Heidelberg AICurriculum

Claude Code Dynamic Workflows Explained for Beginners

Leon van Zyl ·2026-06-09 ·1 min read

Summary written by us from the video's transcript. The video, and everything in it, is Leon van Zyl's work.

Learn how to create and use Claude Code dynamic workflows for scaling repetitive tasks, with best‑practice tips and a step‑by‑step security audit example.

Takeaways

  • Dynamic workflows let Claude Code fan out hundreds of parallel subagents, enabling large‑scale repetitive processing.
  • Only use them for repeatable tasks at scale; simple jobs are cheaper in a single session.
  • Begin prompts with **Ultra Code** and explicitly request a workflow, then test on a small batch before full execution.
  • Specify output locations and formats up front so each agent writes its results consistently.
  • Save workflows via the /run workflows UI to reuse or share across projects.

What Dynamic Workflows Are

Claude Code can generate its own orchestration script that launches hundreds of parallel subagents in a single session. This lets you process many items—like analyzing video transcripts or auditing code—simultaneously, something a normal Claude Code session cannot do.

When to Use (and Avoid) Dynamic Workflows

Use them for tasks that must be repeated at scale, such as handling tens or hundreds of emails, support tickets, or security checks. Avoid them for simple, single‑step jobs because each subagent incurs its own system prompts and token usage, making workflows expensive.

Starting a Workflow – Prompt Syntax

Begin your prompt with **Ultra Code** followed by “use a workflow for the following…”. Then describe the repeated action, any splitting logic, and a small test batch (e.g., first two checks) to verify correctness before scaling up.

Designing a Real‑World Example: OWASP Top 10 Audit

The speaker demonstrates an audit that checks each of the ten OWASP categories across a codebase. For each category, agents receive a description, scan the code, and write findings to markdown files in a designated folder. The workflow includes steps named recon, audit, verify, and report.

Managing Parallel Agents and Output

Agents run in parallel (e.g., two agents checking different OWASP items). After audits finish, each agent verifies results and writes its own report file. If many agents modify the same codebase, add instructions to work on separate branches or trees to avoid contention.

Saving, Sharing, and Reusing Workflows

Press **/run workflows** to view active workflows, then press **S** to save one with a name like “OWASP audit”. Saved workflows appear in the .claude folder under workflows, where they can be shared with teammates or reused by starting a new Claude Code session and invoking the saved workflow.

Best‑Practice Checklist

- Start with a tiny batch to confirm behavior. - Define clear final output format (e.g., markdown reports). - Avoid human‑in‑the‑loop tasks; workflows run unattended. - Plan for file‑conflict avoidance when many agents edit the same repository.