What Is “ultracode” Mode in Claude Code? Explained

The word ultracode — highlighted in violet — has been turning up inside Claude Code prompts, and search interest in “Ultra Code Mode” has spiked along with it. One clarification first: Anthropic's own docs and changelog never use the spelled-out phrase “Ultra Code Mode.” The real feature is a Claude Code setting called ultracode, and per the Claude Code workflows documentation it pairs xhigh reasoning effort with automatic workflow orchestration. Switch it on and Claude plans a workflow for each substantive task on its own, instead of waiting for you to ask. This is the focused explainer of how that works — not a rehash of the May 28, 2026 launch of Opus 4.8 and Dynamic Workflows, which BitsMinds has already covered.

There are two ways to switch it on. To hand the whole session over to Claude's judgment, run /effort ultracode — from then on it plans a workflow for every substantive task until you change the effort level. To run a single task as a workflow without touching the session's effort, just include the keyword ultracode in your prompt; asking in plain words, like “use a workflow,” does exactly the same thing. That trigger keyword was renamed from workflow to ultracode in v2.1.160 (June 2, 2026), and natural-language requests work in either version. If the violet highlight gets in your way, press Alt+W on Windows or Linux (Option+W on macOS) or backspace right after the keyword, and you can disable the trigger entirely from /config.

Under the hood, this is Dynamic Workflows. As the Claude.com announcement puts it, Claude “dynamically writes orchestration scripts that run tens to hundreds of parallel subagents in a single session.” That script is JavaScript, and a runtime executes it in the background while your session stays responsive. Claude breaks your prompt into subtasks and fans them out across subagents, and the verification is adversarial: some agents attack the problem from independent angles while others try to refute what they found, iterating until the results converge. Because the loop, the branching and the intermediate results all live in script variables rather than in the conversation, Claude's context holds only the final answer — which is what lets a plan “stay on track no matter how big the task gets.” The runtime caps work at 16 concurrent agents (fewer on machines with limited CPU cores) and 1,000 agents total per run to prevent runaway loops; every run writes its script to a file under ~/.claude/projects/ and can be resumed within the same session. A few hard limits are worth internalizing:

• No mid-run user input — only agent permission prompts can pause a run.
• Spawned subagents always run in acceptEdits mode and inherit your tool allowlist regardless of your session's mode, so file edits are auto-approved.
• Resume is session-bound — exit Claude Code mid-run and the next session restarts the workflow fresh.
• Cost and time scale up — Anthropic warns Dynamic Workflows can burn through substantially more tokens than a typical session, and each request takes longer.

On availability the sources mostly agree, with one caveat worth stating conservatively. Dynamic Workflows are a research preview that needs Claude Code v2.1.154 or later, and they work across the CLI, Desktop app, IDE extensions, non-interactive claude -p and the Agent SDK, plus the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. The launch blog and InfoQ list Max, Team and admin-enabled Enterprise plans; the workflows doc additionally lists Pro (“on Pro, turn them on from the Dynamic workflows row in /config”) — a claim the launch announcement does not echo. Either way, ultracode only appears on models that support xhigh; on the others, the /effort menu does not offer it.

So when should you reach for it? The good fits are exactly the jobs too big for a single agent: codebase-wide migrations, security audits, large refactors, performance reviews, hunting widespread bugs and deep research — work the launch blog says can stretch into hours or even days. Claude Code even ships /deep-research as a bundled workflow that fans web searches across angles, cross-checks sources, votes on each claim and drops the ones that do not survive. Knowing when not to use it matters just as much: for routine edits the workflows doc is blunt — “drop back with /effort high when you return to routine work” — and InfoQ suggests starting with smaller, well-scoped tasks before pointing it at a large project.

One last piece of context. Dynamic Workflows shipped on May 28 in v2.1.154 alongside Opus 4.8; ultracode is the convenience layer added on top, with the keyword rename and the /effort ultracode wiring landing later, in the June 2 v2.1.160 release.