add-core-subagent

Add Subagent to Core Orchestrator

Overview

Automates wiring a new subagent into the core dispatcher. Core is a slim orchestrator that routes tasks to specialized subagents -- it never does the work itself. Each subagent needs a dispatch rule (so core knows when to route) and a recipe registration (so ruly generates the agent file).

When to Use

• "Add X as a core subagent"
• "Wire up the Y subagent to core"
• "Core should dispatch to Z"
• "Register W in the core recipe"
• Creating a new specialized agent that core should delegate to

Gather Information

Before creating anything, collect these from the user (ask if not provided):

Input	Example	Required
Subagent name	qa	Yes
Recipe name (if different)	qa-testing	Only if differs from name
Work type heading	QA & Testing	Yes
Trigger conditions	Test planning, QA reports, test execution	Yes
Has slash commands?	yes/no	Yes
Capabilities summary	test planning, execution, QA reports	Yes

Naming convention:

• subagent_type in Task tool uses underscores: core_debugger, context_fetcher
• recipe name uses hyphens: core-debugger, context-fetcher
• Dispatch rule filename uses hyphens: use-core-debugger.md
• Simple names are the same everywhere: bug, comms, merger

Step 1: Create the Dispatch Rule

Create the file at:

/Users/patrick/Projects/ruly/rules/workaxle/core/standards/use-{name}.md

Use this template, filling in the placeholders:

---
description: Forces {task_type} tasks to dispatch {subagent_type} subagent for {capabilities}
alwaysApply: true
---
# {Work Type}: Dispatch to Subagent

## When This Applies

Any time you are asked to or encounter:
{trigger_list -- bullet points of actions, keywords, slash commands}

## The Rule

**You MUST dispatch the `{subagent_type}` subagent.** Do NOT:
{do_not_list -- bullet points of what core must not do itself}

The subagent has {capabilities_description}. You are a dispatcher -- route the work.

## CRITICAL: Slash Command Override

> **Only include this section if the subagent has slash commands.**

When a {work_type_lower} slash command is invoked, the command file content will load into your context. **Do NOT follow those instructions yourself.** Instead:

1. Extract the user's intent and any arguments from the command
2. Dispatch the `{subagent_type}` subagent with those details
3. The subagent will follow the command workflow (it has the same files)

## How to Dispatch

{dispatch_examples -- one or more ### subsections with Task tool examples like:}

### For {Action Name}

```
Task tool:
  subagent_type: "{subagent_type}"
  prompt: |
    {Action description}:
    - {relevant parameters}
```

## After Subagent Returns

1. Present the subagent's results to the user
2. If the subagent needs confirmation, relay to user, then dispatch again
3. If additional work needed, dispatch again with updated context

## Does NOT Apply When

{boundary_conditions -- bullet list deferring to other subagents}

## Red Flags -- You Are Violating This Rule

{red_flags -- bullet list of anti-patterns}

**All of these mean: STOP. Dispatch `{subagent_type}`.**