model-rate-limit-recovery（模型速率限制恢复）

Model Rate Limit Recovery Skill

When to Use

• Cron jobs fail with "⚠️ You have hit your ChatGPT usage limit (free plan). Try again in ~9500 min"
• Agent sessions fail with 429, rate_limit, quota, or resource exhausted errors
• Need to recover scheduled tasks after model provider limits are hit
• Setting up resilient agent workflows with automatic fallbacks

Diagnosis Steps

1. Check Error Type

# Check cron job runs
openclaw cron runs --jobId <job_id>

# Look for error messages containing:
# - "usage limit"
# - "Try again in ~"
# - "429"
# - "rate_limit"
# - "quota"
# - "resource exhausted"

2. Verify Current Configuration

# Check current model configuration
openclaw status | grep -A5 "Model"

# Check environment for API keys
env | grep -i "OPENAI\|ANTHROPIC\|DEEPSEEK"

3. Identify Root Cause

Common causes:

• Free plan limits: ChatGPT free tier has usage caps
• No API key rotation: Single key exhausted
• No fallback model: Default model fails, no alternative
• Cron jobs using default model: Not specifying resilient model

Recovery Procedures

Immediate Recovery (Manual)

# 1. Run failed task manually with alternative model
openclaw sessions spawn \
  --task "Your task here" \
  --model "deepseek/deepseek-chat" \
  --label "Manual recovery"

# 2. Update cron job to specify model
openclaw cron update --jobId <job_id> --patch '{
  "payload": {
    "kind": "agentTurn",
    "message": "...",
    "model": "deepseek/deepseek-chat",
    "timeoutSeconds": 180
  }
}'

API Key Rotation Setup

# Add multiple API keys for rotation
export OPENAI_API_KEYS="key1,key2,key3"
export OPENAI_API_KEY_1="sk-..."
export OPENAI_API_KEY_2="sk-..."
export OPENCLAW_LIVE_OPENAI_KEY="sk-..."  # Highest priority

# OpenClaw will automatically rotate through keys on rate limits
# 429, rate_limit, quota, resource exhausted → tries next key
# Other failures → fails immediately

Model Fallback Configuration

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "openai-codex/gpt-5.4",
        "fallback": "deepseek/deepseek-chat"
      },
      "models": {
        "openai-codex/gpt-5.4": {
          "params": {
            "maxRetries": 2,
            "retryOnRateLimit": true
          }
        }
      }
    }
  }
}

Cron Job Best Practices

{
  "name": "Resilient Scheduled Task",
  "schedule": {
    "kind": "cron",
    "expr": "0 * * * *",
    "tz": "Africa/Johannesburg"
  },
  "payload": {
    "kind": "agentTurn",
    "message": "Task instructions...",
    "model": "deepseek/deepseek-chat",  // Specify model explicitly
    "timeoutSeconds": 300               // Set reasonable timeout
  },
  "sessionTarget": "isolated",
  "delivery": {
    "mode": "announce"
  }
}

Prevention Strategies

1. Model Selection

• Primary: openai-codex/gpt-5.4 (when available)
• Fallback: deepseek/deepseek-chat (no usage limits)
• Backup: anthropic/claude-sonnet-4-6 (if available)

2. Cron Job Configuration

• Always specify model in payload
• Set reasonable timeoutSeconds
• Use deleteAfterRun: true for one-shot tasks
• Enable delivery.mode: "announce" for notifications

3. Monitoring

# Regular cron job health checks
openclaw cron list
openclaw cron runs --jobId <job_id> --limit 5

# Check for recent failures
grep -i "usage limit\|429\|rate_limit" /tmp/openclaw/openclaw-*.log

4. Skill Integration

# Create recovery script
cat > /root/.openclaw/workspace/scripts/recover-failed-cron.sh <<'EOF'
#!/bin/bash
JOB_ID="$1"
NEW_MODEL="${2:-deepseek/deepseek-chat}"

# Get failed runs
FAILED_RUNS=$(openclaw cron runs --jobId "$JOB_ID" | grep -c "status.*error")

if [ "$FAILED_RUNS" -gt 0 ]; then
  echo "Recovering $FAILED_RUNS failed runs for job $JOB_ID"
  openclaw cron update --jobId "$JOB_ID" --patch "{\"payload\":{\"model\":\"$NEW_MODEL\"}}"
  openclaw cron run --jobId "$JOB_ID"
fi
EOF
chmod +x /root/.openclaw/workspace/scripts/recover-failed-cron.sh

Templates

Resilient Cron Job Template

See scripts/resilient-cron-template.json

Model Fallback Config

See references/model-fallback-config.json

Recovery Script

See scripts/recover-failed-cron.sh

Notes

• ChatGPT free plan has strict usage limits (~3 requests/hour)
• DeepSeek has no usage limits but may have different capabilities
• API key rotation only works for rate limits (429), not other errors
• Always verify recovery by checking created files/outputs
• Document failures and recoveries in memory/ for future reference

References

• OpenClaw Model Providers Docs
• API Key Rotation Documentation
• Cron Job Management