Academy Prompt Library Recommender

Prompt Library Recommender — Personalized Prompt Discovery

When to use this

Invoke when:

• A new user asks "where do I start?" or "what should I try first?"
• A returning user asks "what prompts are useful for [practice area]?"
• A user's usage profile shows they are only using a fraction of relevant capabilities
• A user selects a role or jurisdiction during onboarding and needs a contextual starting pack
• An in-product empty state needs to surface relevant entry points

What the Prompt Library is

The Louis Prompt Library contains 152 curated prompts — pre-written starting points for common legal tasks. These are not templates that substitute for skill judgment; they are well-formed entry points that help users:

• Understand how to phrase requests to get better outputs
• Discover capabilities they did not know existed
• Build their own prompt habits faster by starting from proven examples

Prompts are organized by category, practice area, jurisdiction, and role.

Recommendation algorithm — how it works

Inputs

| Input | Source | Weight | |---|---|---| | User role | Profile / onboarding | High | | Primary jurisdictions | Profile / onboarding | High | | Recent activity | Session history (anonymized) | Medium | | Practice areas worked in | Inferred from recent docs / skills used | Medium | | Prompts already used | Usage log | Excludes already-used from recommendations | | Explicit "not useful" signals | User dismiss / thumbs-down | Excludes category |

Logic

1. Filter by jurisdiction: surface only prompts relevant to the user's stated jurisdictions. A Beirut lawyer does not need KSA-specific prompts as a priority.
2. Filter by role: a paralegal and a senior partner have different entry points.
3. Surface by practice area: if the user has been working on employment matters this week, prioritize employment-adjacent prompts.
4. Progressive disclosure: show 3–5 prompts, not 50. The goal is actionable discovery, not an overwhelming catalog.
5. Learn and adjust: if a user clicks and runs a prompt, note the category as high-interest. If a user dismisses repeatedly, reduce that category's frequency.

Recommendation output format

Each recommendation surfaces:

1. Prompt title (e.g., "Review a UAE employment contract for mandatory clause gaps")
2. One-sentence description (what it does and when to use it)
3. Practice area and jurisdiction tags
4. "Try it" button that pre-fills the prompt into the chat input
5. Related prompts (1–2 links to similar prompts the user might also want)

Prompt packs by role — defaults

BigLaw / Large-Firm Associate

• Draft an NDA (jurisdiction-selectable)
• Review a services agreement for risk (jurisdiction-selectable)
• Summarize a contract in plain language for a client briefing
• Flag missing standard clauses in a draft SPA
• Generate a redline comparison summary

In-House Counsel

• Review a vendor contract against standard in-house positions
• Generate a contract register entry from an uploaded agreement
• Summarize key obligations and renewal dates from a services contract
• Draft an employment offer letter (jurisdiction-selectable)
• Identify regulatory compliance obligations in a new market

Solo Practitioner

• Draft a standard retainer / engagement letter
• Explain a legal provision in plain language (for client)
• Summarize what a court judgment means in practical terms
• Draft a simple services agreement for a client

Law Student / Bar Candidate

• Generate a case brief in IRAC format
• Explain a legal principle with examples
• Quiz me on [subject area] at [difficulty level]
• Draft a sample exam answer for [fact pattern]
• Build a study plan for bar exam preparation

Paralegal

• Extract key dates, parties, and obligations from a contract
• Summarize a document for file management
• Check a draft for formatting and cross-reference consistency
• Generate a document checklist for a closing

Quality standards for prompt recommendations

• Relevance over completeness. Three relevant prompts are more valuable than 15 marginally relevant ones.
• Don't re-recommend. Once a user has run a prompt, replace it with the next-best option in that category.
• Explain the recommendation. A brief "because you recently worked on employment matters in UAE" anchors the recommendation in the user's context and builds trust.
• Not prescriptive. Recommendations are suggestions. The user sets direction; the recommender surfaces options.

Edge cases

• New user with no history: default to the role-based starter pack. Prompt for role during onboarding if not set.
• User with highly varied practice: rotate recommendations across practice areas rather than locking to a single area.
• User who dismisses everything: surface a "help me find what I need" prompt that asks the user directly what they want to accomplish.

Related skills

• [[academy-use-case-explainer]]
• [[academy-feature-explainer]]
• [[academy-legal-ai-skills-catalog]]
• [[academy-solutions-by-persona]]

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sboghossian
• Source: sboghossian/mini-claude-for-legal
• License: MIT

Install and usage instructions live in the source repository linked above.