How to Use Claude Routines for Repeatable Tasks

Claude does not have a built-in task scheduler that runs prompts automatically on a timer. A "routine" with Claude means a structured, repeatable session: the same context, the same prompt, and a review step each time. Tools like fenbs keep the reusable procedure, the current task context, and the recorded outcomes in one place so each run of a routine starts from the agreed setup rather than a blank page.

The value of a routine is consistency. The same brief, the same constraints, and the same review checklist produce output you can compare from week to week. Without that structure, repeating a task with Claude tends to produce different results each time, making it hard to spot improvements or problems.

What "routines" means for Claude users

Claude.ai offers Projects, which store a custom system prompt and a set of documents that persist across all conversations in that project. That persistent context is the foundation of a Claude routine: the system prompt holds the standing instructions, the documents hold any reference material, and each conversation is one run of the routine.

For users on plans without Projects, or when working with Claude via the API, a routine is simply a saved prompt file you paste at the start of each session. The mechanics differ; the principle is the same: the same setup, every time.

Setting up a routine in Claude Projects

Create a Project for the routine. In the Project instructions (the system prompt), write the standing brief: what the task is, who the output is for, any constraints on format or language, and what a good result looks like. Add any reference documents the routine depends on — a style guide, a data schema, a checklist.

Each time you run the routine, open a new conversation in the Project. The standing instructions and documents are already there. Paste or describe the new input, run the task, and review the output against your checklist before using it.

Writing a saved prompt for a routine

A good routine prompt includes: the goal in one sentence, the format of the input you will supply, the format of the output you expect, any standing constraints, and an example of a good result. Save this as a text file or in a notes app. When you run the routine, paste the prompt, then add the input below it.

## Weekly summary routine
Goal: Summarize the project updates from the past week into a three-paragraph status report for a non-technical manager.
Input: I will paste the raw updates below.
Format: Three short paragraphs — completed work, in-progress items, blockers. Plain language. No bullet points.
Constraints: No jargon; US English; no prices.

--- Updates start here ---

Review the prompt after each run. If Claude consistently misunderstands a constraint or produces a format you correct every time, update the prompt now rather than correcting the output next time.

Scheduling the trigger

Because Claude does not run prompts on a timer, something else must trigger the routine. That can be a calendar reminder, a task in your project tool, or a script that calls the Claude API and stores the result. For API-based scheduling, the Claude API accepts the same prompt structure and returns the output programmatically. See claude-api for how the API works.

Keep the trigger simple enough that you will actually use it. A calendar event that says "run weekly summary" is more reliable than a complex script that breaks when an input format changes.

Reviewing routine output

Every routine needs a review step. Claude's output can drift from the intended result over time as context changes or as you modify the prompt. Read each output against the standing brief before using it or sharing it. For a two-minute task, even a thirty-second review catches the errors that matter.

When the output is wrong, note whether the error is a prompt problem (fix the standing instructions), an input problem (the data you supplied was incomplete), or a new Claude behavior. Update the prompt for the first two; for the third, test the change before committing to it.

Recording routine outcomes

Save a record of each run: the date, the input summary, and whether the output met the standard. When a routine starts producing different results, that record tells you when it changed and what the last good run looked like. Without a record, debugging a degraded routine means guessing.

How to do this with fenbs

1. Create a free account and add a project for this routine.
2. Add the routine's prompt to the project as a skill, with the standing instructions and format requirements.
3. Record decisions — such as approved changes to the prompt format or output criteria — as Decisions and rules entries.
4. Log each run as a task completion note: what input was used, whether the output met the standard, and any prompt change made afterward.
5. Review the skill and decision log before making changes to ensure you know the current approved version.

What Claude routines cannot automate

Claude does not send emails, post to a calendar, update a database, or take any action outside the conversation without a tool connection or API integration. A Claude routine produces output you review and act on. If you need actions taken automatically, that requires either the Claude API with additional code, or a tool that connects Claude to other services. See what is Claude AI for what Claude can and cannot do by default.