export const OPTIMIZATION_SKILL = "# Iris Optimization Skill\n\n## Context\nThis project has the `@foxruv/iris` optimization engine installed. This engine provides a pluggable architecture for hyperparameter and prompt optimization, prioritizing:\n1. **Ax (Bayesian Optimization)** - For numeric/structural parameters.\n2. **DSPy (MIPROv2)** - For prompt/LLM optimization.\n3. **Grid Search** - For simple fallbacks.\n\n## Your Role\nWhen the user mentions \"optimization\", \"tuning\", or \"improving performance\", you must:\n\n1. **Analyze the Target:**\n   - Identify the script or function needing optimization.\n   - Look for an exported `evaluate(params)` function.\n\n2. **Check Configuration (`iris-config.yaml`):**\n   - If no config exists, propose creating one using the template below.\n   - Respect the `strategy` order (default: `['ax', 'dspy', 'grid']`).\n\n3. **Select the Strategy:**\n   - **Prefer Ax** if the project has a Python environment and numeric parameters.\n   - **Prefer DSPy** if optimizing prompts/text.\n   - **Use Grid** if the search space is small (<20 combinations) or dependencies are missing.\n\n4. **Execute:**\n   - Use the CLI: `npx iris optimize --config <config.yaml> --target <script.ts>`\n   - Do NOT write custom optimization loops unless the CLI is insufficient.\n\n## Configuration Template\n```yaml\noptimization:\n  strategy: ['ax', 'grid']\n  searchSpace:\n    parameters:\n      - name: \"learning_rate\"\n        type: \"range\"\n        bounds: [0.0001, 0.1]\n        log_scale: true\n```\n\n## Dependency Check\nBefore running Ax or DSPy, verify dependencies:\n- Ax: `pip install ax-platform` (Python service must be running).\n- DSPy: `pip install dspy-ai`.\n";

export const GEMINI_OPTIMIZATION_SKILL = "# Gemini Optimization Protocol\n\n## System Capabilities\nYou are interacting with a project enabled with the `@foxruv/iris` Optimization Engine. This engine allows you to autonomously tune hyperparameters and prompts using:\n- **Ax** (Bayesian Optimization)\n- **DSPy** (Prompt Optimization)\n- **Grid Search** (Fallback)\n\n## Your Optimization Mandate\nIf the user asks to \"optimize\", \"tune\", or \"improve\" any component:\n\n1.  **Discovery Phase:**\n    *   Scan the codebase for the target script.\n    *   Verify if it exports an `evaluate(params)` function.\n    *   Check for `iris-config.yaml`. If missing, generate one using the standard template.\n\n2.  **Strategy Selection (Strict Priority):**\n    *   **Priority 1: Ax.** Check if `ax-platform` is installed and the Python service is running. If yes, configure `iris-config.yaml` to use `['ax', 'grid']`.\n    *   **Priority 2: DSPy.** If optimizing text/prompts, check for `dspy-ai`. Use `['dspy', 'grid']`.\n    *   **Priority 3: Grid Search.** Use this if dependencies are missing or the search space is small.\n\n3.  **Execution:**\n    *   Construct the CLI command: `npx iris optimize --config <config> --target <script>`\n    *   Run the command using your shell tool.\n    *   Analyze the JSON output (Best Score/Configuration).\n\n4.  **Refinement:**\n    *   If the score is unsatisfactory, propose widening the search bounds in `iris-config.yaml` and re-running.\n\n## Config Reference\n```yaml\noptimization:\n  strategy: ['ax', 'dspy', 'grid']\n  searchSpace:\n    parameters:\n      - name: \"param_x\"\n        type: \"range\"\n        bounds: [0, 10]\n```\n";