fix(miner): handle CHALLENGE_UNAVAILABLE without burning LLM retries

When the LLM verifier is temporarily down, skill-inscribe returns
CHALLENGE_UNAVAILABLE with a new challenge. Previously this code
entered the standard challenge retry loop (maxChallengeRetries=5),
which immediately called answerChallenge 5 times while the verifier
was still unavailable — wasting LLM calls and leaving miners stuck.

Fix: remove CHALLENGE_UNAVAILABLE from IsChallenge(). Instead, handle
it explicitly before the retry loop: save the new challenge to state
and return an error so the outer loop applies networkBackoff. On the
next cycle, the saved challenge is answered and submitted normally.

This prevents the cascading failure mode where all miners enter an
infinite loop of requesting new challenges without submitting answers.

Author: clawplaza

internal/api/types.go

@@ -180,7 +180,11 @@ func (r *InscribeResponse) ErrorMessage() string {
 func (r *InscribeResponse) IsChallenge() bool {
 	switch r.ErrorCode() {
 	case "CHALLENGE_REQUIRED", "CHALLENGE_FAILED", "CHALLENGE_EXPIRED",
-		"CHALLENGE_INVALID", "CHALLENGE_USED", "CHALLENGE_UNAVAILABLE":
+		"CHALLENGE_INVALID", "CHALLENGE_USED":
+		// Note: CHALLENGE_UNAVAILABLE is intentionally excluded.
+		// It means the LLM verifier is temporarily down — handled separately
+		// in mineOnce to save the challenge and back off, rather than burning
+		// LLM retries while the verifier is still unavailable.
 		return true
 	}
 	return false

internal/miner/loop.go

@@ -377,6 +377,21 @@ func (m *Miner) mineOnce(ctx context.Context) (*api.InscribeResponse, error) {
 		return nil, err
 	}
 
+	// CHALLENGE_UNAVAILABLE: verifier is temporarily down.
+	// Save the new challenge for the next cycle and back off immediately.
+	// Entering the retry loop would burn LLM calls while the verifier is still down.
+	if resp.ErrorCode() == "CHALLENGE_UNAVAILABLE" {
+		if ch := resp.GetChallenge(); ch != nil {
+			m.State.LastChallenge = ch
+			_ = m.State.Save()
+			slog.Info("verifier unavailable, challenge saved for next cycle", "id", shortID(ch.ID))
+		} else {
+			m.State.LastChallenge = nil
+			_ = m.State.Save()
+		}
+		return nil, fmt.Errorf("verification service temporarily unavailable, will retry next cycle")
+	}
+
 	// Challenge retry loop
 	for i := 0; resp.IsChallenge() && i < maxChallengeRetries; i++ {
 		challenge := resp.GetChallenge()