{"version":"2.1.307","anchor":"the-agent-tool-description-in-the-command-security-review-pr","canonical_anchor":"the-agent-tool-description-in-the-command-security-review-pr","heading":"Agent tool guidance rewritten in the security review prompt","tier":"notice","area":"Subagents","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.307\/e\/the-agent-tool-description-in-the-command-security-review-pr","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.307","markdown":"### Agent tool guidance rewritten in the security review prompt\n\nThe instructions telling Claude when to hand work to a helper agent were reworded in the security review prompt, including a warning about token cost\n\n**What**\n\nThe description of the `Agent` tool was rewritten in the prompt Claude Code sends for the security review command. The `Agent` tool is how Claude hands a piece of work to a subagent, a separate helper copy of Claude that works on it and reports back.\n\nThe old text began by telling Claude to use the tool when a task matched an available agent type, when work could run in parallel, or when an answer meant reading across several files. The new text adds a caution about starting several helpers at once:\n\n- \"Several at once spend tokens in a burst the user didn't ask for.\"\n\nTokens are the units of text Claude reads and writes, which is what usage is measured in.\n\n**Why**\n\nThis text is part of Claude's own instructions, so rewording it changes how Claude decides whether to delegate work. In practice the new wording points Claude away from launching many helpers at the same time without being asked.\n\n- Area: Subagents\n- Tier: You'll notice\n- Useful: 1\/5\n- Signal: 1\/5\n- Scope: individual\n- Heads-up: no"}