sboghossian Justice — Product Demo: Context Memory
Purpose
A demo that reuses facts the user shared earlier is dramatically more persuasive than one using generic examples. This skill defines how the assistant captures, stores, and applies session context during a product demo, creating a coherent "this was built for me" experience.
Inputs / signals to capture
Listen throughout the demo conversation for the following and store them in the session context object:
| Signal | Example | Store as |
|---|---|---|
| Industry / sector | "we do MSAs for marketing agencies" | industry: marketing, doc_type: MSA |
| Firm type | "I'm a solo practitioner", "we're a 20-person firm" | firm_size: solo or firm_size: 20 |
| Jurisdiction | "most of our clients are in Lebanon" | jurisdiction: LB |
| Practice area | "we focus on corporate M&A" | practice_area: corporate_ma |
| Pain point | "contract review takes us forever" | pain_point: contract_review_time |
| Current tools | "we use Word and email" | current_stack: word, email |
| Language preference | responds in Arabic | language: ar |
| Specific documents | "we do employment contracts a lot" | doc_focus: employment_contracts |
| Deal size or volume | "about 30 contracts a month" | volume: 30_monthly |
Memory rules
Capture
- Capture context passively — do not run a questionnaire at the start. Let context emerge naturally.
- If the user explicitly says something ("I work in construction law in Saudi Arabia"), treat it as high-confidence context.
- If context is implied ("our Riyadh clients" → likely KSA jurisdiction), treat it as lower-confidence; confirm if needed.
Apply
- In all subsequent demo examples, substitute the captured context for generic placeholders:
- Generic: "Let me show you an NDA draft"
- Personalized: "Let me show you an employment contract draft — typical for a Saudi-based firm like yours"
- When invoking a skill for demonstration, pre-fill jurisdiction, document type, and language from context
- When suggesting next steps, reference the user's stated pain points ("Given that contract review is your main bottleneck, here's what the review workflow looks like")
Do not re-ask
- If the user already stated their jurisdiction, do not ask again later in the session
- If the user said they are a solo practitioner, do not show team-workflow steps as the primary demo
Memory scope
- Session only: context memory lives for the duration of the demo conversation
- Cleared at session close unless the user creates an account — do not persist to local storage or backend without user consent
- If the user signs up, prompt: "Shall I save your preferences to your new account?" (opt-in)
Integration with demo flow
The context memory feeds into:
- Example generation: all example contracts, clauses, and research topics are drawn from the user's stated industry and jurisdiction
- Skill suggestions (Step 4 of the demo flow): skills shown are filtered to the user's practice area
- ROI framing: time-saved estimates reference the user's stated volume ("at 30 contracts/month, this saves you ~40 hours")
- Demo recap: [[justice-product-demo-recap]] uses the session context to generate a tailored wrap-up
Failure modes
| Situation | Handling |
|---|---|
| Context is ambiguous or contradictory | Use the more recent / more specific signal; proceed with a soft assumption and confirm if high-stakes |
| No context captured by Step 3 of demo | Ask one targeted question: "What type of work do you do most — drafting, reviewing, or research?" |
| User changes context mid-demo | Update immediately; acknowledge the change naturally ("Got it — let me switch to a construction contract example instead") |
Related skills
- [[justice-product-demo-recap]]
- [[justice-intent-product-demo-request]]
- [[justice-result-page-routing]]