Anthropic's Claude-Code native binary update hints at a new `tool_runner` for agentic loop

EXHIBIT 098 · FLASH

Jul 8, 2026 01:01 UTC binary @anthropic-ai/claude-code-darwin-arm64

Anthropic's Claude-Code native binary update hints at a new `tool_runner` for agentic loops with per-turn hooks, a new `ant` CLI for API access and agent creation from YAML, and expanded tool schema options.

✓ VERIFIED — BEAT THE ANNOUNCEMENT BY 36.5H · AUTO-VERIFIED AGAINST ANTHROPIC NEWS

READY TO POST:

anthropic's latest claude-code update seems to be building out a `tool_runner` for agentic loops, with hooks for approvals, error handling, and streaming. this could make building complex agents much smoother!

a new `ant` cli for claude is surfacing in their latest binary. looks like it'll allow for creating agents and environments directly from version-controlled yaml. super interesting for infra-as-code fans.

⚠️ FROM COMPILED BINARY — lower confidence, verify before posting. (May include bundled-dependency strings, not just first-party changes.) @anthropic-ai/claude-code-darwin-arm64 2.1.202 -> 2.1.204: +6282 strings, -6603 strings, 34 interesting [feature_flag] (10) + - Tool Runner (\`client.beta.messages.tool_runner\`): the SDK helper that runs the agentic loop over tools you define \u2014 with per-turn hooks for approval gates, error interception, result modification, retries, and streaming (you do NOT need the manual loop for those) + /home/runner/work/claude-cli-internal/claude-cli-internal/src/frame + <${$f}>${n.length} background ${r} task(s) from the previous session have no completion record. ${l} They have been marked ${e}. First ${hcn} task ids: ${s.join(", ")}. Task ids in this notification beginning with "${$Fo}" are internal scan markers, not tasks.</${$f}> + Emitted by /clear, plan-mode exit, and fresh-session flows. The surface should mount a fresh transcript under new_conversation_id and reset any cached session title. From internal QueryEvent 'conversation_reset'. + Emitted when the user's /goal Stop hook reports met (clears) or not-yet-met (bumps iterations + last_reason). Any surface with a goal indicator re-renders from this. value is null when the goal is cleared. From internal QueryEvent 'active_goal'. + See: https://chevrotain.io/docs/guide/internals.html#grammar-recording for details`,-1,-1,-1,-1,-1,-1);Object.freeze(Lke);Z0t={name:`This CSTNode indicates the Parser is in Recording Phase + The \`ant\` CLI provides terminal access to the Claude API. Every API resource is exposed as a subcommand. It is the recommended way to create agents and environments from version-controlled YAML (\`ant beta:agents create < agent.yaml\` \u2014 see \`shared/anthropic-cli.md\`), and also exposes sessi + Zod is optional \u2014 \`betaTool()\` from \`@anthropic-ai/sdk/helpers/beta/json-schema\` accepts a raw JSON Schema \`inputSchema\` plus a \`run