Why did the "Routines" section disappear from the left sidebar? (Claude Code)?

Your Routines Section Didn't Vanish — It Moved (Probably)

If you opened Claude Code recently and found the Routines section missing from your left sidebar, you're not imagining things — but you may be misreading what happened. Here's the reassuring part: no official changelog documents a removal. The curated release notes for recent versions like 2.1.300 and 2.1.301 contain no mention of Routines being pulled or of any sidebar changes at all, according to the Claude Code release notes feed.

The most likely explanation is simpler than a disappearance: Routines moved. Current official documentation places Routines inside the Desktop app's Code tab, accessible "in the sidebar or in the sidebar's More menu" — meaning the feature may now be tucked behind an overflow menu rather than sitting as a top-level sidebar item. If you're hunting for a dedicated Routines entry and not finding one, the "More" menu is the first place to look.

There's a second reason relocations like this should surprise no one: Routines is officially a research preview. The documentation states plainly that "behavior, limits, and the API surface may change" (per Anthropic's docs). The feature only launched on April 14, 2026, alongside a major desktop redesign that added an integrated terminal, a faster diff viewer, an in-app file editor, and multi-session support. When a feature is weeks old and its host app has just been rebuilt, UI shuffles are expected behavior, not bugs.

A few concrete data points underline how actively the feature is changing: prompt handling changed at version 2.1.213, and CLI trigger creation requires v2.1.225 or later. In other words, the ground under Routines has been shifting since day one.

So before assuming the worst, run through this quick check:

• Click the "More" overflow menu in the Desktop app's Code tab sidebar — Routines may simply live there now.
• Update Claude Desktop to at least version 1.1.5368, since local scheduled tasks aren't available on earlier builds.
• Confirm your plan qualifies — Routines is limited to Pro, Max, Team, and Enterprise.
• Check whether an org admin disabled Routines, which stops existing routines and blocks new ones for everyone.

If the sidebar still refuses to cooperate, remember that the Desktop app is only one of several doors into the feature. The web dashboard at claude.ai/code/routines and the CLI's /schedule command both reach the same routines — and the web interface is arguably the most stable surface while the desktop UI settles down.

One practical note as you re-establish your setup: scheduled desktop tasks are stored as SKILL.md instruction files, and community guides suggest skills as a lighter-weight alternative to token-heavy routines. If you're rebuilding automations around either approach, SkillzBench's free directory of vetted agent skills — with security scanning and benchmarking in development — is worth keeping on your radar before you wire unfamiliar instructions into a scheduled agent.

The 4-Point Diagnostic Checklist: Why Routines Isn't Showing for You

Before you assume the feature is gone, work through four documented causes. The good news: most "disappearances" turn out to be one of these, and each takes under a minute to rule out.

Check this first: the More menu. Official documentation states that in the Desktop app's Code tab, you click Routines "in the sidebar or in the sidebar's More menu." Because Routines launched as a research preview on April 14, 2026 — with docs explicitly warning that "behavior, limits, and the API surface may change" — UI relocation is expected behavior, not a bug. Overflow menus are exactly where features land during that kind of churn.

If the More menu comes up empty, move down the list:

• Update Claude Desktop. On versions before 1.1.5368, local scheduled tasks simply aren't available — an outdated install is one of the most common culprits.
• Confirm your plan. Routines is limited to Pro, Max, Team, and Enterprise subscribers. If you're on the free tier, the section won't appear at all.
• Ask your admin. On Team and Enterprise plans, Owners can disable Routines for everyone via the admin toggle — and when disabled, existing routines stop running and members can't create new ones. This is the only cause that actively kills routines you've already built.
• Check your CLI version. The /schedule command (alias /routines) requires Claude Code v2.1.225 or later, so an older CLI blocks that access path too.

One subtlety worth flagging: Desktop "Local" tasks run on your machine, while cloud routines run remotely — and conflating the two is a documented source of confusion. If you're looking for a cloud routine in the Desktop sidebar, it may simply be in the wrong place.

The most stable access path, meanwhile, is the web dashboard at claude.ai/code/routines, which sidesteps sidebar shuffling entirely. And if your routines are stored as SKILL.md instruction packages — which is exactly how Desktop scheduled tasks are saved — it's worth knowing which skills are trustworthy: roughly 26% of published agent skills carry at least one vulnerability. SkillzBench is building a free directory that security-scans, benchmarks, and version-pins skills so you can equip your agents with confidence.

Three Reliable Ways to Reach Routines Right Now

Whatever happened to your sidebar, you don't have to wait for Anthropic to put Routines back. Three documented access paths work today, and one of them is far more stable than the others.

The web dashboard at claude.ai/code/routines is your best bet. According to the official Routines documentation, this is where routines are created and managed, and because Routines is still a research preview — "behavior, limits, and the API surface may change" — the web surface is the least likely to shift under you. If the sidebar entry vanished on you once already, anchor your workflow here.

The CLI route requires a recent version: the /schedule command (alias /routines) needs Claude Code v2.1.225 or later, per the same docs. The Desktop app is the third option — Anthropic's scheduled-tasks guide says to click Routines in the Code tab's sidebar or in its "More" menu, which is likely where your missing entry went. Just make sure you're on version 1.1.5368 or newer, since local scheduled tasks don't exist on older builds.

Whichever path you choose, the same limits apply:

• Daily routine caps by plan: 5 for Pro, 15 for Max, 25 for Team and Enterprise, with extra billing beyond the cap, per SiliconANGLE's launch coverage
• Cloud routines enforce a one-hour minimum schedule interval — more frequent cron expressions are rejected outright
• Routines run only on Pro, Max, Team, and Enterprise plans, so check eligibility before troubleshooting anything else
• Admins can disable Routines org-wide at claude.ai/admin-settings/claude-code, which stops existing routines cold

One more thing worth noticing: this isn't random UI churn. Claude Code has been increasingly nudging users toward cloud execution for autonomous work, as practitioner analysis notes — Anthropic's Boris Cherny has recommended running Claude Code in the cloud for tasks spanning hours or days, so you can close your laptop. Expect the cloud routine surface to keep getting the polish while local surfaces shuffle around.

If token-heavy routines feel like overkill for your task, the mrprompts training guide suggests Claude Skills as a lighter alternative — and desktop scheduled tasks are themselves stored as SKILL.md files. SkillzBench is building exactly this kind of infrastructure: a free directory and MCP connector for vetted, security-scanned agent skills, so you can compare and equip skills with confidence rather than guesswork.

Alternatives When Routines Won't Cooperate

If Routines won't load or your admin has switched it off, you're not stuck. Anthropic's own documentation points to three scheduling alternatives that work without the sidebar entry at all.

The first is Claude Code GitHub Actions, which the official docs recommend for running Claude on a schedule in CI. This suits repository-centric jobs — nightly test runs, scheduled dependency checks — because the trigger lives in your workflow file rather than Anthropic's UI, making it immune to sidebar reshuffles and research-preview churn.

The second is /loop, Claude Code's in-session polling command. It re-prompts the agent on an interval while your session stays open, with a minimum interval of one minute. That makes it the right tool for watching a build finish or polling a log, but wrong for anything that needs to run while your machine is asleep.

The third path is the Desktop app's local scheduled tasks. These run on your own machine, support intervals down to one minute, and — tellingly — are stored as SKILL.md files in ~/.claude/scheduled-tasks/<task-name>/SKILL.md. If your Desktop is older than version 1.1.5368, these tasks aren't available, so update first. Desktop also checks the last seven days of missed runs and executes one catch-up run for the most recent missed time, which softens the blow of an offline laptop.

Before you rebuild anything, keep two caveats in mind:

• Security: everything a routine does through your connectors appears as you — commits carry your GitHub username and Slack messages post from your account, so a misbehaving schedule acts with your identity.
• Cost: routines operate in the domain of tokens, which is more expensive than pure compute — fine for a daily digest, wasteful for high-volume webhooks.
• Practitioner guides suggest Claude Skills as a lighter alternative for high-frequency jobs, reserving token-based routines for low-frequency, high-judgment work.

That SKILL.md detail matters more than it looks. Because scheduled tasks are just skill files, the same instruction packages that power one-off automation can drive your schedules — and since research found roughly 26% of published agent skills carry at least one vulnerability, it pays to vet what you install. SkillzBench is building a free directory that security-scans, benchmarks, and version-pins skills so you can pick schedules and automations with confidence.

Turn Scheduled Tasks Into Repeatable Agent Skills

Here's the detail most people miss when they go hunting for the vanished Routines panel: those scheduled tasks were never really a separate feature at all. According to Anthropic's documentation, each desktop scheduled task lives on your machine as a SKILL.md file inside ~/.claude/scheduled-tasks/<task-name>/. In other words, the thing you thought disappeared is actually an agent skill wearing a different interface.

That reframing matters for anyone building repeatable AI workflows. The complete training guide on Routines makes the point bluntly: routines "operate in the domain of tokens, which is more expensive than pure compute," so they're poorly suited to high-volume, repetitive jobs. The guide's suggested alternative? Claude Skills — a lighter, cheaper way to package the same instructions and hand them to an agent on demand, rather than paying for a scheduled run that burns tokens whether or not the task needed doing that hour.

The practical takeaway for business owners and agent builders is to think in terms of a skills library rather than a pile of one-off scheduled prompts:

• Convert recurring routine prompts into reusable SKILL.md packages you can invoke whenever needed, not just on a timer.
• Keep genuinely time-sensitive automations on schedules (cloud routines allow a minimum one-hour interval; desktop tasks and /loop go down to one minute), and handle everything else with skills.
• Write skill instructions the way the experts recommend for routines: "the way you'd write instructions for a smart new hire on their first day."
• Vet every skill before you install it — quality and safety vary wildly across what's published.

That last point deserves emphasis. Skills are spreading across Claude Code, Codex, Cursor, Gemini CLI and chat-app agents, but published research found that roughly 26% of published skills carry at least one vulnerability. Grabbing a random SKILL.md off GitHub and dropping it into your agent is a security decision, not just a convenience. This is the gap SkillzBench is being built to close: a free, independent directory that continuously crawls and dedupes skills from GitHub, pins every version to a commit SHA, runs security scans, license checks and quality judgments, and benchmarks each skill in a sandbox — with and without the skill — so you can see the Lift in percentage points before you trust it. A companion free MCP connector will let you toggle exactly which skills each of your agents can see. All of it is in development now, with the waitlist open ahead of launch.

The bottom line: don't mourn the sidebar. Whether Routines moved to the More menu, got toggled off by an admin, or simply needs a Desktop update past 1.1.5368, the durable version of the workflow is a well-written, well-vetted skill you own. Skills, proven.