The Manager Briefing Agent

Relevant source files

The following files were used as context for generating this wiki page:

• agents/manager_briefing_agent.md
• app.py
• orchestrator.py
• README.md
• agents/triage_agent.md
• memory_utils.py

The Manager Briefing Agent serves as the internal communications hub for the BrightPath AI Operations System. Its primary purpose is to synthesize technical outputs from preceding agents into clean, scannable summaries designed for the business owner or manager. It acts as the final stage of the message processing pipeline, ensuring that automated actions are transparent and that high-priority items requiring human intervention are surfaced immediately.

Operating primarily through the Claude Haiku model for efficiency, the agent extracts the most relevant business data from inbound events and specialist agent actions. It is responsible for flagging revenue opportunities, risks, and escalations, specifically formatting its primary output for delivery via Telegram.
Sources: agents/manager_briefing_agent.md:3-10, README.md:10-23, app.py:175-176

Architecture and Pipeline Integration

The Manager Briefing Agent is the third step in a three-agent sequential pipeline. It consumes the context generated by the Triage Agent and either the Sales or Customer Service specialist agents.

The diagram shows how the Manager Briefing Agent sits at the end of the pipeline, receiving inputs from all previous stages to generate the final summary.
Sources: README.md:12-18, app.py:321-364, orchestrator.py:236-285

Data Input and Processing

The agent receives three primary pieces of information to construct its briefing:

1. The Original Message: The raw inbound text from the customer.
2. Triage Metadata: Classification, urgency, sentiment, and flags (e.g., vip, upsell_opportunity).
3. Agent Results: The draft response and internal actions taken by the specialist agent.

In specialized "Admin" scenarios (such as a manager requesting a daily summary), the orchestrator bypasses specialist agents and provides the Briefing Agent with a "Complete Operational Data" block. This context includes open tasks from the database, recent pipeline events, the decision log, and sales notes.
Sources: app.py:355-359, orchestrator.py:246-255, orchestrator.py:110-141

Technical Specifications and Logic

Output Schema

The agent is constrained to return a strictly structured JSON object. This ensures that the system can programmatically handle flags for UI rendering (e.g., action banners in the Streamlit dashboard) and Telegram bot transmission.

Field	Type	Description
headline	string	One line summary of the event.
urgency	string	high, medium, or low.
what_happened	string	2-3 sentence plain English summary.
needs_human_action	boolean	Indicates if the manager must intervene.
human_action_required	string	Specific instruction for the manager, or null.
telegram_summary	string	Short, punchy message (under 120 words).
value_flag	string	Description of potential revenue or null.
risk_flag	string	Description of risk/concern or null.

Sources: agents/manager_briefing_agent.md:19-35

Messaging Principles and Emojis

The agent follows specific "Briefing Principles" to ensure clarity for a busy manager. It is instructed to use plain English and specific emojis to categorize the Telegram output:

• 💰 Sales/Revenue: Used for new leads or upsell opportunities.
• ✅ Resolved: Used for successful customer service completions.
• ⚠️ Attention: Used for items requiring manual review.
• 🚨 Escalation: Used for hostile sentiment or high-risk issues.
• 📋 Admin: Used for reporting or task summaries.

Sources: agents/manager_briefing_agent.md:12-18, agents/manager_briefing_agent.md:37-45

Implementation in the Orchestrator

The system uses two distinct functions to invoke the agent depending on the request type:

1. run_briefing (Standard Pipeline)

This function handles standard customer interactions. It passes the current message and the immediate results from the specialist agent.
Sources: app.py:301-304, orchestrator.py:218-228

2. run_briefing_admin (Context-Rich Briefing)

Used for requests like "Show me everything urgent today." This implementation pulls data from multiple sources to provide a comprehensive overview.

def run_briefing_admin(message, admin_context):
    """Specialised briefing for admin/manager requests — full DB + memory context."""
    system = load_agent('manager_briefing_agent') + """
    CRITICAL: This is a manager daily briefing request. You have the COMPLETE operational data below.
    Use it to produce a specific detailed briefing with real names, amounts and deadlines.
    """
    user = f"Manager request: {message}\n\nCOMPLETE OPERATIONAL DATA:\n{admin_context}"
    return call_claude(system, user, fast=False, max_tokens=3000)

Sources: app.py:286-294, orchestrator.py:231-243

Summary of Operational Impact

The Manager Briefing Agent bridges the gap between autonomous AI operations and human oversight. By synthesizing database tasks, CRM logs, and agent decisions into a 120-word Telegram summary, it allows the business owner to monitor the system without reviewing individual logs. It specifically monitors for high-value signals (e.g., 10+ people mentioned in triage_agent.md) and high-risk signals (e.g., hostile sentiment) to ensure critical business events are never missed.
Sources: agents/manager_briefing_agent.md:17-18, agents/triage_agent.md:47-49, README.md:3-8