Open agent skill

Reject weak ideas before they become products.

A hard evidence check for the moment before the feature list begins.

progress viewfixed example
IDEA IN REVIEWCourier settlement mismatch

A seller loses money every week because courier settlement reports disagree with delivered orders.

01IncidentConcrete money loss
02FilterChecks failed workaround
03VerdictValidate first
VERDICT
Insufficient evidence

The feature is clear, but the person, incident, workaround, and consequence are missing.

AI roleNot yet determined

Start with the incident.

What happened to the person this week?

Proof, not applause

Different evidence. Different verdict.

The filter can reject an attractive idea, ask for proof, or give a narrow build candidate. These are examples from the product's own evaluation record.

RejectWaiting trackerExisting category, weak wedge
ValidateParcelCheckConcrete cost, seller pull unproven
No evidenceIcon packFormat idea without a user problem

The filter

Buildability is not proof.

Problem Filter checks whether a real person is already paying for the problem with time, money, risk, or repeated work.

PainRecurrenceFailureUser pullCompetitive gapOutcomeBuild wedgeTrustExpansion

Works with your agent

One skill. The agent you already use.

Native skill folders where supported. A ready-to-copy adapter for Cursor.

Open by design

Read it before you trust it.

No hidden service sits behind the verdict. Share the repository with your agent, inspect the instructions, or read the proof.

problem-filter/
├── SKILL.mdcore instructions
├── TEST-REPORT.mdbehaviour proof
├── COMPATIBILITY.mdagent setup
├── adapters/cursor/Cursor formats
└── site/this demo

Use the product

Give your agent a reason to disagree.

Open the repository