Reactive and knowledge-base-grounded; the agent drafts and flags, it never decides what gets submitted to the buyer. This is a general sales RFP/proposal skill — pricing, implementation, integrations, support, company background, references. It is distinct from questionnaire-response, which is scoped to security-specific questionnaires (SIG, CAIQ).

Step 0 — Find what's new

text

# List messages in the RFP label/folder, newest first.
gmail.messages.list(label={{rfp_label}}, order="newest")

# Check each thread for an existing draft reply — a thread that already has
# one has already been worked.
gmail.drafts.list(thread=message.thread_id)

There is no ledger — this is a fresh session per RFP. The Gmail thread itself is the record of what's already been drafted. A thread with no draft is new; work the oldest unhandled one first, respecting any stated submission deadline.

Step 1 — Parse the RFP into individual questions

Pull the attachment or linked file from the thread (Word doc, PDF, or spreadsheet) and open it from google_drive.
Read the whole document — some RFPs spread sections (pricing, technical, implementation, references, …) across separate tabs or headings.
Extract each row or item as a discrete question, keeping its section, item number, and exact wording. Note the source format and structure so the draft can mirror it.

Step 2 — Match each question to the vetted library

Work question by question against our approved past answers and product docs, carried as skills and memory until the team updates them:

Question topic	Vetted source	Typical confidence
Pricing tiers / packaging	`memory/rfp-answers.md#pricing`	High — exact match
Implementation timeline & onboarding	`memory/rfp-answers.md#implementation`	High
Product capabilities / integrations	`memory/rfp-answers.md#product` + live product docs	High — verify against current docs
Support model & SLAs	`memory/rfp-answers.md#support`	High
Company background & references	`memory/rfp-answers.md#company`	High
Security or compliance basics	`memory/rfp-answers.md#security-basics`	Medium — a full security review belongs to the `questionnaire-response` skill, not this one
Anything with no matching entry	—	Flag for a person

A "confident match" means the question maps clearly to one vetted entry, and for product-capability questions, the vetted entry still matches the current product docs. If a question is a close paraphrase of a vetted one, use the vetted answer as written; if it combines two topics, compose from the matching entries rather than inventing new language. If the library has no entry for a topic, that is a flag, not an opening to reason from first principles.

Step 3 — Draft the response in a new Google Doc

Create a new Google Doc inside {{drafts_folder}}, titled with the buyer/RFP name and date. Mirror the RFP's own question order and section headings so the reviewer can follow it next to the source document. Write each matched answer in full under its question, using the vetted wording — adapt only for length or tone the document calls for, never the substance.

Step 4 — Flag low-confidence questions

For every question from Step 2 with no confident match, mark it clearly in the draft (a highlighted heading, a [NEEDS REVIEW] tag, or a comment) and add it to a running list at the top of the doc: item reference, the question text, and why it isn't answered from the vetted library.

Step 5 — Leave a draft reply, never send

Reply on the original Gmail thread with a draft (never sent) linking the new Google Doc and summarizing what was answered and what's flagged. The agent never calls send and never uploads anything to a buyer portal.

Step 6 — Stop and hand off

The run ends here. A person opens the linked Google Doc, reviews every answer, resolves the flagged questions, and submits the RFP through whatever channel the buyer requires — portal, email, or otherwise. That submission never happens in this session.

RFP response