See Every Capability. Control Every Cost. Govern Every Agent.
SkillTrace is the observability layer for AI capabilities — a new category of DevOps that reveals which skills power your AI systems, where your LLM spend originates, and what risks are hiding in production.
Top Capabilities
LLM Cost Attribution
$12.4K
attributed this week
Risk Monitor
filesystem-delete
+2,400% usage spike
db-admin-query
Unusual off-hours activity
Your AI Platform Has a Blind Spot
You can observe your infrastructure. You can trace your prompts. You can monitor your models. But you cannot see the capabilities that drive your AI systems.
What You Can Observe
- ✓Infrastructure — Latency, uptime, error rates
- ✓Applications — APM, traces, logs
- ✓Models — Token usage, prompt tracing, LLM latency
What You Cannot
Which skills, tools, and capabilities are your AI agents actually invoking in production?
Questions you cannot answer today
Without Capability Observability, managing AI at scale is flying blind.
Introducing Capability Observability
A new layer of DevOps for AI-native platforms. Just as Datadog made infrastructure observable, SkillTrace makes AI capabilities observable.
The missing layer in your AI operations stack
Skill Usage
Which capabilities power your AI platform in real time
Cost Attribution
Which capabilities drive your LLM spend — down to the skill
Capability Lifecycle
Which skills are deprecated, unused, or drifting from baseline
Operational Risk
Which sensitive capabilities are spiking without explanation
Zero-Instrumentation Detection
SkillTrace runs inside your AI gateway and passively inspects prompts. No agent changes. No SDK integration. Deploy once at the gateway.
Agents
AI Gateway
SkillTrace
LLM Provider
Telemetry Store
JSONL / SQLite / HTTP / Kafka
<5ms
p99 detection latency
Zero
prompt modification
Fire & forget
telemetry emission
Detection PipelineShort-circuits on first match
Watermark Scan
(microseconds)Detects skill:// watermark comments embedded in prompt content
Frontmatter Hash
(sub-millisecond)SHA-256 hash of YAML frontmatter block against registry
Content Hash
(< 1ms)SHA-256 hash of full prompt content for exact matching
Prefix Hash
(< 1ms)SHA-256 hash of first 500 tokens for partial matching
Watermark example
<!-- skill://acme.github/pr-review@1.2.0 -->
You are a code review assistant...A lightweight HTML comment embedded in the prompt. Detection takes microseconds with zero impact on LLM behavior.
Your Capability Command Center
Total visibility into every capability running across your AI platform. No blind spots. No guesswork.
Capability Usage
Active1,247
skills detected today
+12% vs last week
Know exactly which skills power your AI platform — in real time
Cost Attribution
Spend$3.2K
attributed to top 5 skills
sql-query: 34% of total
Trace every dollar of LLM spend back to the capability that drove it
Capability Drift
Drift18
deprecated skills still active
4 scheduled for removal
Surface deprecated, unused, or drifting skills before they become incidents
Operational Risk
Risk3
anomalies flagged this week
filesystem-delete: +2,400%
Detect anomalous spikes in sensitive capabilities instantly
Deploys Where You Already Run
One middleware. Every gateway. No agent changes.
LiteLLM
PythonPython callback plugin for the most popular AI gateway proxy
Vercel AI SDK
TypeScriptNative LanguageModelMiddleware for Vercel-powered AI apps
Express
Node.jsDrop-in middleware for Express-based OpenAI proxy servers
Cloudflare Workers
EdgeEdge-native adapter for Cloudflare AI Gateway Workers
Deployment typically requires one configuration change. No agent instrumentation required.
Built for Enterprise AI Governance
From multi-tenant gateway routing to capability audit trails — SkillTrace is designed for organizations that take AI operations seriously.
Scale
Built for high-throughput, multi-tenant AI platforms
- →Multi-registry federation
- →Multi-tenant gateway routing
- →Deterministic sampling
- →Hot-reload without restarts
Observe
Deep telemetry across every capability interaction
- →Streaming telemetry pipelines
- →Prometheus metrics endpoint
- →OpenTelemetry integration
- →Async fire-and-forget emission
Govern
Enterprise-grade controls for capability management
- →Capability audit trails
- →Lifecycle management
- →Risk anomaly alerts
- →Read-only prompt inspection
Security by Design
The DevSecOps Platform for AI Capabilities
From skill creation to production governance — one platform. SkillTrace works alongside skills-check to cover the full capability lifecycle.
Develop
Author and iterate on AI agent skills
Validate
Analyze quality, generate fingerprints
Detect
Runtime capability detection at the gateway
Analyze
Usage, cost, drift, and risk analytics
Govern
Lifecycle management and compliance
Deploy in Minutes
One middleware. One configuration. Full capability visibility.
Lightweight gateway middleware
SkillTrace runs as a thin middleware layer inside your existing AI gateway. Point it at your fingerprint registry and telemetry endpoint. Within minutes, every capability flowing through your gateway becomes visible.
- →No agent instrumentation required
- →No prompt modification — ever
- →No performance impact (<5ms p99)
import express from "express";
import { createMiddleware } from "@skills-trace/express";
const app = express();
app.use(
"/v1/chat/completions",
createMiddleware({
registry: "./skills-fingerprints.json",
emitter: "http://telemetry.internal/api/events",
})
);