← back
Skills are new features: Building Skill-Centric Harness — Yogendra Miraje, FactSet
Takeaway
Treat skills as evaluated, model-dependent product features and build the harness that reliably discovers and executes them.
Summary
- FactSet replaced its proprietary workflow blueprints with standardized skills that encode how agent-driven product features work.
- A minimal skill-aware harness needs a registry, system-prompt metadata, and a file-reading tool; scripts additionally require execution support.
- Progressive disclosure loads skill bodies only when needed, making names and descriptions critical routing signals.
- Skills should follow user intent, such as earnings preparation, and undergo evaluation after model changes because instruction-following can regress.
skillsagent-harnessfactset
Original description
Since skills were open sourced, Yogendra Miraje's team at FactSet stopped thinking about shipping features and started thinking about shipping skills. A skill is a capability you hand the agent, and its heart is a short skill.md whose name and description are really routing signals: get them distinct and the agent triggers the right one, blur them and it fires the wrong skill or none at all. He walks through a minimal skill registry, progressive disclosure so the agent only loads what it needs, and trigger words, like asking for a PDF versus an HTML report, that decide which skill runs. The harder lessons show up at scale. Skills without evals drift, because a new model quietly stops obeying them, so he treats skills as contracts you test. Past ten skills you need search and embeddings to keep the library coherent; past a hundred you need real governance, admission, ownership, periodic audits, change management, and a human deciding whether a skill should exist at all. His close is that skills are the interface to your agentic products, and at enterprise scale governing them matters as much as writing them. Speaker info: https://x.com/YogiNotTheBear / mirajey https://yogimiraje.com Timestamps: 0:00 - From blueprints to skills 1:46 - Building skill centric agents 2:52 - Skills in context at enterprise scale 3:46 - Skills as the new features 5:04 - Inside a skill: the skill.md 6:37 - A minimal skill registry 7:42 - Progressive disclosure 9:13 - Trigger words and routing 12:17 - Skills without evals drift 12:56 - Search and embeddings past ten skills 13:46 - Governance past a hundred skills 16:36 - Skills as the product interface