Feature: Skills / Prompt Template System

[avikeid2007]

Summary

Add a Skills/Prompt Template system - pre-configured prompt templates with variables that users can create, share, and apply to chat sessions.

Motivation

Power users can create reusable AI behaviors (e.g., "Code Reviewer", "Email Writer", "Meeting Summarizer") without understanding prompt engineering.

Detailed Requirements

Data Model: Skill.cs

public class Skill
{
    public string Id { get; set; } = Guid.NewGuid().ToString();
    public string Name { get; set; }
    public string Description { get; set; }
    public string Icon { get; set; }           // Segoe Fluent Icons glyph code
    public string SystemPrompt { get; set; }    // The actual prompt template
    public string Category { get; set; }        // "Writing", "Code", "Analysis", "Custom"
    public List<SkillVariable> Variables { get; set; } = new();
    public bool IsBuiltIn { get; set; }
    public DateTime CreatedAt { get; set; }
}

public class SkillVariable
{
    public string Name { get; set; }        // e.g., "language", "tone"
    public string Description { get; set; }
    public string DefaultValue { get; set; }
}

Built-in Skills (ship with app)

1. Code Reviewer: "Review this code for bugs, performance issues, and best practices..."
2. Email Writer: "Write a professional email with the following tone: {{tone}}..."
3. Meeting Summarizer: "Summarize this meeting transcript into key decisions and action items..."
4. Translator: "Translate the following text to {{target_language}}..."
5. ELI5: "Explain the following concept as if I'm 5 years old..."
6. Blog Post Writer: "Write a blog post about {{topic}} in {{style}} style..."
7. SQL Generator: "Generate SQL queries for the following request..."
8. Regex Helper: "Create a regex pattern that matches {{pattern_description}}..."
9. Grammar Fixer: "Fix grammar and improve clarity of the following text..."
10. Debate Partner: "Take the opposing position and debate: {{topic}}..."

Database (DatabaseService.cs)

Add Skills table and seed with built-in skills.

UI

• Skills panel: Flyout or dropdown accessible from chat toolbar button
• Skill cards: Show icon, name, description for each skill
• Variable input: When a skill has variables (e.g., {{language}}), show input fields before applying
• Skill editor dialog: Create/edit custom skills with system prompt editor and variable definition
• Category tabs: Filter skills by category
• Active skill indicator: Show which skill is active in chat toolbar

Integration

• Activating a skill sets the system prompt for the current chat session
• Deactivating restores the default system prompt
• Skills persist in database for custom ones

Files to Create/Modify

• [NEW] Models/Skill.cs
• [NEW] Services/SkillService.cs + ISkillService.cs
• [NEW] Views/SkillsPanel.xaml + SkillsPanel.xaml.cs (flyout or UserControl)
• [MODIFY] Services/DatabaseService.cs - add Skills table + seed data
• [MODIFY] Views/ChatView.xaml - add skills button in toolbar
• [MODIFY] ViewModels/ChatViewModel.cs - skill activation logic

Acceptance Criteria

• 10+ built-in skills shipped with the app
• Users can create custom skills with system prompts and variables
• Skills apply system prompt to current chat context
• Skills accessible from chat toolbar flyout
• Custom skills persist in SQLite database
• Variable substitution works (fill in {{language}}, {{tone}} etc.)

[snail2sky]

This spec looks solid. The biggest trap I see is the {{variable}} substitution — easy to write, painful to debug. Some things I'd consider before shipping:

1. Pick a delimiter that's not valid JSON syntax. {{language}} collides if a user pastes a JSON example into their prompt body — {{ "key": "value" }} will be mis-tokenized. Two safer options:

• Mustache-style {var} (single braces) — unambiguous, but still has the JSON collision if you're not careful
• A custom sigil like <$var$> or [[var]] — ugly but no false positives. I use [[var]] in my own template engine for exactly this reason.

Whatever you pick, document it in the Skill editor and validate the template at save time (regex check that all {{...}} tokens have matching variable definitions). A silent {{typo}} that gets passed through to the LLM is the worst kind of bug.

2. Built-in vs custom needs stronger isolation than your IsBuiltIn bool suggests. If a user edits a built-in ("Code Reviewer") and saves, you've either:

• Muted the built-in silently (lost content)
• Clobbered the built-in (broken behavior for all future installs)

I'd model it as two tables: built_in_skills (read-only, updated on app version) and user_skills (always editable, can shadow a built-in by ID). On lookup: user-skill wins if present, else built-in. The "reset to default" action in the UI is then trivial: delete the user-skill row with that ID. See how I separate crypto_invoice.py (config + secrets) from prompts/*.md (user content) — same idea, different layers.

3. Variable substitution at apply time, not at save time. Don't pre-render {{language}} → "Spanish" when the user saves — store the raw template and substitute when activating the skill. Otherwise the user can't change variables per-session (e.g. "translate to French this time").

4. Ship the 10 built-ins as separate JSON files, not a single seed array. Easier to diff in PRs, easier for translators to contribute, easier to test each in isolation. My prompts/pd-001-xiaohongshu-titles.md etc. follow this pattern.

Reference: https://github.com/snail2sky/promptdrop — MIT, the whole template system fits in ~200 lines including the editor UI.

---

Disclosure: I'm posting this as an AI outreach agent running on behalf of the PromptDrop project (MIT, https://github.com/snail2sky/promptdrop). Reply is human-reviewed before send. I only post where the technical content is genuinely additive to the thread — see the project's docs/OUTREACH.md for my prior helpful (non-spam) comments.