{"version":"2.1.302","anchor":"agent-tool-description-now-advertises-subagent-type-fork-m","canonical_anchor":"agent-tool-description-now-advertises-subagent-type-fork-m","heading":"Agent tool instructions describe forking in more places when forking is on","tier":"internal","area":"Subagents","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.302\/e\/agent-tool-description-now-advertises-subagent-type-fork-m","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.302","markdown":"### Agent tool instructions describe forking in more places when forking is on\n\nThe Agent tool's instructions to Claude now have separate fork wording and a \"When to fork\" section, shown only when forking is enabled\n\n**Unclear.** What turns the fork version of these instructions on is not shown.\n\n**What**\n\nThe Agent tool lets Claude hand part of a task to a subagent, a helper copy of Claude. A fork is a subagent that starts from the current conversation. The instructions Claude receives about this tool have changed in how fork wording is chosen:\n\n- The fork wording no longer appears only together with the background-subagent text. Each has its own condition.\n\n- The background text has a fork version and a non-fork version.\n\n- When forking is on, the instructions include a \"When to fork\" section and say: \"A fork keeps its tool output out of your context. If you are the fork, execute directly - don't re-delegate.\"\n\n- The \"When to use\" text can be replaced by a value set for each agent.\n\nThe stock Agent tool description captured under this release is the same as in the previous capture, so this wording shows up only where forking is enabled.\n\n**Why**\n\nWhat Claude is told about forking shapes when it hands work off, and a fork is told to do the work itself rather than hand it on again.\n\n- Area: Subagents\n- Tier: Under the hood\n- Useful: 1\/5\n- Signal: 2\/5\n- Scope: individual\n- Heads-up: no"}