Scope Guard - ClosedLoop AI

Checks a spec against your customer evidence. Paste a spec, PRD or ticket and it reports which requirements customers actually raised, which nobody asked for, and what is missing.

/closedloop-scope-guard we're about to build tiered export permissions: <paste the spec>

Arguments

Argument	Required	Default
A spec, PRD, ticket, plain description, or a file path	yes

What you get

SCOPE GUARD: Tiered export permissions

YOUR SPEC, AS I READ IT
  1. Admins can restrict which fields appear in an export
  2. Admins can restrict who may run an export
  3. Export runs are logged for audit

  Correct the split before you read anything below it.

1. RESTRICT FIELDS                                       SHARPEN
   Matched: "exports leak internal pricing to resellers"
   "We had to stop resellers exporting at all because the sheet
    carries our margin column."  Contoso Retail, deal blocker

   Customers describe field-level control on ONE column, not a
   configurable set. Say margin, not fields.

2. RESTRICT WHO MAY RUN                                    SPLIT
   Matched two different mechanics that would be built and
   tested separately: turning export off per role, and
   requiring approval per run.

3. AUDIT LOG                                         NO EVIDENCE
   Nothing in the feedback describes this. That is not an
   argument against it. Compliance requirements do not arrive
   through support tickets.

MISSING FROM THE SPEC                                        ADD
   Scheduled exports keep running after someone loses access.
   Raised by several customers. No requirement covers it.

REJECTED MATCHES
   "Export timeout" theme: a performance problem, not a
   permissions one. Dropped, and its numbers with it.

The six verdicts

Verdict	What it means
Sharpen	The evidence describes the need more precisely than the requirement does.
Split	The evidence describes two mechanics that would be built and tested separately.
Merge	Customers describe two of your requirements as steps in one workflow.
Add	Customers raised a need in this area that no requirement covers.
No evidence	Nothing in the feedback describes this requirement.
Build as written	Covered, precise, nothing to change.

It never says “do not build this”

“No evidence” means the feedback is silent. Silence has causes:

A compliance obligation nobody files a support ticket about.
A platform bet.
Something nobody has been asked.

The build decision stays yours.

Numbers sit under theme names

One requirement matches several customer needs, and the themes overlap.
A requirement-level total would double-count customers who raised both, so there is no total to quote.
Reject a theme and its line drops, with nothing to recompute.
Rejected matches are printed, so you can disagree with a line rather than the report.