Agent Skill Usage Tracker

See whether agents actually use the skill you shipped—at runtime

Problem

A lot of SaaS and API platforms publish agent skills, yet the authors have no idea how often those skills are actually used in the wild. Someone on HN openly asks how to measure how often an installed skill gets triggered, how often an agent falls back to web search instead of following the skill, and how long it takes for a skill to ever fire. Teams only have evals under ideal conditions and no runtime behavior. Revenue is affected but invisible.

Solution

Offer a lightweight SDK or proxy endpoint that hooks into the skill's MCP or tool-call path to log whether a skill was invoked, by which agent, whether a fallback occurred, latency and success. A dashboard breaks down trigger rate, bypass rate and failure reasons by skill and agent version, with export support. No user text is collected—only structured hooks.

Users

SaaS platforms shipping agent skills, API companies, indie agent developers, enterprise platform teams

MVP

Hook for one mainstream agent protocol (e.g. MCP); log invoked / skipped / fallback events; one dashboard with trigger rate per skill; a one-line install doc and npm package; CSV export for weekly reports.

Revenue

Usage-based subscription (events) plus a per-seat team dashboard tier; free monthly cap with paid unlocks for indie devs.

Why now

Over the past year agent protocols like MCP have spread fast and every platform is publishing skills, but runtime observability is nearly non-existent. Being the hook layer is easier to land than building yet another model.