Skills, Rules, and the Art of Encoding Procedures
Part 4 of the AI Knowledge Series
Knowledge is only half the equation. The other half is knowing what to do with it.
Every experienced litigator has procedures -- sequences of steps for accomplishing specific tasks. How to analyze a lease for litigation risks. How to prepare a witness for deposition. How to structure a brief for maximum impact.
These procedures are usually implicit. We follow them without consciously thinking about each step. But implicit procedures can't be taught to AI.
Making them explicit -- encoding them as structured "skills" -- is what transforms AI from a general-purpose tool into a practice-specific assistant.
What a Skill Looks Like
Consider the task of drafting a response to a motion for summary judgment in a Texas state court oil and gas case.
An experienced lawyer doesn't just start writing. They follow a procedure:
- Analyze the motion: What grounds? What evidence cited? What's the core theory?
- Identify the response framework: Traditional vs. no-evidence? Mixed?
- Gather authorities: What cases support your position? What distinguishes their cases?
- Review the record: What evidence creates fact issues? What's missing from their presentation?
- Structure the argument: What order? Which points need most development?
- Draft: Following local conventions, citation practices, length constraints
- Verify: Citations accurate? Evidence properly cited? Procedural requirements met?
This isn't just a checklist -- it's a decision tree with branches at each step. If the motion is no-evidence, you need one kind of response. If it's traditional, another. If it's mixed, you need to address both.
Encoding this as an AI skill means capturing not just the steps but the decision logic.
Skills vs. Prompts
You might think: "Can't you just write a detailed prompt that explains all this?"
You can, but it doesn't work as well. Here's why:
Context limits: AI systems have finite context windows. A truly comprehensive prompt might exceed them.
Maintenance: When your procedure changes -- new case law, different court preferences -- you need to update the skill. A monolithic prompt is harder to maintain than a structured skill.
Modularity: Skills can call other skills. Your summary judgment response skill might call your "find distinguishing authorities" skill, which calls your "search knowledge base" skill. This modularity makes complex workflows manageable.
Verification: With structured skills, you can see exactly what the AI did at each step. With a long prompt, the AI's reasoning is opaque.
The Rule Layer
Skills define procedures. Rules define constraints.
Rules are always-on requirements that apply regardless of what skill is running:
- Citation format: Texas citation rules, Bluebook preferences, firm-specific conventions
- Ethical constraints: No misrepresenting authorities, proper adverse authority disclosure
- Quality standards: Minimum research depth, verification requirements
- Style guide: Active voice, concise sentences, structural preferences
Separating rules from skills means you don't have to repeat the same requirements in every skill definition. The rules layer applies automatically.
Encoding Institutional Knowledge
The real power of skills emerges when you encode institutional knowledge -- the stuff that isn't written down anywhere.
Examples from my practice:
Judge-specific preferences: Judge Garcia wants comprehensive fact sections. Judge Wilson prefers jumping straight to legal analysis. These preferences become parameters in my briefing skills.
Opposing counsel patterns: Thompson & Associates always files timing objections. My response preparation skill includes a step to anticipate and address these.
Client considerations: Certain clients need more detailed explanations. Others want bottom-line recommendations. My memo-drafting skills adjust based on client profiles.
This knowledge already exists -- it's in your head. Encoding it makes it persistent, shareable, and accessible to AI assistance.
The Learning Loop
Skills shouldn't be static. Every use is an opportunity for refinement.
My workflow includes a feedback loop:
- AI executes skill
- I review output
- I note what worked and what didn't
- Periodically, I update the skill based on accumulated feedback
Over time, skills get better. The first version of my motion-drafting skill was crude -- basically a structured checklist. The current version reflects hundreds of refinements based on actual use.
Practical Skill Design
Some lessons from building skills:
Start small. Don't try to encode your entire practice at once. Pick one specific, frequent task. Build a skill for it. Refine it. Then expand.
Document decision points. The places where your procedure branches based on circumstances are the places where AI most needs guidance.
Include verification steps. Every skill should have checkpoints where the AI confirms it's on track before proceeding.
Accept iteration. Your first attempt won't be perfect. Build a minimum viable skill, use it, improve it.
The Compound Effect
The more skills you build, the more powerful the system becomes.
A skill for analyzing leases can feed into a skill for identifying litigation risks. That feeds into a skill for developing case theories. Which feeds into a skill for structuring briefs.
Each skill builds on others. The whole becomes greater than the sum of parts.
This is why starting now matters. The compound effect needs time to develop. Firms that wait will face not just a knowledge gap but a skill gap -- years of refinement they haven't done.