- Home
- Skills
- Product Management
- BA Squad: Lead Business Analyst with a Team of Agents
More screenshots
Works with the AI Agents you already use
BA Squad: Lead Business Analyst with a Team of Agents
A spec written by one AI in one pass fails in two predictable ways.
$3
BA Squad: Lead Business Analyst with a Team of Agents
Example session with this skill installed
Use ba-squad. We are rebuilding the online store for Maison Lune (fashion). Below are my notes from the client meeting and the returns text on their live site. Write a functional spec for the Returns and refunds module that dev and QA can start from, and tell me what I need the client to decide.
Meeting notes, 12 Mar
- Customers request a return from Order history, within 30 days of delivery.
- Sale items are final sale. No returns.
- CS approves or rejects each request. Warehouse inspects the parcel when it arrives.
- Refund goes back to the original payment method. COD orders: "we'll figure it out".
- Customer pays return shipping unless the item is faulty.
Returns policy page (live site)
"Returns are accepted within 14 days of receiving your order. Refunds are processed quickly after we receive your item."
- Read your context and instructions
- Compiled the ba squad
- Generated the data export
Squad plan — functional spec for Returns and refunds, for dev and QA to start from
Domain: e-commerce and retail · online fashion store · market [TBD] (pack: seed) · Size: Pair
| Wave | Stage | Team | Output |
|---|---|---|---|
| 1 | Intake | lead reads both sources (each under 3 pages) | project.md, glossary.md |
| 2 | Analysis and spec | domain expert, lead as spec writer | spec/01-returns-and-refunds.md |
| 3 | Gate | QA reviewer, fresh eyes | review/01-qa-review.md |
One module and two short sources is a Pair-sized job, so I am not staffing six agents on it.
Squad report — Returns and refunds functional spec
Delivered:
ba-work/spec/01-returns-and-refunds.md— 3 user stories, 16 acceptance criteria, 4 business rules, 5 messages, status table, flow diagramba-work/spec/rtm.md— traceability matrixba-work/registers.md— 8 open questions (3 blocking), 3 assumptionsba-work/review/01-qa-review.md— 4 findings; the 1 blocker is fixedba-work/backlog/backlog.csv— 1 epic and 3 stories, ready for a tracker's CSV import
Team: domain expert × 1, QA reviewer × 1, lead as spec writer · sequential hats (no sub-agents on this host)
Lint: 6 files · 0 errors · 0 warnings
Decisions needed (blocking):
| ID | Question | Why it matters | Options |
|---|---|---|---|
| OQ-001 | Is the return window 30 days (meeting notes) or 14 days (live policy page), and does the last day count? | The two sources disagree; BR-101 cannot be built or tested | 14 days / 30 days / 30 days and update the policy page |
| OQ-102 | How is a COD order refunded, given there is no original payment method to refund to? | The notes say "we'll figure it out"; COD orders have no refund path | bank transfer / store credit / COD orders not refundable online |
| OQ-106 | What happens when a parcel fails inspection? | The notes cover only the accepted case; the flow has a dead end | send back to the customer / partial refund / reject with no refund |
Open questions (important):
- OQ-103: Within how many days must the refund be completed? The policy page says "quickly", which cannot be tested.
- OQ-104: How does the customer bear the return shipping cost: deducted from the refund, paid to the carrier, or a prepaid label charged back?
- OQ-105: What makes a line a sale item: a markdown price at order time, a product flag, or also any voucher order?
- OQ-107: When part of an order bought with a promotion is returned, is the discount recalculated? Not mentioned by the client; a usual trap in this industry, so raised as a question, not written as a rule.
Assumptions: 3; riskiest: AS-102 (one request covers one order and allows a partial quantity per line) and AS-103 ("faulty" is the reason the customer selects, confirmed at inspection).
Not covered: exchanges and cancellation before shipping (out of scope); message texts (OQ-108); UAT scenarios; consumer-law rules on returns, because no market is named yet.
Next step: send the three blocking questions to the client. US-101 and US-102 are ready to estimate. US-103 is not ready to baseline: three of its six criteria wait on answers.
From spec/01-returns-and-refunds.md
US-101 — Request a return
Epic: EP-01 · Source: meeting notes lines 1–2; policy page sentence 1
As a customer, I want to request a return from my order history, so that I can send back an item I do not want to keep.
Acceptance criteria
- AC1 Show the return action in Order history only for an order that has been delivered.
AC2 Require a return reason and, per order line, a quantity from 1 to the delivered quantity [ASSUMPTION: AS-102].
- AC3 Refuse the request and show MSG-101 when the return window has ended (BR-101).
- AC4 Refuse a line and show MSG-102 when the line is a sale item (BR-102).
AC5 Create the return request in status Requested and show MSG-103 when every check passes [ASSUMPTION: AS-101].
Business rules
| ID | Rule | Condition | Outcome | Applies to | Source |
|---|---|---|---|---|---|
| BR-101 | Return window | request date is within [TBD: OQ-001] days of delivery | request allowed; otherwise refused | US-101 | notes line 1 says 30 days; policy page says 14 days |
| BR-102 | Final sale | the line is a sale item [TBD: OQ-105] | line must not be returned | US-101 | notes line 2 |
| BR-103 | Refund method | inspection accepted | refund goes to the original payment method | US-103 | notes line 4 |
| BR-104 | Return shipping cost | return reason is not "faulty" [ASSUMPTION: AS-103] | customer bears the return shipping cost [TBD: OQ-104] | US-103 | notes line 5 |
From spec/rtm.md (generated by ba_tool.py trace)
| Story | Title | AC | Rules | Open items | Gaps |
|---|---|---|---|---|---|
| US-101 | Request a return | 5 | BR-101, BR-102 | AS-101, AS-102, OQ-001, OQ-105 | — |
| US-102 | Decide a return request | 5 | — | AS-101 | — |
| US-103 | Inspect the parcel and refund | 6 | BR-103, BR-104 | AS-103, OQ-102, OQ-103, OQ-104, OQ-106, OQ-107 | — |
From review/01-qa-review.md
| ID | Severity | Where | Finding |
|---|---|---|---|
| RV-001 | blocker, fixed | US-102 | No criterion stopped a second decision on a request already approved or rejected. Added as AC4. |
| RV-003 | major | US-103 | Three of six criteria are [TBD]; the story cannot be estimated or tested as written. |
backlog.csv
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
A spec written by one AI in one pass fails in two predictable ways. Holes nobody looked for: the happy path is there, the reversal is not. And drift: the same status under three names by page forty. Then someone builds it.
BA Squad makes your agent work the way a senior BA runs a team. It cuts the job into slices, briefs a specialist on each one, runs them in parallel, and then does the part most multi-agent prompts skip: it merges the results into one vocabulary, checks the quotes, lints the requirements and puts a fresh-eyes reviewer on whatever leaves the workspace.
What is in the pack
- 1 lead-BA skill with a squad protocol: how to size a team, brief an agent, merge, gate and report
- 10 agent role briefs: source reader, domain researcher, domain expert, analyst (process, data, integration, experience, quality lenses), spec writer, reviewer (dev, QA, business, user, consistency lenses), stakeholder persona, impact tracer, backlog slicer, test designer
- 7 plays, one per kind of BA job: elicitation, requirements, review, change impact, backlog, UAT, domain expertise
- 8 seed domain packs: banking and fintech, e-commerce and retail, logistics and supply chain, healthcare, insurance, HR and payroll, education, manufacturing. Each has a glossary, actors, core processes with the exceptions that always exist, entity lifecycles, rule patterns and regulations to verify. Any other industry gets a pack built on the spot.
ba_tool.py: one Python file, standard library only, no network access. It scaffolds the workspace, hands out ID blocks so parallel agents never collide, lints requirements, builds the traceability matrix and exports the backlog as CSV.
What it delivers
| You ask for | It runs | You get |
|---|---|---|
| "Analyse these documents and write the SRS" | Intake → Domain → Analysis and spec → Review | source inventories, a conflict log, module specs, traceability matrix |
| "New project, little on paper" | Domain → Discovery | a ranked question bank and a rehearsed workshop |
| "Scope this RFP" | Intake → Domain → Backlog → Discovery | feature list, sizing drivers, assumptions, exclusions, clarification log |
| "The client changed X" | Change | every story, rule, message and test the change touches |
| "Is this spec ready?" | Review → UAT | findings by reader (dev, QA, business, user) and UAT scenarios traced to acceptance criteria |
Everything lands as Markdown in one ba-work/ folder that another BA could pick up tomorrow.
Why you can trust the output
- Nothing is invented. Every statement is a fact with a source, an assumption marked
[ASSUMPTION: AS-nnn], or a gap marked[TBD: OQ-nnn]. Names, numbers, SLAs and legal requirements are never made up. - Conflicts go on the table. When two documents disagree, it quotes both and raises a question with options. It never quietly picks the newer one.
- Industry knowledge stays a hypothesis. What "shops usually do" enters the spec as a question, never as a requirement.
- Ten questions, not eighty. Open questions are de-duplicated and ranked blocking, important or detail, so the stakeholder actually answers them.
- It says "not ready". If a module's core behaviour is undecided, you get the analysis and the clarification log instead of a confident spec built on a guess.
- It right-sizes. One story is solo work, and it will tell you so. It asks before any job that needs more than eight agents.
Who it is for
Business analysts, product owners, presales and delivery leads, and solo founders who receive a pile of documents and owe someone a specification.
Works in English and Vietnamese out of the box, including the lint word lists and the [GIẢ ĐỊNH] / [CẦN LÀM RÕ] markers. On a host without sub-agents the same roles run one at a time, with the same checks.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Product Management.
- Deliverables are Markdown. Word or PDF output needs a separate document skill or tool on your agent.
- It does not connect to Jira or Confluence. The backlog is exported as a CSV that you import yourself.
- Parallel agents need a host with sub-agents, such as Claude Code. Elsewhere the same roles run one at a time: same checks, slower.
- A squad job uses more tokens than a single prompt. The skill sizes the team to the job and asks before any plan that needs more than eight agents.
- Domain packs are general industry practice, written jurisdiction-neutral. They are not legal or compliance advice: regulations are listed for you to verify with the client.
- Scanned or password-protected sources cannot be read. PDF and Word intake depends on what your agent can open.
- The lint word lists cover English and Vietnamese only.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 6 days ago
- Software
- AI agent that can read and write files
- Optional
Python 3.8+ (lint, traceability matrix, backlog CSV)Sub-agent support (parallel squads)Web search (researching an industry without a seed pack)
Without Python the agent applies the same lint and traceability checks by hand. Nothing else to install.
Permissions
File Scopes
Reads the source documents you point it to and writes Markdown into a ba-work/ folder in your project. The terminal is used only to run the bundled ba_tool.py, which uses the Python standard library and makes no network calls. Nothing is sent to Jira, Confluence, email or chat without your yes for that specific action.