Skip to content

Prompting Best Practices

Make your LLM reliable. Stop the retry loops.

Two rules prevent 90% of tool-calling problems. Add them to your system prompt.

System Prompt Guidelines

Include these instructions in your system prompt:

system_prompt: |
  Never retry successful tool calls to get "better" results.
  If a tool call fails, report the error - do not compute the result yourself.

Why these matter:

  • No retries on success: LLMs sometimes want to "improve" results by calling the same tool again. This wastes tokens and can cause loops.
  • No manual computation on failure: When a tool fails, LLMs often try to compute the answer themselves (e.g., calculating a hash). This defeats the purpose of using tools and may produce incorrect results.

Pre-Call Instructions

Add context before tool calls to guide the LLM:

Calculate the SHA-256 hash of the following text:
__onetool__run sha256(text="hello world")

Post-Call Processing

Request specific formatting after tool results:

__onetool__run brave.search(query="latest AI news", count=5)

Summarise the top 3 results in bullet points.

Structured Output

Combine tool execution with output formatting:

__onetool__run
```python
results = {
    "hash": sha256(text="hello"),
    "reversed": reverse(text="hello"),
    "length": count_chars(text="hello")
}
results

Return the results as a markdown table. 

## Batch Operations

For multiple related queries, use batch functions:

```python
# Instead of multiple calls
brave.search(query="topic 1")
brave.search(query="topic 2")

# Use batch
brave.search_batch(queries=["topic 1", "topic 2"])

Error Handling

When tools return errors:

  1. Report the error - Don't try to work around it
  2. Check prerequisites - API keys, dependencies
  3. Validate inputs - Query length, parameter types

Token Efficiency

  • Use short prefixes (__ot) for simple calls
  • Use code fences for multi-step operations
  • Prefer batch operations over multiple calls