@phi-code-admin/phi-code 0.73.0 → 0.74.1

package/agents/test.md

@@ -33,6 +33,9 @@ Use implementation results to know which files were created/modified and what be
 - **Realistic assertions**: Test what matters, not trivial details
 - **Match conventions**: Use the project's test framework, directory structure, and naming patterns
 - **Clean test code**: Tests are documentation — use descriptive names that explain expected behavior
+- Prefer targeted `edit` calls over full file rewrites. When a test fails, fix ONLY the failing test function, not the entire file
+- Maximum 1 full file rewrite per test file. After that, use `edit` for surgical fixes
+- When debugging test failures: read the error → locate the exact failing assertion → fix that specific line
 
 ## Test Writing
 

package/extensions/phi/memory.ts

@@ -43,9 +43,10 @@ export default function memoryExtension(pi: ExtensionAPI) {
 		description: "Search for content in memory using unified search (notes + ontology + vector search)",
 		promptSnippet: "Search project memory (notes, ontology, vector search). ALWAYS call before answering questions about prior work, decisions, or project context.",
 		promptGuidelines: [
-			"Before answering questions about prior work, architecture, decisions, or project context: call memory_search first.",
+			"MANDATORY: Before starting ANY task, call memory_search with relevant keywords. This is not optional.",
 			"When starting work on a topic, search memory for existing notes and learnings.",
 			"After completing important work or learning something new, use memory_write to save it.",
+			"MANDATORY: After completing any significant work, call memory_write to save what you did and what you learned.",
 			"When a command fails or produces an unexpected error, document the error and fix in memory_write (self-improvement).",
 			"When the user corrects you, save the correction in memory_write so you never repeat the mistake.",
 			"After a significant debugging session, write a summary of root cause and solution to memory.",

package/extensions/phi/orchestrator.ts

@@ -21,7 +21,6 @@ import type { ExtensionAPI } from "phi-code";
 import { writeFile, mkdir, readdir, readFile } from "node:fs/promises";
 import { join } from "node:path";
 import { existsSync, readFileSync } from "node:fs";
-// execFile removed — tasks now execute in-session, no subprocess
 import { homedir } from "node:os";
 
 // ─── Types ───────────────────────────────────────────────────────────────
@@ -35,22 +34,6 @@ interface TaskDef {
 	subtasks?: string[];
 }
 
-interface TaskResult {
-	taskIndex: number;
-	title: string;
-	agent: string;
-	status: "success" | "error" | "skipped";
-	output: string;
-	durationMs: number;
-}
-
-interface AgentDef {
-	name: string;
-	description: string;
-	tools: string;
-	systemPrompt: string;
-}
-
 // ─── Extension ───────────────────────────────────────────────────────────
 
 export default function orchestratorExtension(pi: ExtensionAPI) {
@@ -64,218 +47,7 @@ export default function orchestratorExtension(pi: ExtensionAPI) {
 		return new Date().toISOString().replace(/[:.]/g, "-").replace("T", "_").slice(0, 19);
 	}
 
-	// ─── Agent Discovery ─────────────────────────────────────────────
-
-	function loadAgentDefs(): Map<string, AgentDef> {
-		const agents = new Map<string, AgentDef>();
-		const dirs = [
-			join(process.cwd(), ".phi", "agents"),
-			join(homedir(), ".phi", "agent", "agents"),
-			join(__dirname, "..", "..", "..", "agents"),
-		];
-
-		for (const dir of dirs) {
-			if (!existsSync(dir)) continue;
-			try {
-				const files = require("fs").readdirSync(dir) as string[];
-				for (const file of files) {
-					if (!file.endsWith(".md")) continue;
-					const name = file.replace(".md", "");
-					if (agents.has(name)) continue;
-
-					try {
-						const content = readFileSync(join(dir, file), "utf-8");
-						const fmMatch = content.match(/^---\s*\n([\s\S]*?)\n---\s*\n([\s\S]*)$/);
-						if (!fmMatch) continue;
-
-						const frontmatter = fmMatch[1];
-						const body = fmMatch[2].trim();
-						const desc = frontmatter.match(/description:\s*(.+)/)?.[1] || "";
-						const tools = frontmatter.match(/tools:\s*(.+)/)?.[1] || "";
-
-						agents.set(name, { name, description: desc, tools, systemPrompt: body });
-					} catch { /* skip */ }
-				}
-			} catch { /* skip */ }
-		}
-
-		return agents;
-	}
-
-	function resolveAgentModel(agentType: string): string | null {
-		const routingPath = join(homedir(), ".phi", "agent", "routing.json");
-		try {
-			const config = JSON.parse(readFileSync(routingPath, "utf-8"));
-			for (const [_cat, route] of Object.entries(config.routes || {})) {
-				const r = route as any;
-				if (r.agent === agentType) return r.preferredModel || null;
-			}
-			// Map agent type to route category
-			const categoryMap: Record<string, string> = {
-				code: "code", explore: "explore", plan: "plan",
-				test: "test", review: "review", debug: "debug",
-			};
-			const category = categoryMap[agentType];
-			if (category && config.routes?.[category]) {
-				return config.routes[category].preferredModel || null;
-			}
-			return config.default?.model || null;
-		} catch {
-			return null;
-		}
-	}
-
-	function findPhiBinary(): string {
-		// Try the bundled CLI relative to extensions dir
-		const bundledCli = join(__dirname, "..", "..", "..", "dist", "cli.js");
-		if (existsSync(bundledCli)) return bundledCli;
-
-		// Try npm global install paths
-		const npmGlobalPaths = [
-			join(homedir(), "AppData", "Roaming", "npm", "node_modules", "@phi-code-admin", "phi-code", "dist", "cli.js"), // Windows
-			join(homedir(), ".npm-global", "lib", "node_modules", "@phi-code-admin", "phi-code", "dist", "cli.js"), // Linux custom
-			"/usr/local/lib/node_modules/@phi-code-admin/phi-code/dist/cli.js", // Linux/Mac default
-			"/usr/lib/node_modules/@phi-code-admin/phi-code/dist/cli.js", // Some Linux
-		];
-		for (const p of npmGlobalPaths) {
-			if (existsSync(p)) return p;
-		}
-
-		// Try `which phi` (Linux/Mac) or `where phi` (Windows)
-		try {
-			const isWin = process.platform === "win32";
-			const cmd = isWin ? "where" : "which";
-			const result = require("child_process").execSync(`${cmd} phi 2>${isWin ? "NUL" : "/dev/null"}`, { encoding: "utf-8" }).trim();
-			if (result) {
-				const firstLine = result.split("\n")[0].trim();
-				// On Windows, `where phi` returns the .cmd shim; we need the actual JS
-				if (isWin && firstLine.endsWith(".cmd")) {
-					const npmPrefix = require("child_process").execSync("npm prefix -g", { encoding: "utf-8" }).trim();
-					const jsPath = join(npmPrefix, "node_modules", "@phi-code-admin", "phi-code", "dist", "cli.js");
-					if (existsSync(jsPath)) return jsPath;
-				}
-				return firstLine;
-			}
-		} catch { /* not in PATH */ }
-
-		// Last resort: assume phi is in PATH (works with shell:true on Windows)
-		return "phi";
-	}
-
-	// ─── Task Execution (in-session, no subprocess) ─────────────────
-
-	/**
-	 * Execute a task by sending it as a user message to the current session.
-	 * The LLM handles it directly — no subprocess spawning, no cold boot.
-	 * Much faster and more reliable than spawning phi --print processes.
-	 */
-	function executeTaskInSession(
-		task: TaskDef,
-		sharedContext: {
-			projectTitle: string;
-			projectDescription: string;
-			specSummary: string;
-			completedTasks: Array<{ index: number; title: string; agent: string; output: string }>;
-		},
-	): { taskPrompt: string } {
-		const agentType = task.agent || "code";
-
-		// Build prompt with shared context
-		let taskPrompt = `## 🔧 Task: ${task.title} [${agentType}]\n\n`;
-
-		taskPrompt += `**Project:** ${sharedContext.projectTitle}\n\n`;
-
-		if (sharedContext.specSummary) {
-			taskPrompt += `**Spec:** ${sharedContext.specSummary}\n\n`;
-		}
-
-		// Inject results from dependency tasks
-		const deps = task.dependencies || [];
-		if (deps.length > 0) {
-			const depResults = sharedContext.completedTasks.filter(ct => deps.includes(ct.index));
-			if (depResults.length > 0) {
-				taskPrompt += `**Previous results:**\n`;
-				for (const dep of depResults) {
-					const truncated = dep.output.length > 500 ? dep.output.slice(0, 500) + "..." : dep.output;
-					taskPrompt += `- Task ${dep.index} (${dep.title}): ${truncated}\n`;
-				}
-				taskPrompt += "\n";
-			}
-		}
-
-		// The actual task
-		taskPrompt += `### What to do\n\n${task.description}\n`;
-		if (task.subtasks && task.subtasks.length > 0) {
-			taskPrompt += "\n**Sub-tasks:**\n" + task.subtasks.map((st, i) => `${i + 1}. ${st}`).join("\n") + "\n";
-		}
-		taskPrompt += `\n**Instructions:** Execute this task completely. Create/edit all necessary files. Report what you did.\n`;
-
-		return { taskPrompt };
-	}
-
-	// ─── Execute All Tasks (parallel with dependency resolution) ─────
-
-	async function executePlan(
-		tasks: TaskDef[],
-		todoFile: string,
-		notify: (msg: string, type: "info" | "error" | "warning") => void,
-		projectContext?: { title: string; description: string; specSummary: string },
-	): Promise<{ results: TaskResult[]; progressFile: string }> {
-		const progressFile = todoFile.replace("todo-", "progress-");
-		const progressPath = join(plansDir, progressFile);
-		const totalTasks = tasks.length;
-
-		const sharedContext = {
-			projectTitle: projectContext?.title || "Project",
-			projectDescription: projectContext?.description || "",
-			specSummary: projectContext?.specSummary || "",
-			completedTasks: [] as Array<{ index: number; title: string; agent: string; output: string }>,
-		};
-
-		notify(`🚀 Executing ${totalTasks} tasks in-session...`, "info");
-
-		// Build a single comprehensive prompt with ALL tasks
-		let megaPrompt = `# 📋 Project: ${sharedContext.projectTitle}\n\n`;
-		megaPrompt += `${sharedContext.projectDescription}\n\n`;
-		if (sharedContext.specSummary) {
-			megaPrompt += `## Spec\n${sharedContext.specSummary}\n\n`;
-		}
-		megaPrompt += `## Tasks (execute ALL in order)\n\n`;
-
-		const results: TaskResult[] = [];
-
-		for (let i = 0; i < tasks.length; i++) {
-			const task = tasks[i];
-			const { taskPrompt } = executeTaskInSession(task, sharedContext);
-			megaPrompt += `---\n\n${taskPrompt}\n\n`;
-			results.push({
-				taskIndex: i + 1, title: task.title,
-				agent: task.agent || "code", status: "success",
-				output: "(in-session)", durationMs: 0,
-			});
-		}
-
-		megaPrompt += `---\n\n## ⚠️ Instructions\n\n`;
-		megaPrompt += `Execute ALL ${totalTasks} tasks above **sequentially**. For each task:\n`;
-		megaPrompt += `1. Create/edit the required files using your tools\n`;
-		megaPrompt += `2. Report what you did briefly\n`;
-		megaPrompt += `3. Move to the next task\n\n`;
-		megaPrompt += `Do NOT skip any task. Complete the entire project.\n`;
-
-		// Write progress file
-		let progress = `# Progress: ${todoFile}\n\n`;
-		progress += `**Started:** ${new Date().toLocaleString()}\n`;
-		progress += `**Tasks:** ${totalTasks} | **Mode:** in-session\n\n`;
-		for (const r of results) {
-			progress += `- Task ${r.taskIndex}: ${r.title} [${r.agent}]\n`;
-		}
-		await writeFile(progressPath, progress, "utf-8");
-
-		// Return the mega-prompt as tool result — LLM sees it and executes
-		return { results, progressFile, megaPrompt };
-	}
-
-	// ─── Generate Plan Files ─────────────────────────────────────────
+	// ─── Plan File Generators ────────────────────────────────────────
 
 	function generateSpec(p: {
 		title: string; description: string; goals: string[]; requirements: string[];
@@ -331,20 +103,21 @@ export default function orchestratorExtension(pi: ExtensionAPI) {
 		return todo;
 	}
 
-	// ─── Orchestrate Tool (plan + auto-execute) ──────────────────────
+	// ─── Orchestrate Tool — Structured planning with prompt-architect pattern ───
 
 	pi.registerTool({
 		name: "orchestrate",
 		label: "Project Orchestrator",
-		description: "Create a project plan AND automatically execute all tasks with sub-agents in parallel. Each agent gets its own isolated context, model, and system prompt. Tasks without dependencies run simultaneously.",
-		promptSnippet: "Plan + execute projects in parallel waves. Each sub-agent gets isolated context + model. Use prompt-architect patterns for structured task descriptions.",
+		description: "Create a structured project plan with spec, todo, and task breakdown. Forces the model to decompose the project into goals, requirements, architecture, constraints, and success criteria before execution.",
+		promptSnippet: "Plan + structure projects using the prompt-architect pattern. Each task gets [CONTEXT] → [TASK] → [FORMAT] → [CONSTRAINTS].",
 		promptGuidelines: [
-			"When asked to plan or build a project: analyze the request thoroughly, then call the orchestrate tool. It plans AND executes automatically.",
-			"CRITICAL: Each task description must be SELF-CONTAINED. The sub-agent has NO access to this conversation. It receives: (1) project context (title, description, spec summary) automatically, (2) outputs from its dependency tasks automatically, (3) your task description. So include specific details: file paths, expected behavior, code patterns, success criteria. Don't repeat the project description — that's injected automatically.",
+			// NOTE: The old guideline "When asked to plan or build a project, call the orchestrate tool"
+			// was REMOVED because it hijacked the /plan command flow. The orchestrate tool is now
+			// only called explicitly or when the model decides structured planning is needed outside of /plan.
+			"IMPORTANT: Do NOT call orchestrate during /plan orchestration phases. The /plan command handles its own workflow.",
 			"Structure each task description using the prompt-architect pattern: [CONTEXT] what exists and why → [TASK] what to do specifically → [FORMAT] expected output → [CONSTRAINTS] rules and limitations.",
-			"Assign agent types strategically: 'explore' (read-only analysis, codebase understanding), 'plan' (architecture, design decisions), 'code' (implementation, file creation/modification), 'test' (write + run tests, validate behavior), 'review' (security audit, quality check, read-only).",
-			"Set dependencies to maximize parallelism: tasks without dependencies run simultaneously in the same wave. Only add dependencies when a task truly needs another task's output.",
-			"Order tasks logically: explore → plan → code → test → review. But allow independent tasks at each stage to run in parallel.",
+			"Assign agent types strategically: 'explore' (read-only analysis), 'plan' (architecture), 'code' (implementation), 'test' (write+run tests), 'review' (quality audit).",
+			"Set dependencies to maximize parallelism: tasks without dependencies run simultaneously.",
 			"Set priority=high for critical-path tasks, medium for standard work, low for nice-to-haves.",
 		],
 		parameters: Type.Object({
@@ -356,22 +129,29 @@ export default function orchestratorExtension(pi: ExtensionAPI) {
 			tasks: Type.Array(
 				Type.Object({
 					title: Type.String({ description: "Clear, action-oriented task title" }),
-					description: Type.String({ description: "SELF-CONTAINED task description. Include ALL context the sub-agent needs: file paths, expected behavior, code patterns, conventions. The agent has NO shared history." }),
-					agent: Type.Optional(Type.String({ description: "Agent type: explore (read-only analysis), plan (architecture), code (implementation), test (write+run tests), review (quality audit)" })),
-					priority: Type.Optional(Type.String({ description: "high (critical path), medium (standard), low (nice-to-have)" })),
-					dependencies: Type.Optional(Type.Array(Type.Number(), { description: "Task numbers this depends on (1-indexed). Only add when truly needed — fewer dependencies = more parallelism" })),
+					description: Type.String({ description: "SELF-CONTAINED task description. Include ALL context the sub-agent needs." }),
+					agent: Type.Optional(Type.String({ description: "Agent type: explore, plan, code, test, review" })),
+					priority: Type.Optional(Type.String({ description: "high, medium, low" })),
+					dependencies: Type.Optional(Type.Array(Type.Number(), { description: "Task numbers this depends on (1-indexed)" })),
 					subtasks: Type.Optional(Type.Array(Type.String(), { description: "Specific sub-steps within this task" })),
 				}),
-				{ description: "Ordered list of tasks. Independent tasks run in parallel. Dependent tasks wait for prerequisites." }
+				{ description: "Ordered list of tasks" }
 			),
-			constraints: Type.Optional(Type.Union([Type.Array(Type.String()), Type.String()], { description: "Hard constraints: frameworks, patterns, rules, things to avoid" })),
-			successCriteria: Type.Optional(Type.Union([Type.Array(Type.String()), Type.String()], { description: "How to verify the project is complete and correct" })),
+			constraints: Type.Optional(Type.Union([Type.Array(Type.String()), Type.String()], { description: "Hard constraints: things to avoid, rules" })),
+			successCriteria: Type.Optional(Type.Union([Type.Array(Type.String()), Type.String()], { description: "How to verify the project is complete" })),
 		}),
 
-		async execute(_toolCallId, params, _signal, _onUpdate, ctx) {
+		async execute(_toolCallId, params, _signal, _onUpdate, _ctx) {
+			// Block if /plan orchestration is active — prevent hijacking
+			if (orchestrationActive) {
+				return {
+					content: [{ type: "text", text: "⚠️ Orchestration is already active via /plan. Do NOT call orchestrate during /plan phases. Follow the phase instructions instead." }],
+				};
+			}
+
 			const raw = params as any;
 
-			// Normalize string fields to arrays (some models send strings instead of arrays)
+			// Normalize string fields to arrays
 			const toArray = (v: any): string[] => {
 				if (!v) return [];
 				if (Array.isArray(v)) return v;
@@ -396,48 +176,22 @@ export default function orchestratorExtension(pi: ExtensionAPI) {
 				const specFile = `spec-${ts}.md`;
 				const todoFile = `todo-${ts}.md`;
 
-				// Generate and write plan files
+				// Generate and write structured plan files
 				const spec = generateSpec(p);
 				const todo = generateTodo(p.title, p.tasks);
 				await writeFile(join(plansDir, specFile), spec, "utf-8");
 				await writeFile(join(plansDir, todoFile), todo, "utf-8");
 
-				// Notify plan created
-				const notify = (msg: string, type: "info" | "error" | "warning") => {
-					// Use onUpdate for streaming progress to the user
-					if (_onUpdate) {
-						_onUpdate({ content: [{ type: "text", text: msg }] });
-					}
-				};
-
-				notify(`📋 Plan created: **${p.title}** (${p.tasks.length} tasks)\nNow executing with sub-agents...`, "info");
+				const summary = `📋 **Project "${p.title}" — Structured plan created!**\n\n` +
+					`**Spec:** \`${specFile}\` | **Todo:** \`${todoFile}\`\n\n` +
+					`**Goals:** ${p.goals.length} | **Requirements:** ${p.requirements.length} | **Tasks:** ${p.tasks.length}\n\n` +
+					`The spec and todo files are ready in \`.phi/plans/\`. ` +
+					`Use \`/plan ${p.description.slice(0, 100)}\` to execute with the 5-phase orchestrator, ` +
+					`or work through the tasks manually.`;
 
-				// Auto-execute all tasks
-				// Build spec summary for shared context
-				const specSummary = [
-					`Goals: ${p.goals.join("; ")}`,
-					`Requirements: ${p.requirements.join("; ")}`,
-					p.architecture?.length ? `Architecture: ${p.architecture.join("; ")}` : "",
-					p.constraints?.length ? `Constraints: ${p.constraints.join("; ")}` : "",
-				].filter(Boolean).join("\n");
-
-				const { results, progressFile, megaPrompt } = await executePlan(
-					p.tasks, todoFile, notify,
-					{ title: p.title, description: p.description, specSummary },
-				);
-
-				const header = `**📋 Project "${p.title}" — ${p.tasks.length} tasks planned!**\n` +
-					`Plan: \`${specFile}\`, \`${todoFile}\` | Progress: \`${progressFile}\`\n\n` +
-					`---\n\n`;
-
-				// Return the mega-prompt as tool result
-				// The LLM sees this and executes all tasks in its current turn
 				return {
-					content: [{ type: "text", text: header + megaPrompt }],
-					details: {
-						specFile, todoFile, progressFile,
-						taskCount: p.tasks.length,
-					},
+					content: [{ type: "text", text: summary }],
+					details: { specFile, todoFile, taskCount: p.tasks.length },
 				};
 			} catch (error) {
 				return {
@@ -1089,61 +843,12 @@ Tag the note with relevant keywords for vector search.
 	// ─── /run Command — Re-execute existing plan ─────────────────────
 
 	pi.registerCommand("run", {
-		description: "Re-execute an existing plan's tasks with sub-agents",
-		handler: async (args, ctx) => {
-			if (!existsSync(plansDir)) {
-				ctx.ui.notify("No plans found. Use `/plan <description>` to create and execute one.", "warning");
-				return;
-			}
-
-			const files = (await readdir(plansDir)).sort().reverse();
-			const todoFiles = files.filter(f => f.startsWith("todo-") && f.endsWith(".md"));
-
-			if (todoFiles.length === 0) {
-				ctx.ui.notify("No todo files found. Use `/plan <description>` first.", "warning");
-				return;
-			}
-
-			const todoFile = todoFiles[0];
-			const todoContent = await readFile(join(plansDir, todoFile), "utf-8");
-
-			// Parse tasks
-			const tasks: TaskDef[] = [];
-			const sections = todoContent.split(/## Task \d+:/);
-			for (let i = 1; i < sections.length; i++) {
-				const section = sections[i];
-				const titleMatch = section.match(/^(.+?)(?:\s*🔴|\s*🟡|\s*🟢)/);
-				const agentMatch = section.match(/\[(\w+)\]/);
-				const descMatch = section.match(/- \[ \] (.+)/);
-				const subtasks: string[] = [];
-				const stMatches = section.matchAll(/  - \[ \] (.+)/g);
-				for (const m of stMatches) subtasks.push(m[1]);
-
-				if (titleMatch && descMatch) {
-					tasks.push({
-						title: titleMatch[1].trim(),
-						agent: agentMatch?.[1] || "code",
-						description: descMatch[1].trim(),
-						subtasks: subtasks.length > 0 ? subtasks : undefined,
-					});
-				}
-			}
-
-			if (tasks.length === 0) {
-				ctx.ui.notify("Could not parse tasks from todo file.", "error");
-				return;
-			}
-
-			const confirmed = await ctx.ui.confirm(
-				"Re-execute Plan",
-				`${tasks.length} tasks found in \`${todoFile}\`.\nEach will spawn an isolated sub-agent.\n\nProceed?`
+		description: "Re-execute an existing plan (deprecated — use /plan instead)",
+		handler: async (_args, ctx) => {
+			ctx.ui.notify(
+				"⚠️ `/run` is deprecated. Use `/plan <description>` to create and execute a new plan with the 5-phase orchestrator.",
+				"warning",
 			);
-			if (!confirmed) {
-				ctx.ui.notify("Cancelled.", "info");
-				return;
-			}
-
-			await executePlan(tasks, todoFile, (msg, type) => ctx.ui.notify(msg, type));
 		},
 	});
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@phi-code-admin/phi-code",
-  "version": "0.73.0",
+  "version": "0.74.1",
   "description": "Coding agent CLI with read, bash, edit, write tools and session management",
   "type": "module",
   "piConfig": {