Drill · lap 1 · 5/9 · MEM-101 · The summariser that runs every turn

[akash-coded]

5 of 9 · MEM-101 · The summariser that runs every turn

memory · medium · fix · ~12 min · no AWS account

This history trimmer works. Conversations stay under budget and nothing crashes. It also costs a model call on every single turn, forever, and on long conversations it deletes the message the user just sent.

def trim_history(messages, budget_tokens, summarise, keep_recent=4):
    older, recent = messages[:-keep_recent], messages[-keep_recent:]
    summary = summarise(older)                      # collapse the past
    msgs = [summary] + recent
    while sum(m["tokens"] for m in msgs) > budget_tokens:
        msgs.pop()                                  # drop until it fits
    return msgs

summarise(msgs) is provided by the caller and costs a model call. Messages carry a precomputed tokens int.

Find both problems

One is about when work happens. The other is about which end of the list gets cut. Fix the function so that:

• summarise is called only when the history is over budget, and at most once
• the newest message is never removed
• under budget, the input is returned unchanged

Post the fixed function:

/drill MEM-101

```python
def trim_history(messages, budget_tokens, summarise, keep_recent=4):
    ...
```

What this proves

That you can see a permanent tax hiding inside a correct-looking function — and that you know which message in a buffer is the one you must never evict.

---

Submit in this thread

Copy the block from the brief, edit it, and post it as a comment here. The bot replies within a couple of minutes with every check, the skill you demonstrated, and what to do next.

_{Brief in the repo: labs/drills/memory/MEM-101 · next in the sequence: EVAL-101 · how grading works · scoreboard}