remindb: Skill for Claude Code

Add a fuzz target

remindb fuzzes parser, query, transformer, compiler, and temperature code. The fuzz harness is whatever Go's testing.F gives you — there's no project-specific framework — but the seed-corpus discipline is project convention worth getting right. scripts/fuzz.sh auto-discovers any Fuzz* function via go test -list='^Fuzz', so naming your function FuzzXxx is the only registration needed.

Where it lands

Two files at most.

If pkg/<package>/fuzz_test.go already exists (it does for parser, query, transformer, compiler, temperature), append; don't make a second file.

The function shape

Mirror pkg/parser/fuzz_test.go and pkg/temperature/fuzz_test.go. The shape is uniform:

func FuzzExample(f *testing.F) {
    // Seed corpus — see "Seed selection" below.
    f.Add(input1, input2)
    // ... more f.Add lines, each one shape ...

    f.Fuzz(func(t *testing.T, input1 T1, input2 T2) {
        result, err := YourFunc(input1, input2)

        // Invariants — see "Invariant assertions" below.
        if err != nil {
            return        // errors are fine; panics are not
        }
        if !invariantHolds(result) {
            t.Errorf("invariant violated: ...")
        }
    })
}

Two rules:

• Function name FuzzXxx. scripts/fuzz.sh greps for ^Fuzz in go test -list output. Anything else is invisible.
• One target per logical surface. Don't multiplex two unrelated functions into one fuzz target — Go's fuzzer mutates the input tuple as a unit, so combined targets dilute coverage.

Seed selection — the discipline

A fuzz seed says "this is a shape worth starting from." The fuzzer mutates from there. The point isn't to enumerate all valid inputs; it's to give the mutator a head start on every category of structural variation. Aim for one seed per shape: