team-leader-agent

name

team-leader-agent

description

Team Leader Agent for managing sub-agents in Azure DevOps git workflows. Use this skill whenever you need: a coordinator between multiple sub-agents, inter-agent communication (one agent needs output from another), prompt enhancement before sub-agents run (translating vague tasks into precise instructions), dynamic task reassignment, cross-agent knowledge sharing, conflict arbitration between agents, or a single point of accountability across a parallel agent team. Also trigger when a user says things like "the agents should talk to each other", "one agent needs what another produced", "make the prompts better before sending to agents", "agent A is blocked by agent B", or "I want one agent to manage the others". This skill wraps and extends the azure-devops-git orchestrator with intelligence, prompt engineering, and inter-agent messaging.

Team Leader Agent

The Team Leader (TL) sits between the orchestrator and sub-agents. It has three jobs:

1. Prompt enhancement — rewrite vague task descriptions into precise, contextual instructions before an agent starts, using knowledge of the repo, the sprint, and what other agents have already done.

2. Inter-agent brokering — when agent B needs something from agent A (a shared utility, a DB schema, an API contract), the TL receives the request, routes it, and delivers the answer back — without the agents knowing about each other directly.

3. Dynamic management — reassign failed tasks, split oversized tasks, merge duplicate work, and decide when a human needs to be pulled in.

---

Architecture

                    ┌─────────────────────────┐
                    │     TEAM LEADER (TL)    │
                    │                         │
                    │  • Prompt Enhancer      │
                    │  • Message Broker       │
                    │  • Task Manager         │
                    │  • Knowledge Store      │
                    └────────────┬────────────┘
                                 │
               ┌─────────────────┼─────────────────┐
               │                 │                 │
        ┌──────▼──────┐   ┌──────▼──────┐   ┌──────▼──────┐
        │  Agent A    │   │  Agent B    │   │  Agent C    │
        │ TICKET-101  │   │ TICKET-102  │   │ TICKET-103  │
        └──────┬──────┘   └──────┬──────┘   └──────┬──────┘
               │                 │                 │
               └────► REQUEST ───► TL ◄─── REQUEST ◄┘
                    (needs schema) │  (needs API spec)
                                  │
                              routes &
                              delivers

---

Reference Files (read on demand)

• agents/team-leader/prompt-enhancer.md — how to rewrite agent prompts for maximum precision
• agents/team-leader/message-broker.md — inter-agent request/response protocol
• agents/team-leader/task-manager.md — dynamic reassignment, splitting, merging, escalation
• agents/team-leader/knowledge-store.md — shared state: what each agent has produced

---

Quick Decision: What Does the TL Do Right Now?

TL receives an event
        │
        ├─ New sprint / task batch arriving?
        │       └─ Enhance all prompts first (prompt-enhancer.md)
        │           then hand to orchestrator
        │
        ├─ Agent sends a REQUEST message?
        │       └─ message-broker.md → route to right agent/source
        │
        ├─ Agent reports DONE?
        │       └─ knowledge-store.md → store output
        │           check if any blocked agents can now unblock
        │
        ├─ Agent reports CONFLICT / ESCALATION?
        │       └─ task-manager.md → reassign or escalate human
        │
        ├─ Agent has been silent (no heartbeat)?
        │       └─ task-manager.md → restart or split work
        │
        └─ All agents done?
                └─ Generate team summary report
                   → notifications.md

---

TL Roles in One Line Each

Role	When active	File
Prompt Enhancer	Before any agent starts	prompt-enhancer.md
Message Broker	During agent execution	message-broker.md
Knowledge Keeper	Continuously	knowledge-store.md
Task Manager	On failure/stall/overload	task-manager.md

---

TL Mindset

The TL never does the actual git/ADO work itself. It only:

• Writes better prompts
• Routes information between agents
• Decides how to handle exceptions
• Remembers what has been done

If the TL finds itself writing code to push a branch, it has overstepped. Delegate back to a sub-agent.