# Clipping Tool Results — Agent Loops, Stop Conditions and Token Budgets

Source: https://www.geekswithgeeks.com/en/agent-loops/ctx-clip-results

> Cut long tool outputs to a safe size while telling the model what was removed.

## Big outputs are the usual culprit

One careless tool result, like a 50,000-line log file, can use more tokens than the rest of the conversation combined, and it is then **resent on every later turn**. Clip results to a limit and say so, so the model knows to ask for a narrower view.

## Keep only what matters

Trim, summarise and offload so the history stays useful without growing forever.

![Three tools: clip results, summarise old turns, store externally.](assets/figures/agent-loops/section-4-map.svg) — Figure 4.1 — Clip, summarise and offload.

## A clip helper

This ran as shown: the output ends with a note saying how many characters were removed.

```python
def clip(text, limit=200):
    if len(text) <= limit:
        return text
    return text[:limit] + f"\n...[truncated {len(text) - limit} chars]"

print(clip("x" * 250, 200)[-30:])
```

Output:

```
xxxxxx
...[truncated 50 chars]
```

## Make tools paginate

Better than clipping is a tool that returns the first page and a way to ask for more, such as `offset` and `limit` parameters, so the model pulls only what it needs.

**Quiz:** Why is one huge tool result so costly?

- [x] It is resent on every later turn
- [ ] Large results are encrypted
- [ ] It slows the keyboard
- [ ] It is billed once only

*Answer:* It is resent on every later turn. History is resent each turn, so a big result keeps costing tokens until removed.