Clarify Agentbox names and invariants
Co-authored-by: codegirl007 <codegirl-007@users.noreply.github.com>
This commit is contained in:
co-authored by
codegirl007
parent
4d2fb4a733
commit
e702497996
@@ -76,6 +76,8 @@ func greenCentroidX(screenshot []byte) (float64, error) {
|
||||
for y := bounds.Min.Y; y < bounds.Max.Y; y++ {
|
||||
for x := bounds.Min.X; x < bounds.Max.X; x++ {
|
||||
red, green, blue, _ := renderedImage.At(x, y).RGBA()
|
||||
// Match the mover's #00ff66 square. Requiring 1,000 matching pixels
|
||||
// prevents a small green UI detail from passing verification.
|
||||
if green > 0xc000 && red < 0x4000 && blue < 0x8000 {
|
||||
xCoordinateSum += uint64(x)
|
||||
greenPixelCount++
|
||||
|
||||
@@ -87,12 +87,14 @@ func modelPrompt(task string, history []Step, observation Observation) (string,
|
||||
return "", err
|
||||
}
|
||||
return "You control an interactive Linux application from screenshots. " +
|
||||
"Choose exactly one safe input action, or mark done when the task is complete. " +
|
||||
"Choose exactly one allowed input action, or mark done when the task is complete. " +
|
||||
"Use logical key names such as RIGHT, ENTER, or ESCAPE. Keep waits under 5000 ms.\n" +
|
||||
string(data), nil
|
||||
}
|
||||
|
||||
func decisionSchema() map[string]any {
|
||||
// Strict structured output requires every action property to be present,
|
||||
// even when a particular action ignores most of them.
|
||||
actionProperties := map[string]any{
|
||||
"type": map[string]any{
|
||||
"type": "string",
|
||||
|
||||
Reference in New Issue
Block a user