete the S3 bucket
-- Never run rm -rf /
-
-## Custom Instructions
-
-Some other stuff
-`;
-    const rules = extractForbiddenRules(content);
-    expect(rules).toHaveLength(3);
-    expect(rules[0]).toBe('Never destroy the production database');
-    expect(rules[1]).toBe('Never delete the S3 bucket');
-    expect(rules[2]).toBe('Never run rm -rf /');
-  });
-
-  test('returns empty array when no Forbidden section', () => {
-    const content = '## Safety Rules\n\n- Be careful\n';
-    expect(extractForbiddenRules(content)).toEqual([]);
-  });
-
-  test('ignores HTML comment lines', () => {
-    const content = `
-## Forbidden
-
-<!-- Example: - Never destroy the database -->
-- Never touch prod
-`;
-    const rules = extractForbiddenRules(content);
-    expect(rules).toHaveLength(1);
-    expect(rules[0]).toBe('Never touch prod');
-  });
-
-  test('G14: prompt includes HARD CONSTRAINTS block when Forbidden section has entries', () => {
-    const nimbusContent = `
-## Forbidden
-
-- Never destroy the production database
-- Never run terraform destroy in prod
-`;
-    const prompt = buildSystemPrompt({
-      mode: 'build',
-      tools: [],
-      nimbusInstructions: nimbusContent,
-    });
-    expect(prompt).toContain('HARD CONSTRAINTS');
-    expect(prompt).toContain('Never destroy the production database');
-    expect(prompt).toContain('STRICTLY FORBIDDEN');
-  });
-
-  test('G14: prompt does not include HARD CONSTRAINTS when Forbidden section is empty/comments', () => {
-    const nimbusContent = `
-## Forbidden
-
-<!-- List operations Nimbus must never perform in this project -->
-`;
-    const prompt = buildSystemPrompt({
-      mode: 'build',
-      tools: [],
-      nimbusInstructions: nimbusContent,
-    });
-    expect(prompt).not.toContain('HARD CONSTRAINTS');
-  });
-});