## TL;DR

Four hundred line items exceed the agent's per-report budget (memory or time). Process the report's lines in chunks (for example, 50 at a time), aggregate the chunk results, and the large report completes like any other.

## Error

```text
agent crashed with error on a report with 400 line items
```

## Steps

1. Confirm the crash correlates with line-item count: small reports pass, large ones crash. Expected: a clear size threshold.
2. Chunk the line items: iterate in groups of 50, processing and releasing each group. Expected: memory and time stay flat per chunk.
3. Aggregate chunk results into the report-level verdict as you go. Expected: the final verdict covers all 400 lines.
4. Checkpoint progress per chunk so an interruption resumes mid-report. Expected: no reprocessing of finished chunks.
5. Add a guard: reports above a line count get the chunked path automatically. Expected: future large reports just work.

## When to use

- An agent crashes on reports with hundreds of line items.
- The crash is size-dependent, not content-dependent.

## When not to use

- The agent times out across many small reports (see the bulk-scan skill).
- One specific line item crashes the parser (quarantine that line).

## Tool compatibility

- Custom audit agents processing expense reports.

## Variant phrasings

### Agent crashed on large expense report

Same failure, generic wording.

### Too many line items for agent run

Names the size cause directly.

## Why it happens

Agents budget memory and time per report assuming typical sizes. A 400-line report blows past both budgets at once; chunking keeps each unit of work inside the budget.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_1x9rFjX-ZdiZ9raooc46PA
