AI Skill Generation

Instead of configuring each step of the skill builder manually, you can describe what you want in natural language and let alice generate the skill for you. This is the fastest way to go from an idea to a working slash command.

How It Works

AI skill generation uses your connected source metadata to build an accurate query. Here is the process:

1. You describe what you want. Type a plain English description of the data you need, between 10 and 500 characters.
2. alice analyzes your sources. The AI looks at your connected Stripe account data and Supabase database schema (table names, column names, column types) to understand what data is available.
3. alice generates a complete skill. The output includes:
   • Skill name and slug derived from your description
   • A description explaining what the skill does
   • The query in the appropriate mode (SQL for Supabase, API/metric for Stripe, or combined for multi-source)
   • Parameters with types, defaults, and descriptions where appropriate
   • Output configuration (table, summary, list, or text) optimized for the result shape
   • Trigger suggestions based on the query type (e.g., scheduled digest for time-based reports)
4. You review and edit. The generated skill opens in the full skill builder with all six steps pre-populated. Edit anything before saving.
5. You save. Click Create Skill to make it available as a slash command.

Where to Find It

There are two entry points for AI generation:

• Skills page > New Skill -- The skill builder includes an AI generation option at the top of the Setup step. Enter your description and click "Generate."
• Empty state -- If you have no skills yet, the skills page shows a quick-start prompt where you can type a description directly.

Both paths lead to the same generation flow.

Example Prompts

Here are effective prompts along with what alice generates for each:

Revenue and Billing

Customer Intelligence

Application Data (Supabase)

Cross-Source

What Gets Generated

Every AI-generated skill includes these components:

Name and Slug

The AI derives a concise name and URL-safe slug from your description. For example, "users who signed up this week but have not logged in" becomes:

• Name: "Inactive New Signups"
• Slug: inactive-new-signups

You can change both in the Setup step.

Description

A 1-2 sentence summary of what the skill does. This appears in the skills list and in /alice help output.

Query

The AI selects the appropriate query mode based on your description and connected sources:

• Stripe Metric for revenue, subscription, payment, and customer metrics
• Stripe API for raw Stripe data without metric computation
• Raw SQL for Supabase queries involving your application tables
• Combined for cross-source queries referencing both Stripe and Supabase data

Parameters

Dynamic inputs with appropriate types and defaults. For example, a prompt mentioning "last 30 days" generates a days parameter of type number with a default of 30. Parameters appear as flags when running the command: /alice inactive-new-signups --days=14.

Output Configuration

The AI chooses the best output type for the result shape:

• Summary for single-value metrics (MRR, churn rate, counts)
• Table for multi-row tabular data
• List for item-by-item results with titles and details
• Text for narrative or single-string results

Field labels, formats (currency, percentage, date, etc.), and empty-state messages are all configured automatically.

Trigger Suggestions

For time-based queries (digests, daily reports), the AI may suggest a schedule trigger with an appropriate cron expression. For threshold-based queries (failed payments, churn alerts), it may suggest a monitor trigger. You can accept, modify, or remove these suggestions.

Tips for Better Results

Be Specific About Data

The more specific your description, the better the generated query. Compare:

• Vague: "Show me users" -- alice does not know which users, what columns, or what timeframe
• Specific: "Users created in the last 7 days with email and plan, sorted by creation date" -- alice generates a precise query with the right columns and filter

Mention Timeframes

Include time boundaries when relevant. The AI interprets natural language like:

• "last 7 days", "this week", "past month"
• "since January", "in the last quarter"
• "today", "yesterday"

Without a timeframe, the AI defaults to reasonable periods (30 days for most metrics, 7 days for alerts).

Reference Known Fields

If you know your database schema, mention specific table or column names. "Users from the profiles table where subscription_tier is 'pro'" generates a more accurate query than "find pro users."

Specify the Source

If you have multiple connected sources, mention which one to query: "from Stripe" or "from my Supabase database." For combined queries, mention both: "join Stripe revenue with Supabase login data."

Describe the Output You Want

Mention how you want results formatted: "as a table," "show me a summary," "list them with email and status." The AI uses these hints to choose the right output type and field selection.

Keep It Under 500 Characters

The input limit is 500 characters. This is intentional -- concise prompts produce better results than lengthy paragraphs. Focus on the what, not the how. alice figures out the how.

Editing After Generation

The generated skill opens in the full 6-step skill builder. You can edit every aspect:

• Step 1 (Setup) -- Change name, slug, or description
• Step 2 (Query) -- Edit the query, switch modes, adjust filters
• Step 3 (Format) -- Change output type, add/remove columns, adjust formats
• Step 4 (Trigger) -- Add or remove schedule and monitor triggers
• Step 5 (Schedule) -- Fine-tune cron expression, timezone, channel
• Step 6 (Review) -- Final check before saving

Use the test panel to validate the generated query with live data. If the query is not quite right, you can edit it directly or regenerate with a refined prompt.

Limitations

AI generation works best for straightforward queries that map cleanly to your data. It may struggle with:

• Complex multi-table joins with more than 3 tables. Generate a starting point, then refine the SQL manually.
• Custom business logic that requires domain knowledge alice does not have. For example, your definition of "active user" may differ from what the AI assumes.
• Tables with ambiguous names. If multiple tables could match your description (e.g., users, profiles, accounts), the AI picks the most likely one. Verify the table selection in the Query step.

For complex queries, consider starting with a template or building the skill manually.

Next Steps

• Create a Skill -- Full walkthrough of the manual skill builder
• Skill Templates -- Browse 22 pre-built templates as alternatives to AI generation
• Combined Queries -- Build cross-source skills that AI can also generate
• Parameters -- Customize the parameters AI generated for your skill
• Connect Stripe -- Connect Stripe to enable Stripe-based AI generation
• Connect Supabase -- Connect Supabase so the AI can read your database schema
• Core Concepts -- Understand skills, sources, and triggers