# `Condukt.Compactor.ToolResultPrune`
[🔗](https://github.com/tuist/condukt/blob/0.16.5/lib/condukt/compactor/tool_result_prune.ex#L1)

Replaces the content of old tool result messages with a small placeholder,
preserving the message structure (and the assistant's tool call reasoning)
so the model still sees that the tool was invoked.

This is the cheapest compaction strategy: no turns are dropped, only
oversized historical payloads (large file reads, command output, etc.) are
elided. Often enough on its own for coding agents.

## Options

- `:keep_recent` - number of trailing tool results to leave intact
  (default: `5`).
- `:max_size` - tool result payloads larger than this many bytes are elided
  when they fall outside the recent window (default: `4096`).

## Example

    MyApp.Agent.start_link(
      compactor: {Condukt.Compactor.ToolResultPrune, keep_recent: 10, max_size: 2_000}
    )

---

*Consult [api-reference.md](api-reference.md) for complete listing*