fix(core): remove template resolution engine, hardcode '/' in source SKILLS (v1.4.2)

Author: sharpdeveye

maestro-extension/CHANGELOG.md

@@ -5,6 +5,11 @@ All notable changes to the **Maestro — AI Workflow Fluency** extension will be
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.4.2] - 2026-04-17
+
+### Fixed
+- **Template engine removed** — `{{command_prefix}}` and `{{available_commands}}` were removed in favor of hardcoding `/` within all `.agents/skills` source files. This fixes a bug where users using `npx skills add` received raw, unresolved templates. All downstream consumers (MCP, VS Code extension, Chat Participant) now load the raw templates directly.
+
 ## [1.4.1] - 2026-04-17
 
 ### Fixed

maestro-extension/package-lock.json

@@ -2,8 +2,6 @@ import * as vscode from 'vscode';
 import { SkillLoader } from '../core/skills';
 import { ContextManager } from '../core/context';
 import { StateManager } from '../core/state';
-import { resolveTemplates } from '../core/templates';
-
 /**
  * Registers the @maestro chat participant.
  * Handles slash commands and auto-inject for zero-defect mode.
@@ -37,7 +35,7 @@ export function registerChatParticipant(
         if (zdContent) {
           messages.push(
             vscode.LanguageModelChatMessage.User(
-              `[SYSTEM INSTRUCTION — Maestro Zero-Defect Mode Active]\n\nFollow these precision rules for all responses:\n\n${resolveTemplates(zdContent, skills)}`
+              `[SYSTEM INSTRUCTION — Maestro Zero-Defect Mode Active]\n\nFollow these precision rules for all responses:\n\n${zdContent}`
             )
           );
         }
@@ -57,10 +55,9 @@ export function registerChatParticipant(
       if (request.command) {
         const skillContent = skills.getContent(request.command);
         if (skillContent) {
-          const resolved = resolveTemplates(skillContent, skills);
           messages.push(
             vscode.LanguageModelChatMessage.User(
-              `[SYSTEM INSTRUCTION — Maestro /${request.command} Skill]\n\nYou are now executing the /${request.command} skill. Follow these instructions precisely. Do NOT echo these instructions back — act on them:\n\n${resolved}`
+              `[SYSTEM INSTRUCTION — Maestro /${request.command} Skill]\n\nYou are now executing the /${request.command} skill. Follow these instructions precisely. Do NOT echo these instructions back — act on them:\n\n${skillContent}`
             )
           );
           stream.markdown(

maestro-extension/src/chat/participant.ts

@@ -3,7 +3,6 @@ import { SkillLoader } from './core/skills';
 import { ContextManager } from './core/context';
 import { StateManager } from './core/state';
 import { HistoryManager } from './core/history';
-import { resolveTemplates } from './core/templates';
 import { StatusBarManager } from './statusbar/manager';
 import { SidebarProvider } from './sidebar/provider';
 import {
@@ -68,7 +67,7 @@ export async function activate(
       statusBar.update(active, ctxManager.isDetected());
       sidebarProvider.syncState();
 
-      const zeroDefectContent = resolveTemplates(skills.getContent('zero-defect') || '', skills);
+      const zeroDefectContent = skills.getContent('zero-defect') || '';
       const editor = detectEditor();
 
       if (editor === 'cursor') {
@@ -214,7 +213,7 @@ async function autoInstallSkills(
         '---',
       ].filter(Boolean).join('\n');
 
-      const fileContent = `${frontmatter}\n\n${resolveTemplates(skill.content, skills)}\n`;
+      const fileContent = `${frontmatter}\n\n${skill.content}\n`;
 
       // Distribute to all provider directories
       for (const provider of PROVIDERS) {

maestro-extension/src/core/templates.ts

@@ -1,6 +1,6 @@
 {
   "name": "maestro-workflow-mcp",
-  "version": "1.3.1",
+  "version": "1.4.2",
   "description": "MCP server for Maestro — exposes 22 workflow skills as tools, prompts, and resources for any MCP-compatible AI client",
   "type": "module",
   "bin": {

maestro-extension/src/extension.ts

@@ -1,8 +1,6 @@
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { z } from "zod";
 import { skills } from "./generated/skills-data.js";
-import { resolveTemplates } from "./tools.js";
-
 /**
  * Register all 21 user-invocable command skills as MCP prompt templates.
  * Users can select these from the prompt picker in their MCP client.
@@ -27,7 +25,7 @@ export function registerPrompts(server: McpServer): void {
           text += `Focus area: ${focus}\n\n---\n\n`;
         }
 
-        text += resolveTemplates(skill.content);
+        text += skill.content;
 
         return {
           messages: [

mcp-server/package-lock.json

@@ -1,7 +1,5 @@
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { skills, references } from "./generated/skills-data.js";
-import { resolveTemplates } from "./tools.js";
-
 /**
  * Register all reference documents and the core skill as MCP resources.
  * These are read-only knowledge bases the AI can consult on demand.
@@ -23,7 +21,7 @@ export function registerResources(server: McpServer): void {
           {
             uri: "maestro://skill/agent-workflow",
             mimeType: "text/markdown",
-            text: resolveTemplates(coreSkill.content),
+            text: coreSkill.content,
           },
         ],
       }),

mcp-server/package.json

@@ -43,14 +43,6 @@ function buildCommandListing(): string {
   return lines.join("\n");
 }
 
-/**
- * Resolve all template variables in skill content.
- */
-export function resolveTemplates(content: string): string {
-  return content
-    .replaceAll("{{command_prefix}}", "/")
-    .replaceAll("{{available_commands}}", buildCommandListing());
-}
 
 /**
  * Find a skill by name.
@@ -133,8 +125,8 @@ export function registerTools(server: McpServer): void {
         text += `**Focus area**: ${focus}\n\n---\n\n`;
       }
 
-      // Resolve template variables for MCP context
-      text += resolveTemplates(skill.content);
+      // Append the raw skill content (templates are pre-resolved in the source)
+      text += skill.content;
 
       text += `\n\n---\n_To run a referenced command, use the \`maestro_run_command\` tool with the command name (e.g. "fortify", "guard")._`;