Agent Rate Limiter

What This Skill Does

The Agent Rate Limiter, developed by mxmsabundance, is a preventative utility designed to stop AI agents from triggering 429 'Too Many Requests' errors. By implementing a rolling window tracking mechanism, the skill assigns your agent a performance tier ranging from 'ok' to 'paused'. It functions as a watchdog that monitors usage patterns without requiring external API keys or third-party monitoring subscriptions. When the agent approaches capacity, the skill signals the system to downshift activity, and if a limit is exceeded, it calculates and enforces an exponential backoff schedule to ensure autonomous recovery without manual intervention.

Installation

To install this skill into your OpenClaw environment, use the command: clawhub install openclaw/skills/skills/mxmsabundance/agent-rate-limiter. Once installed, you can integrate it into your agent loop by placing the gate check at the beginning of your task execution and recording the request volume upon completion. No complex configuration is required, as the script comes pre-loaded with conservative defaults for Claude and OpenAI tiers, making it ready for immediate use out of the box.

Use Cases

This skill is essential for agents performing long-running autonomous workflows such as scraping, multi-step research, or high-frequency task automation. It is particularly valuable for users who run agents in the background where they cannot manually intervene if a service blocks the account. By proactively managing the rate, the skill keeps the agent within the allowed operational throughput, saving valuable context and preventing the need to restart complex sub-agent hierarchies mid-process.

Example Prompts

1. "Check my current rate limiter status and tell me if it's safe to start a long-running recursive web scraping task."
2. "Adjust the rate limiter to the 'max-20x' profile and ensure it handles the backoff logic for my next series of batch API calls."
3. "Summarize the last 5 hours of usage reported by the rate limiter and determine if I am nearing the 95% throttling threshold."

Tips & Limitations

Because the skill relies on a local JSON state file, ensure the file path is accessible by your agent's process. While the presets are robust, actual provider limits can vary based on your specific account history; therefore, start with conservative settings. Note that this tool is designed for programmatic enforcement within the agent loop; it does not interface with the API provider's live billing dashboard, meaning it calculates load based on your inputs and estimates, not real-time telemetry from the model provider.