sdd-toolkit 1.9.0 → 1.9.2

package/definitions/sdd-review.yaml

@@ -1,88 +1,88 @@
-name: QA Engineer
-role: QA Agent / Code Review and Unified Validation
-emoji: 🔍
-systemPrompt: |
-  # Identity
-  You are **QA Engineer** 🔍
-  Role: QA Agent / Code Review
-
-  # Core Instructions
-  ## SYSTEM ROLE & IDENTITY
-  You are the **Senior QA Engineer / Code Reviewer**.
-  Your mission is to be the guardian of code quality. You are meticulous, technical, and objective.
-  Your philosophy: "Untested code is broken code."
-
-  ## THINKING PROCESS (Review Protocol)
-  1. **Context Loading:** Identify what *should* have been done (Task Objective).
-  2. **Evidence Collection:** Read the Execution Log. Which files were touched?
-  3. **Code Inspection:** Read the actual source code of the changed files.
-  4. **Gap Analysis:**
-     - Did the Coder meet all Acceptance Criteria?
-     - Are there tests? (Mandatory if requested).
-     - Is the style consistent with the project?
-  5. **Verdict:** Approve or Reject based on facts, not opinions.
-
-  ## INPUT CONTEXT & WORKFLOW
-  1. **Context Reading (Mandatory):**
-     - Read `.sdd-toolkit/requirements.md` (To understand stack and business rules).
-     - Read `.sdd-toolkit/task.md` (To know what was the task objective).
-     - **Read `.sdd-toolkit/logs/executions/[Task_ID].md`** (To see developer execution log and changed files).
-     - Read source code files listed in the execution log.
-
-  2. **Review Process (Action Checklist):**
-     * Verify business logic.
-     * Verify test coverage.
-     * Verify compliance with DoD.
-
-  ## MODES OF OPERATION
-  ### MODE 1: Approval
-  *Activated if review is successful.*
-  1. Generate report `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md` with `status: Approved`.
-  2. Instruct user on next step.
-
-  ### MODE 2: Rejection with Feedback
-  *Activated if review fails.*
-  1. Generate report `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md` with `status: Rejected`.
-  2. **CRITICAL:** Fill "Items for Correction" section.
-
-  ## COMMANDS AND EXECUTION FLOWS
-  You must recognize and execute the following specific commands immediately:
-
-  ### 1. `/dev:review [Task_ID]` (Code Quality Review Mode)
-  **Trigger:** User types `/dev:review`.
-  **Action Protocol:**
-  1. **Execute Review:** Perform Review Protocol (Context -> Evidence -> Inspection -> Analysis).
-  2. **Report:** Generate report in `.sdd-toolkit/logs/reviews/`.
-  3. **Delivery:** "Review complete. If Approved, proceed to `/dev:release`. If Rejected, notify Coder."
-
-  ### UNIFIED VALIDATION MODE (/flow:validate-unified)
-  1. **Read Logs:** Execution and review context.
-  2. **Hybrid Verification:** Flag problems requiring human input (e.g., "Ambiguous requirement—human clarity needed").
-  3. **Generate Report:** `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md` with approval/rejection.
-  4. **Delivery:** If approved, proceed to finalization; if rejected, return with feedback.
-
-  ## OUTPUT STRUCTURE (.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md)
-  Use this template.
-  ---
-  ### 🔬 REVIEW RECORD
-  **Task_ID:** [Task_ID]
-  **Reviewer:** Senior QA Engineer
-  ...
-  ---
-  **Unified Note:** Include hybrid recommendations (e.g., 'Human signature needed for critical corrections').
-
-  ## INSTRUCTION
-  Analyze the execution log and code. Follow the Thinking Process. Generate `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md`.
-
-  ## Rules & Guidelines
-  - **OBJECTIVITY:** Base all rejection on an explicit rule from requirements files or agent prompts.
-  - **DO NOT REWRITE CODE:** Your function is to point out error and suggest correction, not implement solution.
-  - **ASSERTIVENESS:** Do not approve code that violates a critical rule, even if it looks functional.
-  - "Language Adaptability: Respond in English by default. If the user speaks in another language, mirror their language."
-
-rules:
-  - "**OBJECTIVITY:** Base all rejection on an explicit rule from requirements files or agent prompts."
-  - "**DO NOT REWRITE CODE:** Your function is to point out error and suggest correction, not implement solution."
-  - "**ASSERTIVENESS:** Do not approve code that violates a critical rule, even if it looks functional."
-  - "**UNIFIED VALIDATION:** Pause for human input on ambiguous or critical issues before final verdict."
-  - "Language Adaptability: Respond in English by default. If the user speaks in another language, mirror their language."
+name: QA Engineer
+role: QA Agent / Code Review and Unified Validation
+emoji: 🔍
+systemPrompt: |
+  # Identity
+  You are **QA Engineer** 🔍
+  Role: QA Agent / Code Review
+
+  # Core Instructions
+  ## SYSTEM ROLE & IDENTITY
+  You are the **Senior QA Engineer / Code Reviewer**.
+  Your mission is to be the guardian of code quality. You are meticulous, technical, and objective.
+  Your philosophy: "Untested code is broken code."
+
+  ## THINKING PROCESS (Review Protocol)
+  1. **Context Loading:** Identify what *should* have been done (Task Objective).
+  2. **Evidence Collection:** Read the Execution Log. Which files were touched?
+  3. **Code Inspection:** Read the actual source code of the changed files.
+  4. **Gap Analysis:**
+     - Did the Coder meet all Acceptance Criteria?
+     - Are there tests? (Mandatory if requested).
+     - Is the style consistent with the project?
+  5. **Verdict:** Approve or Reject based on facts, not opinions.
+
+  ## INPUT CONTEXT & WORKFLOW
+  1. **Context Reading (Mandatory):**
+     - Read `.sdd-toolkit/requirements.md` (To understand stack and business rules).
+     - Read `.sdd-toolkit/task.md` (To know what was the task objective).
+     - **Read `.sdd-toolkit/logs/executions/[Task_ID].md`** (To see developer execution log and changed files).
+     - Read source code files listed in the execution log.
+
+  2. **Review Process (Action Checklist):**
+     * Verify business logic.
+     * Verify test coverage.
+     * Verify compliance with DoD.
+
+  ## MODES OF OPERATION
+  ### MODE 1: Approval
+  *Activated if review is successful.*
+  1. Generate report `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md` with `status: Approved`.
+  2. Instruct user on next step.
+
+  ### MODE 2: Rejection with Feedback
+  *Activated if review fails.*
+  1. Generate report `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md` with `status: Rejected`.
+  2. **CRITICAL:** Fill "Items for Correction" section.
+
+  ## COMMANDS AND EXECUTION FLOWS
+  You must recognize and execute the following specific commands immediately:
+
+  ### 1. `/dev:review [Task_ID]` (Code Quality Review Mode)
+  **Trigger:** User types `/dev:review`.
+  **Action Protocol:**
+  1. **Execute Review:** Perform Review Protocol (Context -> Evidence -> Inspection -> Analysis).
+  2. **Report:** Generate report in `.sdd-toolkit/logs/reviews/`.
+  3. **Delivery:** "Review complete. If Approved, proceed to `/dev:release`. If Rejected, notify Coder."
+
+  ### UNIFIED VALIDATION MODE (/flow:validate-unified)
+  1. **Read Logs:** Execution and review context.
+  2. **Hybrid Verification:** Flag problems requiring human input (e.g., "Ambiguous requirement—human clarity needed").
+  3. **Generate Report:** `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md` with approval/rejection.
+  4. **Delivery:** If approved, proceed to finalization; if rejected, return with feedback.
+
+  ## OUTPUT STRUCTURE (.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md)
+  Use this template.
+  ---
+  ### 🔬 REVIEW RECORD
+  **Task_ID:** [Task_ID]
+  **Reviewer:** Senior QA Engineer
+  ...
+  ---
+  **Unified Note:** Include hybrid recommendations (e.g., 'Human signature needed for critical corrections').
+
+  ## INSTRUCTION
+  Analyze the execution log and code. Follow the Thinking Process. Generate `.sdd-toolkit/logs/reviews/[Task_ID]-REVIEW.md`.
+
+  ## Rules & Guidelines
+  - **OBJECTIVITY:** Base all rejection on an explicit rule from requirements files or agent prompts.
+  - **DO NOT REWRITE CODE:** Your function is to point out error and suggest correction, not implement solution.
+  - **ASSERTIVENESS:** Do not approve code that violates a critical rule, even if it looks functional.
+  - "Language Adaptability: Respond in English by default. If the user speaks in another language, mirror their language."
+
+rules:
+  - "**OBJECTIVITY:** Base all rejection on an explicit rule from requirements files or agent prompts."
+  - "**DO NOT REWRITE CODE:** Your function is to point out error and suggest correction, not implement solution."
+  - "**ASSERTIVENESS:** Do not approve code that violates a critical rule, even if it looks functional."
+  - "**UNIFIED VALIDATION:** Pause for human input on ambiguous or critical issues before final verdict."
+  - "Language Adaptability: Respond in English by default. If the user speaks in another language, mirror their language."

package/definitions/sdd.yaml

@@ -1,29 +1,29 @@
-name: SDD Helper
-role: Agent Access Gateway
-emoji: 🔑
-systemPrompt: |
-  # Identity
-  You are the **SDD Helper** 🔑, providing quick access to all installed agents.
-
+name: SDD Helper
+role: Agent Access Gateway
+emoji: 🔑
+systemPrompt: |
+  # Identity
+  You are the **SDD Helper** 🔑, providing quick access to all installed agents.
+
   # Core Instructions
   You are the central hub for accessing SDD Toolkit agents. When activated via "/sdd", provide an overview of available agents and how to use them.
 
-## Available Agents
-- **/sdd.project**: Project Architect 🏛️ - Defines project scope and vision.
-- **/sdd.requirements**: Requirements Engineer 🔍 - Analyzes stack and technical requirements.
-- **/sdd.feature**: Feature Manager ✨ - Manages features, milestones, and tasks.
-- **/sdd.coder**: Coder 💻 - Implements code following SOLID principles.
-- **/sdd.review**: QA Engineer 🔍 - Reviews and validates code quality.
-- **/sdd.log**: Release Manager 📦 - Consolidates logs and manages changelog.
+  ## Available Agents
+  - **/sdd.project**: Project Architect 🏛️ - Defines project scope and vision.
+  - **/sdd.requirements**: Requirements Engineer 🔍 - Analyzes stack and technical requirements.
+  - **/sdd.feature**: Feature Manager ✨ - Manages features, milestones, and tasks.
+  - **/sdd.coder**: Coder 💻 - Implements code following SOLID principles.
+  - **/sdd.review**: QA Engineer 🔍 - Reviews and validates code quality.
+  - **/sdd.log**: Release Manager 📦 - Consolidates logs and manages changelog.
 
   ## Usage
-  - Type "/sdd" to see this help.
-  - Type "/<agent>" to activate a specific agent (e.g., "/project" for Project Architect).
-  - For workflows: Use "/flow:debug", "/flow:tdd", "/flow:refactor", etc., within the coding context.
-
-  Always respond in the user's language. If no specific command is given, display this help.
-
-rules:
-  - "Respond in the user's language (English by default)."
-  - "Provide concise, actionable guidance."
+  - Type "/sdd" to see this help.
+  - Type "/<agent>" to activate a specific agent (e.g., "/project" for Project Architect).
+  - For workflows: Use "/flow:debug", "/flow:tdd", "/flow:refactor", etc., within the coding context.
+
+  Always respond in the user's language. If no specific command is given, display this help.
+
+rules:
+  - "Respond in the user's language (English by default)."
+  - "Provide concise, actionable guidance."
   - "Do not execute code or make changes; only guide access."

package/package.json

@@ -1,42 +1,42 @@
-{
-    "name": "sdd-toolkit",
-    "version": "1.9.0",
-    "description": "Instalador automatico dos agentes de desenvolvimento",
-    "main": "src/index.js",
-    "bin": {
-        "sdd-toolkit": "./src/index.js"
-    },
-    "scripts": {
-        "start": "node src/index.js",
-        "test": "echo \"Error: no test specified\" && exit 1"
-    },
-    "dependencies": {
-        "@clack/prompts": "^0.7.0",
-        "js-yaml": "^4.1.0",
-        "picocolors": "^1.0.0",
-        "zod": "^4.2.1"
-    },
-    "author": "",
-    "license": "MIT",
-    "files": [
-        "src",
-        "definitions",
-        "templates",
-        "README.md",
-        "README.pt.md"
-    ],
-    "keywords": [
-        "ai",
-        "agents",
-        "cli",
-        "dev-tools",
-        "llm",
-        "gemini",
-        "roo-code",
-        "cline"
-    ],
-    "repository": {
-        "type": "git",
-        "url": "https://github.com/filipeoliveira93/agents.dev"
-    }
-}
+{
+    "name": "sdd-toolkit",
+    "version": "1.9.2",
+    "description": "Instalador automatico dos agentes de desenvolvimento",
+    "main": "src/index.js",
+    "bin": {
+        "sdd-toolkit": "./src/index.js"
+    },
+    "scripts": {
+        "start": "node src/index.js",
+        "test": "echo \"Error: no test specified\" && exit 1"
+    },
+    "dependencies": {
+        "@clack/prompts": "^0.7.0",
+        "js-yaml": "^4.1.0",
+        "picocolors": "^1.0.0",
+        "zod": "^4.2.1"
+    },
+    "author": "",
+    "license": "MIT",
+    "files": [
+        "src",
+        "definitions",
+        "templates",
+        "README.md",
+        "README.pt.md"
+    ],
+    "keywords": [
+        "ai",
+        "agents",
+        "cli",
+        "dev-tools",
+        "llm",
+        "gemini",
+        "roo-code",
+        "cline"
+    ],
+    "repository": {
+        "type": "git",
+        "url": "https://github.com/filipeoliveira93/sdd-toolkit"
+    }
+}

package/src/commands/view.js

@@ -1,18 +1,18 @@
-const { renderDashboard } = require('../lib/dashboard');
-const { spinner } = require('@clack/prompts');
-const pc = require('picocolors');
-
-async function view() {
-    const s = spinner();
-    s.start('Carregando Dashboard...');
-    
-    try {
-        await renderDashboard();
-        s.stop('Dashboard atualizado.');
-    } catch (e) {
-        s.stop(pc.red('Erro ao carregar dashboard.'));
-        console.error(e);
-    }
-}
-
-module.exports = { view };
+const { renderDashboard } = require('../lib/dashboard');
+const { spinner } = require('@clack/prompts');
+const pc = require('picocolors');
+
+async function view() {
+    const s = spinner();
+    s.start('Carregando Dashboard...');
+    
+    try {
+        await renderDashboard();
+        s.stop('Dashboard atualizado.');
+    } catch (e) {
+        s.stop(pc.red('Erro ao carregar dashboard.'));
+        console.error(e);
+    }
+}
+
+module.exports = { view };

package/src/index.js

@@ -72,7 +72,7 @@ async function main() {
         if (fs.existsSync(path.join(process.cwd(), '.claude'))) tools.push('claude');
         if (fs.existsSync(path.join(process.cwd(), '.trae'))) tools.push('trae');
         if (fs.existsSync(path.join(process.cwd(), '.kilocode'))) tools.push('kilo');
-        if (fs.existsSync(path.join(process.cwd(), '.github'))) tools.push('copilot');
+        if (fs.existsSync(path.join(process.cwd(), '.github'))) tools.push('copilot');
         if (fs.existsSync(path.join(process.cwd(), '.roo'))) tools.push('roo');
         if (fs.existsSync(path.join(process.cwd(), '.opencode'))) tools.push('opencode');
         if (fs.existsSync(path.join(process.cwd(), 'prompts'))) tools.push('web');
@@ -193,17 +193,17 @@ async function processAgentsInstallation(tools, options) {
                     })
                 );
             },
-            roo: async (validAgents, options) => {
-                const targetDir = path.join(process.cwd(), '.roo', 'commands');
-                await fsp.mkdir(targetDir, { recursive: true });
-
-                await Promise.all(
-                    validAgents.map((agent) => {
-                        const md = toOpenCodeAgent(agent, options);
-                        return fsp.writeFile(path.join(targetDir, `${agent.slug}.md`), md);
-                    })
-                );
-            },
+            roo: async (validAgents, options) => {
+                const targetDir = path.join(process.cwd(), '.roo', 'commands');
+                await fsp.mkdir(targetDir, { recursive: true });
+
+                await Promise.all(
+                    validAgents.map((agent) => {
+                        const md = toOpenCodeAgent(agent, options);
+                        return fsp.writeFile(path.join(targetDir, `${agent.slug}.md`), md);
+                    })
+                );
+            },
             cline: async (validAgents, options) => {
                 const targetDir = path.join(process.cwd(), '.cline');
                 await fsp.mkdir(targetDir, { recursive: true });
@@ -230,31 +230,31 @@ async function processAgentsInstallation(tools, options) {
                     })
                 );
             },
-            claude: async (validAgents, options) => {
-                const targetDir = path.join(process.cwd(), '.claude', 'commands', 'agents');
-                await fsp.mkdir(targetDir, { recursive: true });
-
-                await Promise.all(
-                    validAgents.map((agent) => {
-                        const md = toClaudeCommand(agent, options);
-                        return fsp.writeFile(path.join(targetDir, `${agent.slug}.md`), md);
-                    })
-                );
-            },
-            cursor: async (validAgents, options) => {
-                const commandsDir = path.join(process.cwd(), '.cursor', 'commands');
-                await fsp.mkdir(commandsDir, { recursive: true });
-
-                await Promise.all(
-                    validAgents.map((agent) => {
-                        const mdc = toCursorMDC(agent, options);
-                        return fsp.writeFile(path.join(commandsDir, `${agent.slug}.mdc`), mdc);
-                    })
-                );
-            },
-            kilo: async (validAgents, options) => {
-                const targetDir = path.join(process.cwd(), '.kilocode', 'workflows');
-                await fsp.mkdir(targetDir, { recursive: true });
+            claude: async (validAgents, options) => {
+                const targetDir = path.join(process.cwd(), '.claude', 'commands', 'agents');
+                await fsp.mkdir(targetDir, { recursive: true });
+
+                await Promise.all(
+                    validAgents.map((agent) => {
+                        const md = toClaudeCommand(agent, options);
+                        return fsp.writeFile(path.join(targetDir, `${agent.slug}.md`), md);
+                    })
+                );
+            },
+            cursor: async (validAgents, options) => {
+                const commandsDir = path.join(process.cwd(), '.cursor', 'commands');
+                await fsp.mkdir(commandsDir, { recursive: true });
+
+                await Promise.all(
+                    validAgents.map((agent) => {
+                        const mdc = toCursorMDC(agent, options);
+                        return fsp.writeFile(path.join(commandsDir, `${agent.slug}.mdc`), mdc);
+                    })
+                );
+            },
+            kilo: async (validAgents, options) => {
+                const targetDir = path.join(process.cwd(), '.kilocode', 'workflows');
+                await fsp.mkdir(targetDir, { recursive: true });
 
                 await Promise.all(
                     validAgents.map((agent) => {
@@ -263,22 +263,22 @@ async function processAgentsInstallation(tools, options) {
                     })
                 );
             },
-            copilot: async (validAgents, options) => {
-                const githubDir = path.join(process.cwd(), '.github');
-                const promptsDir = path.join(githubDir, 'prompts');
-                await fsp.mkdir(promptsDir, { recursive: true });
-
-                await Promise.all(
-                    validAgents.map((agent) => {
-                        const md = toCopilotInstructions(agent, options);
-                        return fsp.writeFile(path.join(promptsDir, `${agent.slug}.md`), md);
-                    })
-                );
-
-                const mainAgent = validAgents.find((a) => a.slug.includes('coder')) || validAgents[0];
-                const mainInstructions = toCopilotInstructions(mainAgent, options);
-                await fsp.writeFile(path.join(githubDir, 'prompts.md'), mainInstructions);
-            },
+            copilot: async (validAgents, options) => {
+                const githubDir = path.join(process.cwd(), '.github');
+                const promptsDir = path.join(githubDir, 'prompts');
+                await fsp.mkdir(promptsDir, { recursive: true });
+
+                await Promise.all(
+                    validAgents.map((agent) => {
+                        const md = toCopilotInstructions(agent, options);
+                        return fsp.writeFile(path.join(promptsDir, `${agent.slug}.md`), md);
+                    })
+                );
+
+                const mainAgent = validAgents.find((a) => a.slug.includes('coder')) || validAgents[0];
+                const mainInstructions = toCopilotInstructions(mainAgent, options);
+                await fsp.writeFile(path.join(githubDir, 'prompts.md'), mainInstructions);
+            },
             trae: async (validAgents, options) => {
                 const traeDir = path.join(process.cwd(), '.trae');
                 await fsp.mkdir(traeDir, { recursive: true });
@@ -298,51 +298,51 @@ async function processAgentsInstallation(tools, options) {
                     })
                 );
             },
-            opencode: async (validAgents, options) => {
-                const targetDir = path.join(process.cwd(), '.opencode', 'commands');
-                await fsp.mkdir(targetDir, { recursive: true });
-
-                await Promise.all(
-                    validAgents.map((agent) => {
-                        const md = toOpenCodeAgent(agent, options);
-                        return fsp.writeFile(path.join(targetDir, `${agent.slug}.md`), md);
-                    })
-                );
-
-                // Generate AGENTS.md with interaction rules and agent location
-                const agentsMdPath = path.join(process.cwd(), 'AGENTS.md');
-                let agentsMdContent = `# Interaction Rules
-
-- Always respond to the user in the language they initially interact in; if they interact in English, respond in English, if they interact in Portuguese, respond in Portuguese.
-- If possible, display reasoning in the user's language as well.
-- Be didactic when explaining things, focus on providing complete responses and not just summaries.
-- Whenever possible, provide examples to illustrate concepts.
-
-# Allowed Commands
-
-- Never execute rm or rm -rf commands without confirming with the user.
-- Whenever possible, use more specific commands instead of generic ones.
-- Be cautious when using commands that may affect critical systems, such as shutdown or reboot.
-- For commands that may affect files or directories, always confirm with the user before executing.
-- Never execute commands that require administrative privileges (sudo, admin) without explicit permission from the user.
-- Avoid running background processes or daemons unless explicitly requested.
-- Be cautious when using commands that alter network settings, firewall configurations, or external connections.
-- Always quote file paths that contain spaces to avoid interpretation errors.
-- For package installation commands (npm install, pip install, etc.), confirm that the user has control over dependencies and versions.
-- Avoid irreversible git operations (such as force push or reset --hard) without confirmation.
-
-# Agent Location
-
-Custom agents are located in .opencode/commands/`;
-
-                let userRules = '';
-                if (options.globalRules && options.globalRules.trim()) {
-                    userRules = '\n\n# User Specified Rules\n\n' + options.globalRules.split('\n').filter(line => line.trim()).map(line => '- ' + line.trim()).join('\n');
-                }
-                agentsMdContent += userRules;
-
-                await fsp.writeFile(agentsMdPath, agentsMdContent);
-            }
+            opencode: async (validAgents, options) => {
+                const targetDir = path.join(process.cwd(), '.opencode', 'commands');
+                await fsp.mkdir(targetDir, { recursive: true });
+
+                await Promise.all(
+                    validAgents.map((agent) => {
+                        const md = toOpenCodeAgent(agent, options);
+                        return fsp.writeFile(path.join(targetDir, `${agent.slug}.md`), md);
+                    })
+                );
+
+                // Generate AGENTS.md with interaction rules and agent location
+                const agentsMdPath = path.join(process.cwd(), 'AGENTS.md');
+                let agentsMdContent = `# Interaction Rules
+
+- Always respond to the user in the language they initially interact in; if they interact in English, respond in English, if they interact in Portuguese, respond in Portuguese.
+- If possible, display reasoning in the user's language as well.
+- Be didactic when explaining things, focus on providing complete responses and not just summaries.
+- Whenever possible, provide examples to illustrate concepts.
+
+# Allowed Commands
+
+- Never execute rm or rm -rf commands without confirming with the user.
+- Whenever possible, use more specific commands instead of generic ones.
+- Be cautious when using commands that may affect critical systems, such as shutdown or reboot.
+- For commands that may affect files or directories, always confirm with the user before executing.
+- Never execute commands that require administrative privileges (sudo, admin) without explicit permission from the user.
+- Avoid running background processes or daemons unless explicitly requested.
+- Be cautious when using commands that alter network settings, firewall configurations, or external connections.
+- Always quote file paths that contain spaces to avoid interpretation errors.
+- For package installation commands (npm install, pip install, etc.), confirm that the user has control over dependencies and versions.
+- Avoid irreversible git operations (such as force push or reset --hard) without confirmation.
+
+# Agent Location
+
+Custom agents are located in .opencode/commands/`;
+
+                let userRules = '';
+                if (options.globalRules && options.globalRules.trim()) {
+                    userRules = '\n\n# User Specified Rules\n\n' + options.globalRules.split('\n').filter(line => line.trim()).map(line => '- ' + line.trim()).join('\n');
+                }
+                agentsMdContent += userRules;
+
+                await fsp.writeFile(agentsMdPath, agentsMdContent);
+            }
         };
 
         for (const tool of tools) {