> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cometly.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Prompts Overview

> Reusable prompt templates clients surface as slash-commands — pre-filled multi-step analyses.

MCP **Prompts** are templates the client can offer to users as one-click slash-commands. Each prompt embeds a multi-step instruction the agent follows, calling the right tools in the right order.

## Available prompts

| Name                         | Description                                                                            | Arguments                                                  |
| ---------------------------- | -------------------------------------------------------------------------------------- | ---------------------------------------------------------- |
| `analyze_performance`        | Analyze ad performance for a date range across one or more sources, focused on a goal. | `date_range` (required), `sources`, `goal`                 |
| `compare_attribution_models` | Show the same KPIs across multiple attribution models.                                 | `date_range` (required), `metrics` (required)              |
| `weekly_review`              | One-shot dashboard-style summary for the most recent complete week vs the prior week.  | None                                                       |
| `find_underperformers`       | Surface ads / adsets / campaigns above a CPA threshold or below a ROAS threshold.      | `date_range` (required), `cpa_threshold`, `roas_threshold` |

## How they work

Each prompt expands into a structured instruction that drives the agent through a series of tool calls. For example, `weekly_review` resolves to:

```
Generate a weekly performance review.

Steps:
1. Call resolve_date_range with `preset: last_week`.
2. Call query_ad_metrics_compare with `compare_to: previous_period`.
3. Call query_metric_breakdown grouped by source.
4. Write a Key Metrics + Insights + Recommendations summary suitable for a Monday-morning Slack post.
```

The agent then executes those steps and writes the final summary.

## Invoking a prompt

Via MCP `prompts/get`:

```json theme={null}
{
  "method": "prompts/get",
  "params": {
    "name": "analyze_performance",
    "arguments": {
      "date_range": "last 30 days",
      "sources":    "facebook_ads,google_ads",
      "goal":       "decrease CPA"
    }
  }
}
```

The returned `messages` array is the multi-turn template the agent receives.

## Cost

Prompts themselves are free — they're text templates. The tool calls the agent runs as a result of the prompt are subject to the usual rate limits and (for `analyze_report`) AI credit deduction.