Engineering Manager Agent

File: .github/agents/tsh-engineering-manager.agent.md

The Engineering Manager agent orchestrates the implementation phase. It does not write code itself — instead, it reads implementation plans and delegates tasks to specialized agents based on the type of work required. It ensures each task reaches the right agent with the right context, tracks progress, and triggers quality checks automatically.

Delegation Flow

Implementation Plan
        │
        ▼
┌──────────────────────┐
│  Engineering Manager  │ ← Reads plan, delegates tasks
└──────┬───────────────┘
       │
  ┌────┼────┬────┬────┬────┬────┐
  ▼    ▼    ▼    ▼    ▼    ▼    ▼
 SE   E2E  DevOps Arch  CR  UIR  PE

The Engineering Manager is bound to the /tsh-implement public prompt. When invoked, it parses the implementation plan, identifies individual tasks, and routes each one to the appropriate specialized agent. For UI tasks with Figma references, it uses the internal /tsh-implement-ui prompt to orchestrate the verification loop.

Agent Delegation

Agent	Handles	When Used
Software Engineer	Application code, features, bug fixes	Task involves writing or modifying application code
E2E Engineer	End-to-end tests	Task requires writing or updating E2E test suites
DevOps Engineer	Infrastructure, CI/CD, deployments	Task involves infrastructure changes or pipeline configuration
Architect	Codebase analysis, technical context	Before implementation starts, to establish codebase understanding
Code Reviewer	Code quality, best practices	After implementation completes, or when no review phase is defined
Prompt Engineer	LLM application prompts	Task involves designing, optimizing, or securing LLM prompts
UI Reviewer	Figma verification, visual correctness	After UI implementation, to verify against design specifications

How to Use

All delegation happens automatically when you run /tsh-implement. The Engineering Manager reads the implementation plan, identifies each task type, and routes it to the right agent with full context — plan details, acceptance criteria, and technical constraints are packaged automatically.

Tool Access

Tool	Usage
Atlassian	Gathers Jira tickets and Confluence documentation for task context
Sequential Thinking	Decides which agent to delegate to when the task type is ambiguous
Sub-agents	Invokes specialized agents for delegation
Terminal	Runs commands to verify environment state
File Read/Edit/Search	Navigates the codebase and implementation plan
Todo	Tracks task completion within the plan
VS Code Commands	Executes VS Code commands and asks user for clarification

Key Behaviors

• Delegates every task — Never writes code itself; all implementation is routed to specialized agents.
• Routes based on task type — Application code → Software Engineer, E2E tests → E2E Engineer, infrastructure → DevOps Engineer, LLM prompts → Prompt Engineer.
• Packages context automatically — Each delegation includes structured context from the implementation plan.
• Runs codebase analysis first — Invokes the Architect agent to establish technical context before starting implementation.
• Auto-triggers Code Reviewer — Automatically runs the Code Reviewer at the end of implementation if no review phase is defined in the plan.
• Tracks progress — Updates plan checkboxes after each completed task.
• Asks before deviating — Requests user confirmation before deviating from the implementation plan.
• Uses Sequential Thinking for ambiguous routing — When the task type doesn't clearly map to one agent, uses the Sequential Thinking tool to reason through the delegation decision.

Handoffs

The Engineering Manager receives work from:

• Architect — Provides the implementation plan that the Engineering Manager executes against.

After completing all tasks, the Engineering Manager can hand off to:

• Code Reviewer — For a final quality review of the full implementation (triggered automatically if not already planned).
• UI Reviewer — For Figma verification of UI changes.