Files
Orchestrator/.agents/skills/model_advisor/SKILL.md
T

36 lines
1.6 KiB
Markdown

---
name: model_advisor
description: Evaluates a task and orchestrates execution across available CLI models based on cost, quotas, and strengths.
---
# Instructions
You have been invoked to act as a central orchestrator. You must evaluate the task, check available models and quotas, and delegate the work via CLI.
## Step 1: Check Quotas & Models
Run the following scripts from your workspace:
- `.agents/scripts/check_quotas.sh`
- Review `.agents/models_config.json` to see currently available models.
## Step 2: Evaluate the Task
Consider the following criteria:
1. **Complexity & Context**: Does it require massive context (Gemini Pro) or deep nuance (Claude Sonnet)?
2. **Cost & Quota**: Are quotas low? Use cheaper models if the task is simple.
## Step 3: Output a Recommendation & Delegate
If the task is best suited for a CLI-driven model (e.g., Codex or Antigravity with a specific model), you MUST proactively write and execute the terminal command to delegate the task.
Example output:
> [!TIP]
> **Model Advisor Recommendation**: The quota for Antigravity is healthy. I am delegating this refactoring task to Claude 3.5 Sonnet via the CLI.
Example command you might run:
```bash
agy run --model claude-3-5-sonnet "refactor the payment gateway"
```
If the task is best suited for the *current* IDE model, advise the user to proceed normally without CLI delegation. If it's suited for an IDE model you are *not* currently using, advise the user to switch models manually in the IDE.
## Step 4: Proceed
Once the CLI command completes, or if you decided to handle it locally, complete the user's request.