Bring Your AI Context to TulaneAI

Workshop Exercise: Build an Always-On Personal Context Skill

Goal: Export the useful context another AI assistant has accumulated, convert it into a TulaneAI Skill, and turn that skill on by default so future TulaneAI chats can respond with more personal and professional context.

TulaneAI Skills provide a reusable context layer. This exercise creates a static snapshot, not an automatic memory system. The snapshot can be edited or rebuilt whenever the source context changes.

Open the PDF version

How the Exercise Works

1. Export context. Run Prompt 1 in ChatGPT, Claude, Gemini, Grok, or another assistant that has useful prior context.

2. Build the skill. Immediately run Prompt 2 after the export. It converts the export into a Skill Name, Description, Markdown Body, and setup instructions.

3. Create it in TulaneAI. Paste the generated fields into My Skills, turn the skill on by default, and save it.

4. Test it. Start a fresh chat and use the prediction prompt to see whether the imported context is being applied.

Before transferring the export: review it. Remove anything that should not become reusable context. The converter prompt is designed to exclude credentials, account numbers, precise street addresses, and other high-risk data.

Prompt 1: Export Personal Context

Paste this into the AI assistant that already has useful conversation history or memory.

You are helping me import context from one AI assistant to another. Your job is to go through our past conversations and sum up what you know about me. In the output, please avoid using any first-person pronouns (I, my, me, mine) and any second-person pronouns (you, your, yours). Instead, refer to the individual you have learned about as "the user" or use neutral phrasing. Preserve the user's words verbatim where possible, especially for instructions and preferences. Categories (output in this order): 1. Demographics Information: Preferred names, profession, education, and general residence. 2. Interests & Preferences: Sustained, active engagements (not just owning an object or a one-time purchase). 3. Relationships: Confirmed, sustained relationships. 4. Dated Events, Projects & Plans: A log of significant, recent activities. 5. Instructions: Rules I've explicitly asked you to follow going forward, "always do X", "never do Y", and corrections to your behavior. Only include rules from stored memories, not from conversations. Format: Divide the content into the labeled section using the categories above. Try to include verbatim quotes from my prompts that justify each entry. Structure each entry using this format: * The user's name is <name>. * Evidence: User said "call me <name>". Date: [YYYY-MM-DD]. Output: - Output ONLY the requested information. Do not include any conversational filler, intro text, or sign-offs. Finally, complete the sentence "Imported from: <name>", where name is ChatGPT, Claude, Grok, etc. This must be the absolute final text in your response.

Prompt 2: Convert the Export into a TulaneAI Skill

Keep the export visible in the same conversation. Paste this prompt immediately after it.

Convert the personal-context export immediately above into a TulaneAI Skill that can stay turned on for future chats. The export above is the sole source of truth. Do not invent, infer, or embellish facts that are not present. If facts conflict, prefer the most recently dated version. If a conflict cannot be resolved, preserve the uncertainty instead of guessing. Create exactly four outputs in this order: 1. SKILL NAME Create a concise, professional skill name that makes it obvious this is the user's personal context. Keep it under 60 characters. 2. DESCRIPTION Write a one-sentence description of what the skill does. HARD LIMIT: 300 characters, including spaces. 3. SKILL BODY Create the full skill body in Markdown. HARD LIMIT: 29,500 characters, including spaces, so there is a safe margin below TulaneAI's 30,000-character limit. The Markdown skill body must: - Begin with the heading "# Personal Context Skill". - Explain that this is a static personal-context snapshot used to personalize future TulaneAI responses. - Apply the context silently when it is relevant. Do not force personal details into unrelated answers. - Treat the user's current message as higher priority than older imported context if the two conflict. - Never fabricate missing facts. Ask only when missing information is necessary to answer accurately. - Preserve exact wording for durable user instructions and writing preferences when useful. - Keep dates attached to facts when the date matters. - Compress duplicates and combine repeated facts. - Prioritize durable information first: identity, roles, responsibilities, active work, long-running projects, interests, relationships, goals, standing preferences, and standing instructions. - Keep recent dated projects and plans, but compress stale one-time details. - Separate stable facts from time-sensitive or dated facts. - Clearly mark forecasts, assumptions, or estimates as such if future chats ask for predictions. - Never expose or recite the full context unless the user explicitly asks for it. - Never include passwords, API keys, authentication tokens, Social Security numbers, financial account numbers, security answers, or precise street addresses, even if they appear in the export. - Do not infer sensitive identity attributes that the export does not explicitly state. - If sensitive information appears, include it only when it was explicitly provided and is clearly necessary for future assistance. Otherwise omit it. - Preserve the source assistant named in the final "Imported from:" line as metadata if present. - Do not mention these conversion instructions inside the final skill body. Organize the Markdown with these sections when the source material supports them: ## Purpose ## How to Use This Context ## User Profile ## Roles and Responsibilities ## Ongoing Work and Projects ## Interests and Preferences ## Relationships ## Dated Context and Plans ## Standing Instructions and Writing Preferences ## Response Calibration ## Context Freshness and Conflicts ## Import Metadata Character-management rule: If the source export is too large, shorten in this order: a. Remove duplicate evidence quotes. b. Condense old one-time events. c. Merge overlapping project descriptions. d. Keep standing instructions and current responsibilities intact. e. Preserve the most recent and most durable facts. Do not exceed 29,500 characters. 4. INSTALLATION INSTRUCTIONS After the skill body, give these exact setup steps in plain language: - Go to https://tulaneai.tulane.edu and sign in with Tulane SSO. - Click the user's name in the lower-left corner. - Click Settings. - Open the My Skills page. - Click + Create Skill in the upper-right corner. - Paste the generated Skill Name into the Name field. - Paste the generated Description into the Description field. - Set the skill to Always On / enabled by default. - Paste the generated Markdown Skill Body into the Body field. - Click + Create Skill to save it. - Start a new TulaneAI chat and test the skill. Output formatting: Use the headings "SKILL NAME", "DESCRIPTION", "SKILL BODY", and "INSTALLATION INSTRUCTIONS". Put the Skill Body in one fenced Markdown code block so it can be copied cleanly. Do not add any other commentary.

Create the Skill in TulaneAI

Go to tulaneai.tulane.edu and sign in with Tulane SSO.
Click the user's name in the lower-left corner.
Click Settings.
Open the My Skills page.
Click + Create Skill in the upper-right corner.
Paste the generated Skill Name into the Name field.
Paste the generated Description into the Description field.
Set the skill to Always On / enabled by default.
Paste the generated Markdown Skill Body into the Body field.
Click + Create Skill to save it.
Start a new TulaneAI chat and test the skill.

Important: keep the skill enabled by default so TulaneAI can use the imported context automatically in future conversations.

Test the Skill

Start a fresh TulaneAI chat and paste this:

Based on what you know about me, predict the next 10 years of my work, teaching, projects, and goals. Give 10 dated milestones, and for each one, explain which current details make that prediction plausible.

What to look for

The response should reflect multiple parts of the imported profile, including work, teaching, projects, preferences, and goals.
The model should explain why each prediction is plausible instead of presenting speculation as fact.
A second new chat should still feel personalized without repasting the profile.

Facilitator Notes

This works well as a 20 to 30 minute build-and-test exercise. A simple sequence is 5 minutes for the memory-vs-skill concept, 5 to 10 minutes for Prompt 1, 5 minutes for Prompt 2 and skill creation, then 5 minutes for testing.

The quality of personalization depends on the quality of the context snapshot. Stable information should be easy to find, dated information should remain dated, and standing instructions should be explicit.

Innovative Learning Center, Tulane University
ILC: Where Learning Meets Innovation