Over the past few weeks, we've gutted and rebuilt the entire Agent Skills library that ships with case.dev. Every skill has been rewritten, pressure-tested through our review pipeline, and tuned for production use inside CaseMark for Legal Teams, our flagship product for legal professionals.
If you build on case.dev, this matters to you. Here's the short version of what changed, what it unlocks, and where skills fit in the platform.
What we actually did
We didn't just add skills. We threw out the old ones and started over.
The previous library had grown organically. Skills referenced documents that didn't exist, paraphrased compliance language where precision was required, and carried inconsistencies that only surfaced under real-world agent workloads. We caught these issues through our automated review pipeline and decided the right move was a clean rewrite rather than incremental patches.
The result is a tighter, more reliable set of skills built to a single standard: can an autonomous agent use this skill to produce work a legal professional would actually trust?
What skills are (and aren't)
Skills are structured prompts and methodologies that give AI agents domain-specific expertise on demand. They're available through the Skills API and as an MCP server at skills.case.dev/api/mcp, which means any MCP-compatible agent (Claude Desktop, Cursor, your own tooling) can pull them in at runtime.
They are not fine-tuned models. They're not RAG. They're portable, composable instructions that an agent loads when it needs to do something specific, like analyze a deposition, draft a demand letter, build a medical chronology, or navigate SAR filing requirements. The agent gets the expertise it needs for the task at hand without burning context on domains that aren't relevant.
Why this matters for case.dev developers
Skills are the connective tissue between raw AI capability and domain-correct output. When you combine them with the rest of the case.dev platform, the workflow gets interesting fast:
Vault stores the documents. OCR extracts the text. Voice transcribes the recordings. The LLM API does the reasoning. Skills tell the agent how to reason about legal work specifically -- what to look for in a deposition, how to structure a chronology, which compliance rules actually apply and what the precise language needs to be. Memory lets the agent retain context across sessions. Summarize runs the structured analysis workflows.
Without skills, you have powerful general-purpose infrastructure. With skills, you have infrastructure that knows how to practice law.
What the rewrite focused on
Three things drove every decision in the rebuild:
Compliance precision. Legal and regulatory skills now carry exact language where exact language is required. We flagged and fixed every instance where paraphrasing had introduced ambiguity in areas like filing deadlines, disclosure requirements, and statutory citations. In legal work, "close enough" is a liability.
Agent-first design. These skills aren't documentation for humans. They're instructions for autonomous agents operating under real constraints: limited context windows, multi-step workflows, and the need to produce output that passes professional review. Every skill is structured to minimize ambiguity and maximize the agent's ability to execute without hand-holding.
Reference integrity. Every skill that references supporting material now has that material present and accounted for. No dead links, no phantom documents. If a skill says "see reference," the reference exists.
How to use them
If you're already on case.dev, you can start pulling skills today:
GET /skills/resolve?q=deposition+analysis&limit=5
Or point any MCP client at https://skills.case.dev/api/mcp and let your agent discover what's available. The free tier gives you 50 requests per day with no authentication required, so you can evaluate before you commit anything.
If you're building with Thurgood, skills load progressively as your agent works. You don't need to manage them manually.
What's next
The skills library is open source on GitHub. We're continuing to add coverage across practice areas and jurisdictions, and we're tightening the automated review pipeline that caught the issues in the first place. If you're building legal AI and want to contribute skills or flag gaps, the repo is open.
The goal hasn't changed: make case.dev the infrastructure layer where legal tech developers build, so you can focus on your product instead of reinventing domain expertise from scratch.
Start building at case.dev. Skills API docs at docs.case.dev/agent-skills.
