{"version":"2.1.314","anchor":"tool-denial-correction-text-and-truncation-continuation-prom","canonical_anchor":"tool-denial-correction-text-and-truncation-continuation-prom","heading":"New messages for stopped tool calls and replies cut off at the length limit","tier":"internal","area":"System Prompt","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.314\/e\/tool-denial-correction-text-and-truncation-continuation-prom","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.314","markdown":"### New messages for stopped tool calls and replies cut off at the length limit\n\nClaude Code has new text telling the model a stopped action may have partly run, and asking it to continue a cut-off reply\n\n**Unclear.** It is not clear whether these messages are already sent to the model or are only defined so far.\n\n**What**\n\nClaude Code contains two new messages written for the model:\n\n- A correction saying an action was stopped while it was running, not refused before it ran, so it may have run partly or fully.\n\n- A prompt to carry on after a reply was cut off for going over the output length limit.\n\n**Why**\n\nThese change what the model is told after an interrupted action. Telling it the action may have partly happened helps it avoid assuming nothing was done.\n\n- Area: System Prompt\n- Tier: Under the hood\n- Useful: 1\/5\n- Signal: 1\/5\n- Scope: individual\n- Heads-up: no"}