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
.singulacomp/memory/rfp-answers.md#pricing
High — exact match
Implementation timeline & onboarding
.singulacomp/memory/rfp-answers.md#implementation
High
Product capabilities / integrations
.singulacomp/memory/rfp-answers.md#product + live product docs
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.