claude-comms 1.0.45 → 1.0.46

package/.agents/tools/workflow/lib/prompts.ts

@@ -0,0 +1,548 @@
+/**
+ * Prompt Building Module
+ *
+ * Handles template loading and prompt assembly for all job types.
+ * Extracted from runner.ts for modularity.
+ */
+
+import { readFileSync } from "fs";
+import { dirname, join } from "path";
+import { fileURLToPath } from "url";
+
+// Types needed for prompt building
+export interface Assignment {
+  _id: string;
+  _creationTime: number;
+  namespaceId: string;
+  northStar: string;
+  status: "pending" | "active" | "blocked" | "complete";
+  blockedReason?: string;
+  independent: boolean;
+  priority: number;
+  artifacts: string;
+  decisions: string;
+  headGroupId?: string;
+  createdAt: number;
+  updatedAt: number;
+}
+
+export interface JobGroup {
+  _id: string;
+  _creationTime: number;
+  assignmentId: string;
+  nextGroupId?: string;
+  status: "pending" | "running" | "complete" | "failed";
+  aggregatedResult?: string;
+  createdAt: number;
+}
+
+export interface Job {
+  _id: string;
+  _creationTime: number;
+  groupId: string;
+  jobType: string;
+  harness: "claude" | "codex" | "gemini";
+  context?: string;
+  status: "pending" | "running" | "complete" | "failed";
+  result?: string;
+  prompt?: string;
+  startedAt?: number;
+  completedAt?: number;
+  createdAt: number;
+}
+
+export interface ChatMessage {
+  _id: string;
+  threadId: string;
+  role: "user" | "assistant" | "pm";
+  content: string;
+  createdAt: number;
+}
+
+export interface ChatJobContext {
+  threadId: string;
+  namespaceId: string;
+  mode: "jam" | "cook" | "guardian";
+  // For differential prompting
+  effectivePromptMode: "jam" | "cook";
+  lastPromptMode?: "jam" | "cook";
+  messages: ChatMessage[];
+  latestUserMessage: string;
+  claudeSessionId?: string;
+  // Guardian mode context
+  assignmentId?: string;
+  isGuardianEvaluation?: boolean; // True when PO is evaluating PM response
+}
+
+// Prompt types for differential prompting
+export type PromptType = "full" | "mode_activation" | "minimal" | "guardian_eval";
+
+// Resolve templates directory relative to this module
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const TEMPLATES_DIR = join(__dirname, "..", "templates");
+
+/**
+ * Load a template file by job type
+ */
+export function loadTemplate(jobType: string): string {
+  const templatePath = join(TEMPLATES_DIR, `${jobType}.md`);
+  try {
+    return readFileSync(templatePath, "utf-8");
+  } catch {
+    console.error(`Template not found: ${jobType}.md`);
+    return "Execute the task as described.\n\n{{CONTEXT}}";
+  }
+}
+
+// Accumulated job result for PM jobs
+export interface AccumulatedJobResult {
+  jobType: string;
+  harness: string; // Kept for internal tracking, not shown to PM
+  result: string;
+  groupId?: string;
+  groupIndex?: number;
+}
+
+/**
+ * Format accumulated results for PM prompt
+ * Uses jobType with A/B/C suffixes for multiple jobs of same type
+ * Does NOT expose harness names - PM sees only jobType labels
+ */
+function formatAccumulatedResults(
+  accumulatedResults: AccumulatedJobResult[]
+): string {
+  if (!accumulatedResults || accumulatedResults.length === 0) {
+    return "(no previous results)";
+  }
+
+  // Group by jobType to determine if we need A/B/C suffixes
+  const byJobType = new Map<string, AccumulatedJobResult[]>();
+  for (const r of accumulatedResults) {
+    const list = byJobType.get(r.jobType) || [];
+    list.push(r);
+    byJobType.set(r.jobType, list);
+  }
+
+  // Build sections with A/B/C suffixes for duplicates
+  const sections: string[] = [];
+  const suffixes = ["A", "B", "C", "D", "E", "F", "G", "H"];
+
+  for (const [jobType, results] of byJobType) {
+    if (results.length === 1) {
+      // Single job of this type - no suffix
+      sections.push(`## ${jobType}\n\n${results[0].result}`);
+    } else {
+      // Multiple jobs of same type - add A/B/C suffixes
+      for (let i = 0; i < results.length; i++) {
+        const suffix = suffixes[i] || `${i + 1}`;
+        sections.push(`## ${jobType} ${suffix}\n\n${results[i].result}`);
+      }
+    }
+  }
+
+  return sections.join("\n\n---\n\n");
+}
+
+interface GroupedResults {
+  groupId: string;
+  groupIndex: number;
+  results: AccumulatedJobResult[];
+}
+
+function groupAccumulatedResults(
+  accumulatedResults: AccumulatedJobResult[]
+): GroupedResults[] {
+  if (!accumulatedResults || accumulatedResults.length === 0) {
+    return [];
+  }
+
+  const groups = new Map<string, GroupedResults>();
+  let fallbackIndex = 0;
+
+  for (const result of accumulatedResults) {
+    const groupId = result.groupId || `unknown-${fallbackIndex}`;
+    const groupIndex =
+      typeof result.groupIndex === "number" ? result.groupIndex : fallbackIndex;
+
+    if (!groups.has(groupId)) {
+      groups.set(groupId, {
+        groupId,
+        groupIndex,
+        results: [],
+      });
+    }
+
+    groups.get(groupId)!.results.push(result);
+    fallbackIndex += 1;
+  }
+
+  return Array.from(groups.values()).sort((a, b) => a.groupIndex - b.groupIndex);
+}
+
+function isReviewJobType(jobType: string): boolean {
+  return jobType === "review" || jobType.endsWith("review");
+}
+
+function includesJobType(results: AccumulatedJobResult[], jobType: string): boolean {
+  return results.some((r) => r.jobType === jobType);
+}
+
+function inferPrimaryJobType(results: AccumulatedJobResult[]): string {
+  if (includesJobType(results, "implement")) return "implement";
+  if (includesJobType(results, "plan")) return "plan";
+  if (includesJobType(results, "document")) return "document";
+  if (includesJobType(results, "uat")) return "uat";
+  return results[0]?.jobType || "unknown";
+}
+
+function buildPmModules(
+  accumulatedResults: AccumulatedJobResult[],
+  r1GroupResults?: AccumulatedJobResult[]
+): string {
+  const grouped = groupAccumulatedResults(accumulatedResults);
+  if (grouped.length === 0) {
+    return "No prior job results found. Ask clarifying questions if needed and decide the first actionable job.";
+  }
+
+  const latestGroup = grouped[grouped.length - 1];
+  const latestHasReview = latestGroup.results.some((r) => isReviewJobType(r.jobType));
+  const latestHasImplement = latestGroup.results.some((r) => r.jobType === "implement");
+  const latestHasPlan = latestGroup.results.some((r) => r.jobType === "plan");
+
+  if (latestHasReview) {
+    let reviewGroupIndex = -1;
+    for (let i = grouped.length - 1; i >= 0; i--) {
+      if (grouped[i].results.some((r) => isReviewJobType(r.jobType))) {
+        reviewGroupIndex = i;
+        break;
+      }
+    }
+
+    const hasR1Override = r1GroupResults && r1GroupResults.length > 0;
+    const priorGroup =
+      reviewGroupIndex > 0 ? grouped[reviewGroupIndex - 1] : undefined;
+    const p1JobType = hasR1Override
+      ? inferPrimaryJobType(r1GroupResults!)
+      : priorGroup
+        ? inferPrimaryJobType(priorGroup.results)
+        : "unknown";
+    const priorResults = hasR1Override
+      ? formatAccumulatedResults(r1GroupResults!)
+      : priorGroup
+        ? formatAccumulatedResults(priorGroup.results)
+        : "(no prior group found)";
+
+    return `### Post-Review Decision\n\n#### R-1 Context (Group Before Review)\n${priorResults}\n\n#### Decision Logic (P-1 = ${p1JobType})\n1. **If fundamental decisions are required** (cannot be inferred from mental-model + north star with high confidence):\n   - Ask clarifying questions\n   - Block the assignment until resolved\n2. **If high-severity issues have a clear optimal solution and reviewers concur**:\n   - Append a new **${p1JobType}** job to address/refine\n   - Include the issues raised and rationale in your PM response\n3. **If only medium/low/no issues and reviewers (and UAT, if present) approve**:\n   - Filter issues for alignment with mental model and real product value\n   - If **P-1 = plan**: update the plan doc yourself, then append **implement** (or complete if planning-only)\n   - If **P-1 = implement**: append **implement** to address items, or append **document** if no further changes are warranted\n\nAlways include the issues raised and your decision rationale.\n`;
+  }
+
+  if (latestHasImplement) {
+    return `### Post-Implement Decision\n\n- Append a **review** job to assess the implementation quality and alignment.\n- If the implementation impacts UX, include **uat** in the same job group.\n- Provide clear context for reviewers/UAT (what changed, success criteria, files/flows).\n`;
+  }
+
+  if (latestHasPlan) {
+    return `### Post-Plan Decision\n\n- If the plan/spec represents **non-trivial change** (5+ files, backend + frontend, foundational schema, or core system building blocks), append a **review** job.\n- If the plan is **trivial** or already reviewed and approved, append **implement** to execute the full plan.\n - Note: the implement job is powerful and can candle multiple WP and sequence accordingly. Assign suffiecient work to the implement job so it can deliver a full vertical slice (ready for review).\n- Ensure the plan doc path is recorded in artifacts.\n`;
+  }
+
+  return "No matching PM module found. Decide next steps based on the latest results and north star alignment.";
+}
+
+/**
+ * Build prompt for assignment-based jobs
+ * Each job has its own jobType and context
+ */
+export function buildPrompt(
+  group: JobGroup,
+  assignment: Assignment,
+  job: Job,
+  accumulatedResults: AccumulatedJobResult[],
+  previousNonPmGroupResults: AccumulatedJobResult[],
+  r1GroupResults: AccumulatedJobResult[]
+): string {
+  const template = loadTemplate(job.jobType);
+
+  // Format accumulated results for PM/retrospect jobs
+  const previousResults =
+    job.jobType === "pm" ? accumulatedResults : previousNonPmGroupResults;
+  const resultText = formatAccumulatedResults(previousResults);
+  const pmModules =
+    job.jobType === "pm" ? buildPmModules(accumulatedResults, r1GroupResults) : "";
+
+  return template
+    .replace(/\{\{NORTH_STAR\}\}/g, assignment.northStar)
+    .replace(/\{\{ARTIFACTS\}\}/g, assignment.artifacts || "(none)")
+    .replace(/\{\{DECISIONS\}\}/g, assignment.decisions || "(none)")
+    .replace(/\{\{CONTEXT\}\}/g, job.context || "(no specific context)")
+    .replace(/\{\{PREVIOUS_RESULT\}\}/g, resultText)
+    .replace(/\{\{ASSIGNMENT_ID\}\}/g, assignment._id)
+    .replace(/\{\{GROUP_ID\}\}/g, group._id)
+    .replace(/\{\{CURRENT_JOB_ID\}\}/g, job._id)
+    .replace(/\{\{HARNESS\}\}/g, job.harness)
+    .replace(/\{\{PM_MODULES\}\}/g, pmModules);
+}
+
+/**
+ * Determine which prompt type to use based on context
+ */
+export function determinePromptType(chatContext: ChatJobContext): PromptType {
+  const isNewSession = !chatContext.claudeSessionId;
+  const isGuardianEval = chatContext.isGuardianEvaluation === true;
+
+  // Guardian evaluation always gets its special prompt
+  if (isGuardianEval) {
+    return "guardian_eval";
+  }
+
+  // New session (no claudeSessionId) - send full prompt
+  if (isNewSession) {
+    return "full";
+  }
+
+  // Check if mode changed since last prompt
+  const modeChanged = chatContext.lastPromptMode !== chatContext.effectivePromptMode;
+
+  if (modeChanged) {
+    return "mode_activation";
+  }
+
+  // Same mode, resuming session - minimal prompt
+  return "minimal";
+}
+
+/**
+ * Build prompt for chat jobs with differential prompting
+ * - Full prompt for new sessions
+ * - Mode activation snippet when mode changes
+ * - Minimal prompt for same-mode resume
+ * - Guardian eval injection for PM report evaluation
+ */
+export function buildChatPrompt(chatContext: ChatJobContext, namespace: string): string {
+  const promptType = determinePromptType(chatContext);
+  const template = loadTemplate("product-owner");
+
+  // Variable replacements used across all prompt types
+  const replaceVars = (text: string): string => {
+    return text
+      .replace(/\{\{THREAD_ID\}\}/g, chatContext.threadId)
+      .replace(/\{\{NAMESPACE\}\}/g, namespace)
+      .replace(/\{\{MODE\}\}/g, chatContext.effectivePromptMode)
+      .replace(/\{\{LATEST_MESSAGE\}\}/g, chatContext.latestUserMessage)
+      .replace(/\{\{ASSIGNMENT_ID\}\}/g, chatContext.assignmentId || "(no assignment linked)");
+  };
+
+  switch (promptType) {
+    case "full":
+      return buildFullPrompt(template, chatContext, replaceVars);
+
+    case "mode_activation":
+      return buildModeActivationPrompt(chatContext, replaceVars);
+
+    case "minimal":
+      return buildMinimalPrompt(chatContext, replaceVars);
+
+    case "guardian_eval":
+      return buildGuardianEvalPrompt(template, chatContext, replaceVars);
+  }
+}
+
+/**
+ * Build full prompt for new sessions
+ */
+function buildFullPrompt(
+  template: string,
+  chatContext: ChatJobContext,
+  replaceVars: (text: string) => string
+): string {
+  let processed = template;
+
+  // Remove GUARDIAN_MODE blocks (not applicable for full prompt)
+  const guardianModeRegex = /\{\{#if GUARDIAN_MODE\}\}[\s\S]*?\{\{\/if\}\}/g;
+  processed = processed.replace(guardianModeRegex, "");
+
+  // Handle {{#if COOK_MODE}}...{{else}}...{{/if}}
+  const cookModeRegex = /\{\{#if COOK_MODE\}\}([\s\S]*?)\{\{else\}\}([\s\S]*?)\{\{\/if\}\}/g;
+  processed = processed.replace(cookModeRegex, (_match, cookContent, jamContent) => {
+    return chatContext.effectivePromptMode === "cook" ? cookContent : jamContent;
+  });
+
+  // Handle simple {{#if COOK_MODE}}...{{/if}} without else
+  const simpleCookModeRegex = /\{\{#if COOK_MODE\}\}([\s\S]*?)\{\{\/if\}\}/g;
+  processed = processed.replace(simpleCookModeRegex, (_match, content) => {
+    return chatContext.effectivePromptMode === "cook" ? content : "";
+  });
+
+  // Handle {{#if NEW_SESSION}}...{{/if}} - include for new sessions
+  const newSessionRegex = /\{\{#if NEW_SESSION\}\}([\s\S]*?)\{\{\/if\}\}/g;
+  processed = processed.replace(newSessionRegex, (_match, content) => content);
+
+  return replaceVars(processed);
+}
+
+/**
+ * Build mode activation prompt when mode changes during session
+ */
+function buildModeActivationPrompt(
+  chatContext: ChatJobContext,
+  replaceVars: (text: string) => string
+): string {
+  const mode = chatContext.effectivePromptMode.toUpperCase();
+
+  let modeContent: string;
+
+  if (chatContext.effectivePromptMode === "cook") {
+    modeContent = `## COOK MODE ACTIVATED
+
+You now have **FULL AUTONOMY** to take action. Adopt COOK mode behavior until explicitly changed.
+
+### Your Powers in Cook Mode
+
+- CREATE new assignments via CLI
+- INSERT jobs into the workflow queue
+- Make product decisions and execute them
+
+When the user wants work to be done:
+1. **Confirm** your understanding of requirements
+2. **Create** an assignment with a **verbose north star** (include user perspective + success criteria)
+3. **Insert** an initial job to begin work (usually \`plan\` type)
+4. **Immediately update** \`docs/project/spec/mental-model.md\` with new insights from the conversation
+5. **Inform** the user what you've initiated
+
+### CLI Commands Available
+
+\`\`\`bash
+# Create a new assignment (auto-linked to this thread)
+npx tsx .agents/tools/workflow/cli.ts create "<north-star-description>" --priority <N>
+
+# Insert job(s) - jobs in the same array run in parallel
+npx tsx .agents/tools/workflow/cli.ts insert-job <assignmentId> \\
+  --jobs '[{"jobType":"plan","context":"<context>"}]'
+
+# Append to existing chain (use --append to link after current tail)
+npx tsx .agents/tools/workflow/cli.ts insert-job <assignmentId> --append \\
+  --jobs '[{"jobType":"implement","context":"Build auth"},{"jobType":"uat","context":"Test login"}]'
+
+# View assignments and queue
+npx tsx .agents/tools/workflow/cli.ts assignments
+npx tsx .agents/tools/workflow/cli.ts queue
+
+# Delete assignment
+npx tsx .agents/tools/workflow/cli.ts delete-assignment <assignmentId>
+\`\`\`
+
+### Job Types You Can Create
+
+| Type | Use When |
+|------|----------|
+| \`plan\` | Need a spec doc and work-package breakdown |
+| \`implement\` | Clear requirements ready for implementation |
+| \`review\` | Engineering quality review of plan/spec or implementation |
+| \`uat\` | Need user-perspective testing |
+| \`document\` | Update docs and finalize assignment |`;
+  } else {
+    modeContent = `## JAM MODE ACTIVATED
+
+You are now in **READ-ONLY** ideation mode. Adopt JAM mode behavior until explicitly changed.
+
+### Your Role in Jam Mode
+
+- You CANNOT create assignments or jobs
+- You CAN help spec out ideas
+- You CAN ask clarifying questions
+- You CAN suggest approaches and trade-offs
+- You CAN help refine requirements
+- You CAN explore the codebase and existing work
+
+Help the user think through their ideas:
+- Ask probing questions to clarify requirements
+- Identify potential challenges and edge cases
+- Suggest technical approaches
+- Help prioritize features
+- Draft acceptance criteria
+- Explore trade-offs between options
+
+### When to Suggest Cook Mode
+
+If the user says things like "Let's do it", "Make it happen", "Start working on this" - suggest they switch to **Cook mode** to take action.`;
+  }
+
+  const prompt = `${modeContent}
+
+---
+
+## Current Message
+
+**User says:**
+${chatContext.latestUserMessage}`;
+
+  return replaceVars(prompt);
+}
+
+/**
+ * Build minimal prompt for same-mode resume
+ */
+function buildMinimalPrompt(
+  chatContext: ChatJobContext,
+  replaceVars: (text: string) => string
+): string {
+  const prompt = `## Current Message
+
+**User says:**
+${chatContext.latestUserMessage}`;
+
+  return replaceVars(prompt);
+}
+
+/**
+ * Build guardian evaluation prompt for PM report evaluation
+ */
+function buildGuardianEvalPrompt(
+  template: string,
+  chatContext: ChatJobContext,
+  replaceVars: (text: string) => string
+): string {
+  // Extract the GUARDIAN_MODE block from template
+  const guardianModeRegex = /\{\{#if GUARDIAN_MODE\}\}([\s\S]*?)\{\{\/if\}\}/;
+  const match = template.match(guardianModeRegex);
+
+  if (match && match[1]) {
+    return replaceVars(match[1].trim());
+  }
+
+  // Fallback if template doesn't have guardian block
+  return replaceVars(`## GUARDIAN MODE - ALIGNMENT EVALUATION
+
+You are monitoring assignment alignment. A PM has reported on work progress.
+
+**Assignment ID:** ${chatContext.assignmentId || "(unknown)"}
+
+### PM Progress Report
+${chatContext.latestUserMessage}
+
+### Alignment Response
+
+Respond with **ONE** of:
+
+**🟢** - Trajectory aligned with intent. Just the emoji, nothing else.
+
+**🟠** - Uncertain. Include 2-3 sentence rationale explaining the concern.
+
+**🔴** - Misaligned. Include rationale and block the assignment.`);
+}
+
+/**
+ * Parse chat context from job context JSON field
+ */
+export function parseChatContext(contextStr: string): ChatJobContext | null {
+  try {
+    return JSON.parse(contextStr) as ChatJobContext;
+  } catch {
+    console.error("Failed to parse chat context:", contextStr);
+    return null;
+  }
+}
+
+/**
+ * Check if a job is a chat job based on its type
+ */
+export function isChatJob(job: Job): boolean {
+  return job.jobType === "chat" || job.jobType === "product-owner";
+}

package/.agents/tools/workflow/lib/streams.test.ts

@@ -0,0 +1,140 @@
+/**
+ * Tests for stream handlers module
+ *
+ * Run with: npx tsx --test lib/streams.test.ts
+ */
+
+import { describe, it } from "node:test";
+import assert from "node:assert";
+import {
+  ClaudeStreamHandler,
+  CodexStreamHandler,
+  GeminiStreamHandler,
+  buildCommand,
+} from "./streams.js";
+
+// ============================================================================
+// ClaudeStreamHandler tests
+// ============================================================================
+
+describe("ClaudeStreamHandler", () => {
+  it("extracts text and captures session_id on success", () => {
+    const handler = new ClaudeStreamHandler();
+
+    // Simulate event stream
+    handler.onEvent({
+      type: "assistant",
+      message: { content: [{ type: "text", text: "Hello" }] },
+    });
+    handler.onEvent({
+      type: "result",
+      subtype: "success",
+      result: "Final result",
+      session_id: "session_123",
+    });
+
+    assert.strictEqual(handler.getResult(), "Final result");
+    assert.strictEqual(handler.isComplete(), true);
+    assert.strictEqual(handler.getSessionId(), "session_123");
+  });
+
+  it("does not mark complete on failure", () => {
+    const handler = new ClaudeStreamHandler();
+
+    handler.onEvent({ type: "result", subtype: "error" });
+
+    assert.strictEqual(handler.isComplete(), false);
+  });
+});
+
+// ============================================================================
+// CodexStreamHandler tests
+// ============================================================================
+
+describe("CodexStreamHandler", () => {
+  it("extracts agent_message text and marks complete on turn.completed", () => {
+    const handler = new CodexStreamHandler();
+
+    handler.onEvent({
+      type: "item.completed",
+      item: { type: "agent_message", text: "First message" },
+    });
+    handler.onEvent({
+      type: "item.completed",
+      item: { type: "agent_message", text: "Second message" },
+    });
+    handler.onEvent({ type: "turn.completed" });
+
+    assert.ok(handler.getResult().includes("First message"));
+    assert.ok(handler.getResult().includes("Second message"));
+    assert.strictEqual(handler.isComplete(), true);
+  });
+
+  it("ignores non-agent_message items", () => {
+    const handler = new CodexStreamHandler();
+
+    handler.onEvent({
+      type: "item.completed",
+      item: { type: "reasoning", text: "Thinking..." },
+    });
+
+    assert.strictEqual(handler.getResult(), "");
+  });
+});
+
+// ============================================================================
+// GeminiStreamHandler tests
+// ============================================================================
+
+describe("GeminiStreamHandler", () => {
+  it("accumulates assistant content and marks complete on result", () => {
+    const handler = new GeminiStreamHandler();
+
+    handler.onEvent({ type: "message", role: "assistant", content: "Hello " });
+    handler.onEvent({ type: "message", role: "assistant", content: "world" });
+    handler.onEvent({ type: "result" });
+
+    assert.strictEqual(handler.getResult(), "Hello world");
+    assert.strictEqual(handler.isComplete(), true);
+  });
+
+  it("ignores non-assistant messages", () => {
+    const handler = new GeminiStreamHandler();
+
+    handler.onEvent({ type: "message", role: "user", content: "User input" });
+    handler.onEvent({ type: "tool_use", tool_name: "shell" });
+
+    assert.strictEqual(handler.getResult(), "");
+  });
+});
+
+// ============================================================================
+// buildCommand tests
+// ============================================================================
+
+describe("buildCommand", () => {
+  it("builds claude command with optional session resume", () => {
+    const basic = buildCommand("claude", "test");
+    assert.strictEqual(basic.cmd, "claude");
+    assert.ok(basic.args.includes("--output-format"));
+    assert.ok(!basic.args.includes("--resume"));
+
+    const withSession = buildCommand("claude", "test", { sessionId: "s123" });
+    assert.ok(withSession.args.includes("--resume"));
+    assert.ok(withSession.args.includes("s123"));
+  });
+
+  it("builds codex and gemini commands", () => {
+    const codex = buildCommand("codex", "test");
+    assert.strictEqual(codex.cmd, "codex");
+    assert.ok(codex.args.includes("--json"));
+
+    const gemini = buildCommand("gemini", "test");
+    assert.strictEqual(gemini.cmd, "gemini");
+    assert.ok(gemini.args.includes("stream-json"));
+  });
+
+  it("throws for unknown harness", () => {
+    assert.throws(() => buildCommand("unknown", "test"), /Unknown harness/);
+  });
+});