Merge pull request #1 from thepeterstone/claude/local-oss-model-agents-MEBqj

Local OSS models as a third runner (epic)

1 files changed, 3 insertions, 77 deletions

diff --git a/internal/executor/ratelimit.go b/internal/executor/ratelimit.go
index 1f38a6d..109aa49 100644
--- a/internal/executor/ratelimit.go
+++ b/internal/executor/ratelimit.go
@@ -1,33 +1,9 @@
 package executor
 
-import (
-	"context"
-	"fmt"
-	"regexp"
-	"strconv"
-	"strings"
-	"time"
-)
+import "strings"
 
-var retryAfterRe = regexp.MustCompile(`(?i)retry[-_ ]after[:\s]+(\d+)`)
-
-const maxBackoffDelay = 5 * time.Minute
-
-// isRateLimitError returns true if err looks like a transient Claude API
-// rate-limit that is worth retrying (e.g. per-minute/per-request throttle).
-func isRateLimitError(err error) bool {
-	if err == nil {
-		return false
-	}
-	msg := strings.ToLower(err.Error())
-	return strings.Contains(msg, "rate limit") ||
-		strings.Contains(msg, "too many requests") ||
-		strings.Contains(msg, "429") ||
-		strings.Contains(msg, "overloaded")
-}
-
-// isQuotaExhausted returns true if err indicates the 5-hour usage quota is
-// fully exhausted. Unlike transient rate limits, these should not be retried.
+// isQuotaExhausted returns true if err indicates the 5-hour Claude usage quota
+// is fully exhausted. Unlike transient rate limits, these should not be retried.
 func isQuotaExhausted(err error) bool {
 	if err == nil {
 		return false
@@ -39,53 +15,3 @@ func isQuotaExhausted(err error) bool {
 		strings.Contains(msg, "rate limit reached (rejected)") ||
 		strings.Contains(msg, "status: rejected")
 }
-
-// parseRetryAfter extracts a Retry-After duration from an error message.
-// Returns 0 if no retry-after value is found.
-func parseRetryAfter(msg string) time.Duration {
-	m := retryAfterRe.FindStringSubmatch(msg)
-	if m == nil {
-		return 0
-	}
-	secs, err := strconv.Atoi(m[1])
-	if err != nil || secs <= 0 {
-		return 0
-	}
-	return time.Duration(secs) * time.Second
-}
-
-// runWithBackoff calls fn repeatedly on rate-limit errors, using exponential backoff.
-// maxRetries is the max number of retry attempts (not counting the initial call).
-// baseDelay is the initial backoff duration (doubled each retry).
-func runWithBackoff(ctx context.Context, maxRetries int, baseDelay time.Duration, fn func() error) error {
-	var lastErr error
-	for attempt := 0; attempt <= maxRetries; attempt++ {
-		lastErr = fn()
-		if lastErr == nil {
-			return nil
-		}
-		if !isRateLimitError(lastErr) {
-			return lastErr
-		}
-		if attempt == maxRetries {
-			break
-		}
-
-		// Compute exponential backoff delay.
-		delay := baseDelay * (1 << attempt)
-		if delay > maxBackoffDelay {
-			delay = maxBackoffDelay
-		}
-		// Use Retry-After header value if present.
-		if ra := parseRetryAfter(lastErr.Error()); ra > 0 {
-			delay = ra
-		}
-
-		select {
-		case <-ctx.Done():
-			return fmt.Errorf("context cancelled during rate-limit backoff: %w", ctx.Err())
-		case <-time.After(delay):
-		}
-	}
-	return lastErr
-}