agentpulse_skill

Overview

This skill tracks every LLM API call your agent makes and aggregates costs, token usage, latency, and errors into a real-time dashboard at agentpulse.dev. It runs silently in the background, reporting events so you can monitor spending and performance without changing your agent workflow. Use it to answer spending, token, and reliability questions with accurate, server-calculated figures.

How this skill works

After each LLM API call, the skill posts a small JSON event to the AgentPulse API containing agent name, provider, model, input/output token counts, latency, and status. AgentPulse calculates costs server-side using an up-to-date pricing table and stores daily breakdowns, top models, error counts, and averages. When you ask about spending or performance, the skill fetches an overview summary from the AgentPulse stats endpoint and presents clear metrics and trends.

When to use it

• You want to know how much your agent spent today, this week, or this month.
• You need a breakdown of token usage or the most expensive models.
• You want to monitor latency and error rates for reliability debugging.
• You need alerts for spending limits, consecutive failures, or rate-limit spikes.
• You want a daily spending trend or per-day API call counts.

Best practices

• Sign up at https://agentpulse.dev and set AGENTPULSE_API_KEY in the environment before running your agent.
• Report events after every API call, including token counts and latency, but do this silently to avoid user noise.
• Use server-side cost data from AgentPulse rather than local estimates for accurate billing.
• Configure alerts on the AgentPulse dashboard for cost thresholds, failure streaks, and rate-limit spikes.
• Limit captured data to metadata (model, counts, status); enable prompt capture only if you explicitly need it and understand privacy implications.

Example use cases

• Show total spend and API call count for the current month and list top models by cost.
• Diagnose spikes in latency or rate-limit errors by inspecting recent daily stats and average latency.
• Track token usage growth over time to decide when to switch to a cheaper model variant.
• Set a daily cost limit and receive alerts when the agent approaches the threshold.
• Audit error counts and consecutive failures to trigger reliability improvements in your agent.