How it works

Inputs in, an ordered run, a result out.

A Sequential Agent takes typed inputs, runs its tasks top to bottom against your tools and tables, and returns a structured result, the same way on every run.

Typed inputs

Values you pass in, validated before the run spends anything.

2 tasks

Each task runs in order, passing its output to the next.

Runs standalone

No external tools or tables needed.

Result + trace

A structured output, with every step recorded.

The run loop

Read→Plan→Act→Check→Write→Tracerepeats until done

What's inside

Everything this template installs, by name.

Tools & tables

Runs standalone, no external tools or tables required

Tasks

2 ordered tasks

What it takes in

The inputs you provide on each run, validated before anything executes.

InputTypeRequired

Proposal Or Quote To ReviewproposalOrQuoteLong textRequired

Deal ContextdealContextLong textOptional

Pricing And Discount RulespricingRulesLong textOptional

The work, step by step

2 tasks, run in order.

Each task passes its result to the next. Here's what each does and what it produces.

01
Review for issues. Examine {{proposalOrQuote}} (using {{dealContext}} for background) and check for: pricing or math errors, missing or unclear terms, scope ambiguity, risky or open-ended commitments, and any violations of {{pricingRules}} if provided. If no pricing rules are given, apply general good-practice checks and note that assumption. List every issue found. Present readably, never JSON.
Expected output
A readable issues list using headings and bullets, no JSON and no code blocks, grouped into: Pricing and Math, Terms and Scope, Risk and Commitments, and Policy Compliance. For each issue, state what is wrong. If a group has no issues, write None found.
02
Prioritize and recommend. Using the Task 1 findings, separate must-fix issues from optional improvements, give a suggested fix for each must-fix issue, and provide an overall recommendation of Approve, Revise, or Escalate with a one-paragraph rationale. Present readably, never JSON.
Expected output
A readable section with no JSON and no code blocks: Must-Fix Before Sending (bullets, each with the issue and a suggested fix), Optional Improvements (bullets), and Recommendation (Approve, Revise, or Escalate, with a short rationale).