{"version":"2.1.284","anchor":"context-management-edits-can-carry-tool-clearing","canonical_anchor":"context-management-edits-can-carry-tool-clearing","heading":"Requests to the model can now ask for old tool results to be cleared","tier":"internal","area":"Context Management","url":"https:\/\/changelogs.core-directive.com\/v\/2.1.284\/e\/context-management-edits-can-carry-tool-clearing","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.284","markdown":"### Requests to the model can now ask for old tool results to be cleared\n\nClaude Code's requests can now include an instruction to clear old tool results from context, not only old thinking\n\n**Unclear.** It is not clear what supplies the tool-clearing instruction or whether a flag controls it.\n\n**What**\n\nClaude Code can send context management instructions with its requests to the model. These tell the server what to drop from the conversation to save room. Before, the only one clears Claude's earlier thinking. The requests can now also carry a tool-clearing instruction. Tool results are the output Claude gets back when it runs a command or reads a file.\n\n**Why**\n\nThe request format can now ask for old tool output to be removed as well as old thinking. Whether this happens in any session depends on what supplies the instruction.\n\n- Area: Context Management\n- Tier: Under the hood\n- Useful: 3\/5\n- Signal: 4\/5"}