Official
Orchestrate subagents at scale with dynamic workflows
The canonical guide for when to use workflows, how `/deep-research` runs, how `/workflows` exposes progress, and how saved workflows become slash commands.
Claude Code DocsOriginal
Claude Code Dynamic Workflows research preview
A living hub for agentic workflows, not imaginary CLI commands.
workflows.fit tracks Claude Code dynamic workflows today, then leaves room for Pi, Codex, Antigravity, and other workflow runtimes as the pattern spreads. Current Claude workflows are launched inside Claude Code through prompts and slash commands such as /deep-research, viewed with /workflows, and saved as reusable /<name> commands.
/deep-research What changed in dynamic workflows?
CURRENTRuntime Map
Designed for a multi-agent future
Claude Code
Live research preview
Pi
Community runtime experiments
Codex
Watchlist
Antigravity
Watchlist
The site model now separates runtime, article resources, translated reading notes, tutorials, and reusable workflow examples so new ecosystems can be added without rewriting the whole information architecture.
Parallel Orchestration
Coordinate many specialized subagents for audits, migrations, and research where independent branches can be checked before they converge.
Custom Scripting Runtimes
Treat workflows as JavaScript orchestration: state, loops, branching, and intermediate results live in the runtime instead of the main chat context.
Adversarial Verification
Use independent reviewers, source cross-checking, and test loops to filter weak findings before the final report reaches the user.
Articles and sources
Official and community reading shelf
Curated source links with hosted multilingual translation pages. English originals expose Chinese, Spanish, French, Japanese, and German versions so readers do not have to leave the site just to understand the material.
Official
Introducing dynamic workflows in Claude Code
Announcement post positioning dynamic workflows as research-preview orchestration for large migrations, audits, and high-stakes work that benefits from independent verification.
Claude BlogOriginal
Case Study
Rewriting Bun with dynamic workflows
The most visible migration case: a large Zig-to-Rust port shaped as staged mapping, generation, reviewer agents, and build/test repair loops.
Claude Blog and HN discussionOriginal
Community
Claude Code Dynamic Workflow hands-on notes
A practical look at `/deep-research`, including the research workflow shape of scope, search, fetch, verify, and synthesize.
azukiazusa.devOriginal
Ecosystem
Pi dynamic workflows runtime experiment
A community implementation for Pi that explores primitives such as `agent()`, `parallel()`, `pipeline()`, and `phase()`.
Michaelliv/pi-dynamic-workflowsOriginal
Community
Awesome Claude Dynamic Workflows
A community reading list and concept map covering workflow concepts, primitives, opt-in behavior, resume semantics, and ultracode.
peymanvahidi/awesome-claude-dynamic-workflowsOriginal
Tutorials and teardown
How to use and make dynamic workflows
This section turns official docs and community experiments into teachable operating principles: when to run a workflow, how to scope it, and how the runtime differs from ordinary subagent delegation.
Start01
Current Claude Code usage model
Use workflows inside an interactive Claude Code session. Run bundled workflows such as `/deep-research`, inspect runs with `/workflows`, and save successful scripts as slash commands.
- Start Claude Code in the project where the work should happen.
- Ask Claude to run a workflow for a scoped task, or run `/deep-research <question>` for the bundled research workflow.
- Open `/workflows` to inspect phases, agent counts, token totals, and completed results.
- Save a useful run from `/workflows`; later invoke the saved workflow as `/<name>`.
Build02
How the orchestration model works
A workflow is not just a longer prompt. The important shift is that branching, loops, intermediate results, and verification policy move into a JavaScript orchestration script.
- Use the script to hold state that would otherwise flood the main conversation.
- Use subagents for isolated perspectives, file groups, source groups, or reviewer roles.
- Use parallel phases only where tasks are independent.
- End with a convergence phase that filters findings and explains confidence.
Operate03
Cost and scope checklist
Dynamic workflows are expensive enough that the site should teach restraint: small edits, single-file changes, and unclear success criteria are usually poor fits.
- Prefer workflows for migrations, audits, research, and review tasks with many independent branches.
- Start with a narrow folder, package, or file sample before scaling up.
- Check the raw script or planned phases before approving a large run.
- Give agents a test, lint, benchmark, or source-validation bar whenever possible.
Explore Workflows
Browse and copy dynamic workflow scripts contributed by developers around the world.
Showing 4 workflows
Claude Code /deep-research Built-in Workflow
MediumA teardown of Claude Code's own deep research workflow: scope the question, fan out web search agents, fetch and extract claims, run adversarial verification, then synthesize a cited report.
OfficialResearchWebSearchVerification
5122400
View detailsParallel Test Refactoring and Visualizer
MediumFans out subagents to migrate legacy Jest tests to Vitest in parallel, verifying syntax correctness and running tests at each phase.
RefactoringTestingVitest
128890
View detailsAdversarial Codebase Security Audit
HardDeploys one group of agents to scan for vulnerabilities (XSS, SQL Injection, Dependency issues) and another group of adversarial agents trying to exploit the findings to confirm exploits.
SecurityAuditDevOps
2541240
View detailsLCP & Image Performance Optimizer
EasyAutomated Core Web Vitals audit. Analyzes LCP components, optimizes image formats, adds fetchpriority, and validates layouts.
PerformanceLCPImages
89450
View details