Agent Skills' code review skill now drops findings your code already rules out

In Addy Osmani's Agent Skills, the code review and quality skill now checks each finding against the callers, guards and tests that reach the changed code, and drops it when they rule it out. So a review stops flagging bugs your code already prevents.

In Step 4, before any finding is reported, the reviewer now checks the callers, guards and tests. Bugs already in the base stay apart from new regressions. And zero findings is a valid answer.

### Step 4: Categorize Findings

Before reporting a finding, check it against the callers, guards and tests that reach the changed code, and drop it if they rule it out.
Label defects that already exist in the base separately from regressions this change introduces.
Zero findings is a valid result; do not invent one.

One new test hands the reviewer a claim: the lookup function has no owner check, so anyone can withdraw. But its only caller refuses anyone who isn't the owner. So the claim is rejected.

C: lookup has no owner check, so any actor can withdraw.

def withdraw(actor, owner, amount, accounts):
    if actor != owner:
        raise PermissionError("owner required")
    balance = lookup(owner, accounts)

In the same test, less than or equal becomes less than, so withdrawing your whole balance now fails. That bug is new. The fee was already wrong before the change, so it is labelled pre-existing.

A third test rewrites the check without changing what it allows. The expected review reports zero new defects and invents no blocker.

    return 0 < amount <= balance
    return amount > 0 and amount <= balance

It merged November first, the day after release 0.6.15, so it is on main, not yet in a tagged release, and the review command gets it too. No graded model run was made, so the gain is unmeasured.

False alarms are what agent reviews cost you most. I would take this rule. A review that checks the caller and can report nothing is one you can believe when it does flag something.