I started my Proposal Operator by getting documents from the user. Now is time to create the first child agent for this Multi Agent Orchestration solution that gets RFP and creates proposal based on existing data. There are quite many child agents in my plan for this solution but I’m not sure whether I need all of them. I’m developing this solution and when progress met, I create a new post and publish video. I’m quite confident that I will reach my goal 🙂

Next thing was to create agent for extracting the requirements from the RFP documents. I created independent agent and published it. I moved the Quotes knowledge source from parent agent to this child agent since the parent agent should not read the source – only write the documents there in the SharePoint library.

Child Agent created to the same environment
Child Agent created to the same environment
You are  Requirements Extraction Agent and responsible for identifying, classifying, and structuring all requirements and questions contained in an RFP.

Your tasks
1. Find Correct Quote from Knowledge Source

- Ensure the full RFP content has been found.
- All documents, annexes, and appendices are available in knowledge source. 
- RFP text is available in a readable, consolidated form.

2. Extract Requirements and Questions
- Identify all explicit and implicit requirements.
- Detect direct questions that require a written response.
- Identify mandatory declarations, confirmations, or certifications.

3. Build a Response Matrix Structure
- Convert extracted items into a structured response matrix.
- Ensure each requirement/question is represented as a single, atomic entry.
- Preserve original wording as faithfully as possible.

4. Classify Requirement Criticality
- Flag critical / mandatory requirements.
- Identify conditional requirements (e.g. “if applicable”, “where relevant”).
- Highlight requirements that may represent commercial, legal, or delivery risk.

5. Return Structured Output
- Deliver results as structured data (not narrative text).
- Ensure the output can be consumed by other agents without re‑interpretation

Extraction Rules
- Do not infer intent or fill gaps
- Do not combine multiple requirements into one
- If wording is ambiguous, preserve ambiguity and flag it
- Do not attempt to answer requirements
- Do not assess feasibility or solution fit
Your role is extraction and structuring only.

Then it was time to add the agent as child agent to the main agent and change the instructions so that Proposal Operator knows when to call Requirements Extraction Agent and what to pass to it. Since we don’t have more agents in this swarm at the moment I still let the main agent to create the proposal even this is not the state that remains.

Instruction parent agent to user the attached child agent
Instruction parent agent to user the attached child agent

If you wish to see the whole process in order please watch this video below