Skills do the work.Humans review the output.
A skill is a focused, prompt-engineered unit of work (think: a clear set of instructions) that runs on the LLM you choose to use. Token cost is bounded — no recursive self-checking, no agentic loops burning through context to plan-retry-replan. The cost varies with the work (a longer input or a larger output uses more tokens), but it doesn't spiral the way an autonomous agent's cost does. The skill produces a deliverable; you review it; you decide. The supermajority of the analytical and drafting work: done by a tool. The final call: still yours.
Skills vs. agents — the practical difference
Agents iterate. They plan, call themselves recursively, and burn tokens at a multiple of the work itself. That's the right architecture for some problems — open-ended research, multi-step automation across systems — and the wrong one when you want a structured first-pass deliverable, more bounded cost, and a human in the loop on the output.
A skill is a scoped instruction set with defined inputs and a defined output. Execution is bounded; the deliverable comes back to you in a form you can review and revise yourself rather than re-prompting the LLM until it converges on what you wanted.
For much legal business management work — matter profitability, client rollups, partner book reviews, pricing memos, professional liability insurance comparisons, client attrition analyses, and more — the skill shape fits the work needed. The work is structured. The output gets reviewed by a licensed professional anyway. And most of the time savings comes from the automated first-pass draft, not from agentic iteration.
Output quality is the seller's responsibility, not ours. Skills on this marketplace differ in how they show their work — some include formulas, sources, and assumptions inline; others don't. The quality of each skill depends on the author's attention to detail and ability to accurately explain how to perform a task. If you've ever played the game Tell Me How To Make a Peanut Butter and Jelly Sandwich at a work retreat, you know how often there is a disconnect between the process someone thinks is required to perform work vs. the process communicated to someone else when asking them to perform the work. Clear communication is key when delegating a specific process and expecting a certain result. The same applies to skills. Skills on this marketplace are meant to be starting places based on one author's experience and ability to communicate how to do something. Buyers of these skills may prefer to do things differently or may learn how to do something better from others. Either way, buyers should purchase these skills with the expectation that they are skills they can shape to their own use cases and liking. Buyers should read each skill's detail page and free preview before they buy so they understand the basis of the skill they will be building upon.
Built for the business of law. Not a general AI store.
We're not trying to compete on breadth. We're the multi-LLM marketplace built for legal business management — managing partners, COOs, finance directors, legal-ops leads, IT, marketing, and HR. We don't vouch for the accuracy of any individual skill — sellers attest to the quality of their own work. What we provide is the marketplace infrastructure: a focused audience, payment plumbing, recourse flow, and disclaimers that match the use case.
A general AI marketplace will eventually have more skills — that is fine. We're not trying to be the biggest catalog. We're trying to be the catalog law firms and in-house legal operations teams can justify procuring tools from. That requires curation, firm-context calibration, multi-LLM coverage, and peer-customer review.
If you run the business of a law firm and tried a general AI marketplace and found yourself thinking “this isn't built for our workflows” — that's the gap Counsel Commons™ fills.