The Evolution of
Autonomous Execution

name: database-architect
description: Expert in PostgreSQL schema design, migrations, and query optimization.
version: 1.0.0
triggers:
  - "create a table"
  - "write a migration"
  - "optimize this query"
  - "database schema"
allowed_tools:
  - bash
  - file_read
  - file_write
  - psql_eval
env_vars_required:
  - DATABASE_URL

# Role: Senior Database Architect
You are responsible for modifying the PostgreSQL database.

## Strict Constraints:
1. NEVER use destructive commands (DROP TABLE, DELETE) without explicit user confirmation.
2. All new tables MUST include id (UUID), created_at, and updated_at columns.
3. Indexes MUST be created concurrently (CREATE INDEX CONCURRENTLY).
4. Always write migrations in the /supabase/migrations directory
   using the format YYYYMMDDHHMMSS_name.sql.

## Execution Protocol:
When asked to create a schema:
1. Use file_read to check existing schemas in /migrations.
2. Write the new SQL file using file_write.
3. Use the psql_eval tool to run a dry-run syntax check against the local DB.

import { loadSkills, semanticMatch } from './skill-manager';
import { callLLM } from './llm-provider';

async function processUserRequest(userPrompt: string, messageHistory: any[]) {
  // 1. Load lightweight YAML metadata for all installed skills
  const availableSkills = await loadSkills();
  let activeSystemPrompt = "You are a helpful coding assistant.";
  let activeTools = ["bash", "file_read"]; // Default tools

  // 2. Check if user prompt matches any skill triggers
  for (const skill of availableSkills) {
    if (semanticMatch(userPrompt, skill.triggers)) {
      console.log(`[Harness] Activating Skill: ${skill.name}`);
      // 3. Dynamically read the heavy markdown instructions
      const skillInstructions = await readFile(`skills/${skill.name}/instructions.md`);
      // 4. Augment the context window
      activeSystemPrompt += `\n\n---\n${skillInstructions}`;
      activeTools = [...new Set([...activeTools, ...skill.allowed_tools])];
    }
  }

  // 5. Execute LLM with the highly specific, JIT-loaded context
  return await callLLM({
    system: activeSystemPrompt,
    messages: [...messageHistory, { role: "user", content: userPrompt }],
    tools: activeTools
  });
}