Skip to main content
    Kriv AI

    Claude Code

    What is a Claude Code workflow? The two kinds Anthropic documents

    Anthropic uses the word workflow for everyday recipes and for a script that runs many subagents. This page puts both in order and says what we could not read.

    Anthropic's Claude Code documentation uses "workflow" in two ways. Its Common workflows page describes its recipes as "Step-by-step guides for exploring codebases, fixing bugs, refactoring, testing, and other everyday tasks with Claude Code." Its workflows page says: "A dynamic workflow is a JavaScript script that orchestrates many subagents at once." This page walks through the everyday steps in order, then the dynamic kind. Source: Anthropic Claude Code documentation, read 18 November 2026.

    plain answer

    A Claude Code workflow starts by understanding the code

    Anthropic's first recipe is for a codebase you do not know yet. It tells you to start Claude Code in the project root and ask for "a high-level overview", with the example prompt "give me an overview of this codebase". Its tips say: "Start with broad questions, then narrow down to specific areas."

    For large searches, Anthropic points to subagents: "Exploring a large codebase fills your context with file reads. Delegate the exploration so only the findings come back." It adds that the subagent "reads files in its own context window and reports a summary."

    plain answer

    Plan mode lets you review a plan before any edit

    Anthropic's permission modes page says: "Plan mode tells Claude to research and propose changes without making them." It says Claude "reads files, runs shell commands to explore, and writes a plan, but does not edit your source." Edits stay blocked until you approve the plan, except in interactive terminal sessions with bypass permissions available.

    You enter plan mode by pressing Shift+Tab, by prefixing one prompt with /plan, or by starting with the command claude --permission-mode plan.

    When you do not yet know what you want, the best practices page suggests asking Claude to interview you: "Interview me until you could write the spec, then write it to SPEC.md." It then advises starting "a fresh session to build from it, so the build begins with a clean context and a written definition of done."

    plain answer

    Give Claude a check that comes back pass or fail

    Anthropic's advice is to give Claude something it can run that comes back pass or fail; our guide to Claude Code best practices covers it.

    plain answer

    Fix, refactor and test in small steps, then review the pull request

    The Common workflows page writes each task as short steps. To fix a bug it says to share the error, ask for fix recommendations, then apply the fix, and its tip is to "Tell Claude the command to reproduce the issue and get a stack trace." For a refactor it ends with "run tests for the refactored code" and advises: "Do refactoring in small, testable increments." For tests, its last step is "run the new tests and fix any failures".

    For pull requests, Anthropic says you can ask Claude directly ("create a pr for my changes") and then: "Review Claude's generated PR before submitting and ask Claude to highlight potential risks or considerations." Our own pages on Claude Code review and the security review command cover that last step.

    plain answer

    Resume, run in parallel and run from scripts when the work is long

    Anthropic says that when a task spans several sittings you can "pick up where you left off instead of re-explaining context", and that claude --continue "resumes the most recent session in the current directory."

    For parallel work it says each git worktree is "a separate checkout on its own branch", started with claude --worktree followed by a name, so edits do not collide.

    For scripts and CI, the documentation says to pass -p to run Claude Code in non-interactive mode, for example claude -p "Find and fix the bug in auth.py" with the tools you allow. It also says that in a Manual mode claude -p run, "a call that needs approval is denied unless something approves it."

    plain answer

    Run Claude on a schedule when nobody is at the keyboard

    The Common workflows page has a section called "Run Claude on a schedule". It lists four places a scheduled task can run. Cloud scheduled tasks run in the cloud, "Anthropic-managed by default", for "Tasks that should run even when your computer is off." Desktop scheduled tasks run on your machine for "Tasks that need direct access to local files, tools, or uncommitted changes." GitHub Actions run in your CI pipeline for "Tasks tied to repo events like opened PRs". The /loop command runs in "The current CLI session" for "Quick polling while a session is open."

    Its tip for the prompt: "be explicit about what success looks like and what to do with results. The task runs autonomously, so it can't ask clarifying questions."

    plain answer

    A dynamic workflow is a script Claude writes that runs many subagents

    Anthropic's workflows page says: "A dynamic workflow is a JavaScript script that orchestrates many subagents at once. Claude writes the script for the task you describe, and a runtime executes it in the background while your session stays responsive."

    It says to reach for one "when a task needs more agents than one conversation can coordinate, or when you want the orchestration codified as a script you can read and rerun." Its examples include a codebase-wide bug sweep and a 500-file migration. The page's table of who decides what runs next gives "Claude, turn by turn" for subagents and "The script" for workflows.

    To start one, you ask in your prompt: include the keyword ultracode, or say "use a workflow". One of Anthropic's example prompts reads: "use a workflow to run npx tsc --noEmit and keep fixing the reported errors until the type check passes or two rounds in a row make no progress". It says Claude Code asks whether to allow the workflow before it runs, and that you can watch it with /workflows.

    The page is direct about cost: "a single run can use meaningfully more tokens than working through the same task in conversation." It lists a limit of 1,000 agents total per run. It says dynamic workflows are "available on all paid plans, with Anthropic API access, and on Amazon Bedrock, Google Cloud's Agent Platform, and Microsoft Foundry", and that on Pro you turn them on from the Dynamic workflows row in /config.

    plain answer

    Skill, subagent or hook: which one for which job

    Anthropic's Extend Claude Code page sorts the pieces you add around a workflow. A skill is "Reusable instructions, knowledge, or workflows" and a subagent is an "Isolated worker with its own context". The page says: "Use a subagent when you need context isolation or when your context window is getting full."

    For a step that must happen every time, it points to a hook: "Use a hook when the action must happen the same way every time and doesn't need Claude to think." Its comparison table says a hook "Fires on its event without Claude deciding", while for a skill "Claude interprets the instructions; outcome can vary". Our Claude Code hooks page covers hooks in detail.

    When a skill is worth writing is covered in our guide to Claude Code skills.

    limits

    What we could not confirm

    We did not run these steps for this page and make no claim about how long any of them takes or what results they give. We read the Common workflows, Run Claude Code programmatically, workflows and Extend Claude Code pages in full on 18 November 2026. We read the Best practices page for the sections quoted here, and the first 100,000 of 120,338 characters of the permission modes page, which includes the plan mode section.

    None of the pages shows a date. Anthropic updates them often and some flags need a minimum Claude Code version, so re-read the pages before you copy a command.

    Straight answers

    Frequently asked questions about What is a Claude Code workflow? The two kinds Anthropic documents

    What is a Claude Code workflow?

    Anthropic uses the word in two ways. Its Common workflows page describes its recipes as "Step-by-step guides for exploring codebases, fixing bugs, refactoring, testing, and other everyday tasks with Claude Code." Its workflows page defines a dynamic workflow as "a JavaScript script that orchestrates many subagents at once." Source: S1 and S5, read 18 November 2026.

    How do I make Claude Code plan before it edits?

    Anthropic says plan mode makes Claude "research and propose changes without making them". You press Shift+Tab, prefix one prompt with /plan, or start with claude --permission-mode plan. Source: S3, read 18 November 2026.

    When should I use a dynamic workflow instead of subagents?

    Anthropic says to reach for a workflow "when a task needs more agents than one conversation can coordinate, or when you want the orchestration codified as a script you can read and rerun." Source: S5, read 18 November 2026.

    Can Claude Code run on a schedule?

    Anthropic lists four options on its Common workflows page: cloud scheduled tasks, desktop scheduled tasks, GitHub Actions and /loop, which differ in where the task runs. Source: S1, read 18 November 2026.

    Can I run two Claude Code sessions on one repository?

    Anthropic says to give each its own checkout: claude --worktree followed by a name starts a session in a separate git worktree on its own branch. Source: S1, read 18 November 2026.

    Can Claude Code run in a script or in CI?

    Anthropic says to add the -p flag to run it non-interactively, and recommends --bare for scripted and SDK calls so the result is the same on every machine. Source: S4, read 18 November 2026.

    Talk to the team that would do the work

    Bring your requirements to a working session with the person who'll actually deliver.

    Book a Discovery Call