Skip to main content
ニュースに戻る
ニュース

Day-One Support for Claude Code Dynamic Workflows

When Claude Code fans a session out into many subagents, MadoHub keeps the canvas readable — one progress line per tile, no badge flooding, with token usage rolled into the Context Gauge.

Claude Code shipped Dynamic Workflows alongside Opus 4.8 on 2026-05-28 — its own mechanism for fanning a session out into many concurrent subagents, all inside a single tile. It's Claude Code's feature, not MadoHub's. What MadoHub ships today is compatibility: it watches for those subagent runs, keeps the canvas readable while they run, and folds their token usage into the numbers you already check — so a workflow with dozens of subagents doesn't bury your screen under notifications.

What you can do now

  • See workflow progress per tile, not per subagent — a single line like Review 3/5 under the tile tells you how many subagents have finished out of how many have started, without opening the tile.
  • Avoid badge flooding — a workflow that spins up dozens of subagents shows as one tidy progress indicator instead of stacking a notification per subagent.
  • Spot an active workflow at a glance — the tile's title bar shows a small pulsing dot next to the Context Gauge while a workflow is running.
  • Track workflow token usage separately — subagents run in their own contexts, so their tokens are reported as a separate figure in the Context Gauge tooltip rather than mixed into your main session's budget.

How to try it

  1. Run a Claude Code tile and trigger a Dynamic Workflow (Claude Code's own command, e.g. an ultracode-style run).
  2. Watch the tile's title bar for the pulsing workflow dot, and Mission Control for a doneCount/totalSeen line like Review 3/5.
  3. Hover the Context Gauge to see workflow token usage as its own figure, separate from your main session's context.

Why it matters / use case

Without this, a single workflow run would look indistinguishable from dozens of agents finishing at once — each subagent reaching a terminal state triggering the same "agent done" notification a normal tile does. On a 65-subagent workflow, that's 65 stacked notifications on one tile. MadoHub recognizes the workflow shape and rolls all that activity into one progress line, so you can keep an eye on a big run without dismissing an endless stream of badges, and you can tell how full your main session's context actually is without subagent tokens inflating the number.

What's next / limitations

Detection is pattern-based and deliberately defensive — if Claude Code changes its workflow file format, MadoHub degrades gracefully instead of breaking. Workflow token usage is shown separately because subagents have their own context windows; folding them into your main session's percentage would misrepresent how full that session actually is.