Insights & Updates

Most agent skills ship with zero tests. Here's why that's catastrophic — and the four-part eval loop that fixes it, borrowed from a Braintrust talk that should be required watching for anyone writing skills.

OpenAI's Agents SDK team found that improving skill descriptions boosted routing accuracy more than changing skill logic. This is exactly what selftune automates.

Anthropic published their internal lessons from using Claude Code skills at scale. Here's what every skill developer needs to know — and what it means for how you build.

Install, initialize, and get your first skill health check in under 2 minutes. A step-by-step guide to SelfTune.

TanStack Intent solves skill distribution. selftune solves skill quality. Together they complete the agent skill lifecycle. Here's how the three-layer stack works.

Agent skills and MCP tools both extend AI agents, but they solve different problems. Compare when to use skills vs MCP server tools.

Agent skills fail silently. Diagnose why your skills don't trigger, debug common SKILL.md issues, and fix undertriggering with real data.

Learn how to write agent skills that trigger reliably. 8 best practices for SKILL.md descriptions, trigger phrases, and skill testing.

Compare the top agent skill marketplaces and registries. Where to find, share, and manage skills for Claude Code, Codex, and OpenCode.

Agent skills are reusable instruction sets that extend AI coding agents. Learn how SKILL.md works, what skills do, and how to evaluate them.

270K+ published skills. $7.84B agent market. Not a single skill that learns from its users. A data-driven look at where the skill ecosystem stands today.

When a skill misses a trigger, there is no error, no log, no signal. The agent just skips it. Here is why that happens and what you can do about it.

There are 270K+ skills across agent marketplaces and none of them learn how you talk. SelfTune makes your skills adapt to you automatically.