Skip to main content

3 Prompts to Test for Agent Readiness

· 3 min read
Angie Jones
Angie Jones
Head of Developer Relations

blog cover

Goose is LLM-agnostic, meaning you can plug in the model of your choice. However, not every LLM is suitable to work with agents. Some may be great at answering things, but not actually doing things. If you're considering which model to use with an agent, these 3 prompts can quickly give you a sense of the model's capabilities.

Tool Calling​

This initial prompt tests for tool calling capabilities. Its ask is forceful to reduce a shy model's hesitation to make function calls.

Create a file at ~/workspace/loose-goose/tool-test.txt with the contents "Hello World".

Use the write tool. Do not ask for confirmation. Just do it.

✅ tool-test.txt was created

❌ the agent responds by telling you the code to write yourself

Example of successful response

─── text_editor | developer ──────────────────────────
path: ~/workspace/loose-goose/tool-test.txt
command: write
file_text: Hello World

The file has been created successfully with the following content:

"Hello World"

The model emits a structured tool call in JSON.

Memory Awareness​

Next, test whether the agent can recall what it’s doing. It's critical that the model can remember previous actions and continues logically.

Now append a new line that says: "I know what I'm doing"

✅ tool-test.txt was updated

❌ the agent responds by asking you which file

Example of successful response

─── text_editor | developer ──────────────────────────
path: ~/workspace/loose-goose/tool-test.txt
command: write
file_text: Hello World
I know what I'm doing

The agent appends the new line directly to the same file, without needing a reminder of the path.

File system reasoning​

The last prompt tests whether the model can infer file locations by resolving relative and absolute paths based on context. You don't want the agent deleting important directories because the model is hallucinating about where it is.

What is the current content of tool-test.txt?

✅ content of tool-test.txt

❌ confusion about where to find the file

Example of successful response

─── text_editor | developer ──────────────────────────
path: ~/workspace/loose-goose/tool-test.txt
command: read

Hello World
I know what I'm doing

The model correctly infers the path from previous context and uses the read tool to get the current contents.

If a model passes this multi-turn prompt sequence, it's safe to assume that it is suitable for agentic AI.