Requirements cite real feedback
Each requirement is traceable to something a user said — feedback, a community thread, a support ticket. A requirement with nobody behind it gets flagged as an assumption rather than dressed up as a need.
ClawBot Crew · Product Agent
The Product Agent writes specs where each requirement points at something a real user actually said — a support ticket, a community thread, a piece of feedback. When it cannot find anyone who asked for a feature, it says so instead of writing it up in confident product language.
One iteration is one shippable slice. Every spec ships with a non-goals section, because scope creeps through the gaps you did not name.
It behaves like a product manager who keeps receipts.
Each requirement is traceable to something a user said — feedback, a community thread, a support ticket. A requirement with nobody behind it gets flagged as an assumption rather than dressed up as a need.
It cuts hard. A spec that cannot ship in one iteration gets split, and you are told what was left out and why, rather than handed a document nobody can finish.
Every PRD names what it is deliberately not doing. That section is where most scope creep gets caught, which is why it is not optional.
As feedback arrives it versions the document. When your coding agent hits an ambiguity mid-build, it answers and writes the decision back into the spec, so the PRD does not quietly become fiction.
From a pile of feedback to something a coding agent can actually execute.
A product idea plus whatever real user input you have — messages, tickets, threads, interview notes. Messy is fine; it is better than tidy and invented.
YouBoth are useful, but they are not the same thing, and specs go wrong when they get blurred. What users actually said becomes evidence; what you concluded becomes an assumption, labelled as one.
AgentRequirements with citations, a non-goals section, and an explicit cut list of what is waiting for a later iteration.
If you disagree with the cut, argue with it — that argument is much cheaper here than three weeks into a build.
AgentYour coding agent asks clarifying questions; this one answers and updates the spec. New feedback arriving mid-build gets versioned in rather than lost.
AgentBring evidence, get a spec.
The limits that keep the output honest.
If nobody asked for a feature, it says so. You can still build it — plenty of good products started as a founder's conviction — but it will be labelled an assumption, not evidence.
Ruthless scoping means being told your one iteration is actually three. That is the value, and it is also the part people find annoying.
It owns the spec, not the implementation. Pair it with the Coding Agent — tag them both in a chat and they hand work to each other.
Ten users saying "make it better" produces a spec that says ten users want it better. Evidence quality caps spec quality.
One subscription per crew agent, with its own monthly credit allowance.
Includes 1,500 credits per month for spec writing, versioning and answering build-time questions. Free trial available.
Crew agents are hired on top of a paid ClawBot plan. One base plan covers every crew agent you hire.
No — it works standalone, and plenty of people use it just to get a scoped spec they hand to their own team.
Together is where it gets interesting: the spec owner answers the builder's questions mid-build, and the document stays true to what was actually built.
Then most of the spec will be labelled assumptions, which is an accurate picture of where you are. Pair it with the GTM Agent to go get the evidence.
Yes. Paste it in and ask what has no evidence behind it, or what to cut to ship in a week. That is usually the fastest way to see how it thinks.
It versions as new feedback and build-time decisions come in. It is a living document by design, not a file that goes stale the day after it is written.
Each crew agent is its own subscription, with its own memory and its own credits. They can hand work to each other when you tag them in chat.
Not a tidy summary — the actual pile. The first thing worth seeing is which of your planned features turn out to have nobody behind them.