Workflow Loader Skill

This skill provides the capability for any Gemini Agent to robustly load, read, and execute external workflow files defined in the mitra/agents/{agent}/workflows/ directory.

🛑 The Problem

By default, an instruction like "Action: Load backend.md" is ambiguous. The LLM might:

1. Hallucinate: Pretend it read the file and invent steps.
2. Fail: Not know where the file is located.
3. Guess: Use general knowledge instead of your specific process.

✅ The Solution

This skill provides strict Tools (Python scripts) that you MUST use when a workflow is requested.

🛠️ Usage Instructions

1. Listing Available Workflows

When a user asks "What can you do?" or you need to see available modules for an agent:

Run:

bash
1./.gemini/skills/workflow-loader/scripts/list_workflows.sh --agent {agent_name}

Example: ./.gemini/skills/workflow-loader/scripts/list_workflows.sh --agent architect

2. Loading a Workflow

When you trigger a menu handler (e.g., *backend) or need to execute a specific process:

Run:

bash
1./.gemini/skills/workflow-loader/scripts/load_workflow.sh --agent {agent_name} --workflow {workflow_name}

Example: ./.gemini/skills/workflow-loader/scripts/load_workflow.sh --agent architect --workflow backend

3. Execution Protocol

Once the content is loaded (printed to your context):

1. Acknowledge: "Workflow loaded: [Title]"
2. Step-by-Step: Execute the instructions in the markdown file sequentially.
3. Context: Treat the loaded content as the "Active Protocol" for the current turn.

⚠️ Critical Rules

• NEVER pretend to load a file. If you haven't run the load_workflow.sh script, you haven't loaded it.
• ALWAYS check the output of the script. If it says "File not found", inform the user.