Slash command support has quietly become the most consequential change to Claude Code on iOS since the mobile experience launched. For roughly six months, anyone driving a local Claude Code session from an iPhone hit the same wall: type /compact into the app, and the literal text /compact was handed to the model as an ordinary chat message. The slash command never ran. The session kept filling up. You walked back to your desk.
That wall is gone. Boris Cherny, who works on Claude Code at Anthropic, confirmed on 17 August 2026 that slash commands typed into the mobile app and claude.ai now execute against the local CLI process, and pointed at release 2.1.202 as the version carrying the fix. The documentation has since been rewritten to spell out exactly which ones travel and which ones do not.
This article covers what actually shipped, the complete list of what works from a phone today, the handful that still demand a terminal, the 173-day trail of community reports that pushed it there, and the one gap that is still open. If you track releases like this, they land in our AI models and tools hub as they happen. The short version for anyone deciding whether to rely on it: the capability is real, the coverage is broad, and the missing piece is not the slash commands themselves but the ability to discover them on a small screen.
Table of contents
- What Actually Changed in the Claude Code iOS Slash Command Story
- Every Slash Command That Works From Claude Code on iOS
- The Slash Commands That Still Need a Local Terminal
- The Missing Autocomplete Is the Real Friction
- How the Request Became One of the Repository’s Loudest
- Custom Slash Commands and Skills Are the Unfinished Half
- What Claude Code on iOS Actually Is
- How to Get Slash Commands Working From Your iPhone Today
- What This Means for Development Teams
- What We Could Not Verify
- Claude Code Slash Command FAQ
- References
What Actually Changed in the Claude Code iOS Slash Command Story
The fix is narrower than “mobile got slash commands” and more useful than it sounds.
The confirmation and the version behind it
On 17 August 2026, Cherny posted on the long-running feature request that had tracked this since February. His wording was specific: from the mobile app or claude.ai you can run /compact, /clear, /context, /usage, /recap, /model <name>, /effort <level>, /fast, /config key=value, /mcp, /autocompact <size> “and more; they run in the local CLI process.” Terminal-only commands such as /rewind, /resume, and /plugin still need the local terminal.
The changelog line is the real receipt
Release 2.1.202 lists it plainly: “Fixed commands sent from Remote Control (mobile/web) into an interactive session failing with Unknown command.” That phrasing matters. The slash command plumbing was not absent — the routing existed and the interactive session rejected what arrived. Two other Remote Control fixes shipped in the same release, covering images sent without a caption being silently dropped and the wrong permission mode showing in the mobile and web apps.
“Runs in the local CLI process” is the important clause
A slash command sent from an iPhone is not interpreted by the phone. It crosses the Remote Control connection and executes inside the claude process on your machine, against that session’s real context window, real MCP servers and real settings. A /compact from the sofa compacts the same conversation your laptop is holding. That is why the fix was a routing fix rather than a mobile feature.
It is still being extended
The pattern has continued past 2.1.202. Release 2.1.260, the current version, added a text form of /advisor explicitly “for the desktop app, Remote Control, and other headless sessions” — the same design move, applied to another slash command. Each release converts one more terminal-shaped interaction into something that survives the trip to a phone.
| Date | Event in the slash command story |
|---|---|
| 24 Feb 2026 | Remote Control launches, requiring Claude Code v2.1.52 |
| 25 Feb 2026 | Issue #28379 filed: slash commands treated as plain text in the remote UI |
| 4 Mar 2026 | Issue #30674 asks for /compact from mobile, marked “Critical – Blocking my work” |
| 8 Mar 2026 | Issue #32051 reports no autocomplete for a project with 145+ custom commands |
| 4 May 2026 | A user reports slash commands executing on v2.1.126, but with no picker |
| 13 Jun 2026 | Issue #68252 filed: slash commands intermittently route as plain text |
| 17 Aug 2026 | Anthropic confirms support, naming release 2.1.202 |
Every Slash Command That Works From Claude Code on iOS
The Remote Control documentation now carries an explicit inventory rather than a vague caveat. It splits into three shapes.
Text-output slash commands, which behave normally
These need no picker, no dialog and no keyboard shortcut, so they cross the connection unchanged: /compact, /clear, /context, /usage, /exit, /usage-credits, /recap and /reload-plugins. One footnote: /usage-credits prints the billing URL rather than opening a browser, because the phone is not where the browser lives.
Slash commands that take an argument instead of a picker
/model, /effort, /fast, /color and /rename all work, but you pass the value inline — /model sonnet, /effort high — because the terminal versions open a picker or a slider that has nowhere to render. This is the single most useful adaptation in the set. Changing the model or the effort level mid-task was previously the clearest reason to walk back to the desk.
The three that behave differently by surface
/mcp has worked since v2.1.166 and returns a text summary of server status on mobile rather than opening the picker; its reconnect, enable and disable subcommands work from both mobile and web, and reconnect with no server name reconnects every server that has failed. /config has worked since v2.1.181 and takes key=value on mobile, or lists the settable keys when run bare. /autocompact arrived in v2.1.221 and takes the window size as an argument, as in /autocompact 500k.
| Slash command | Form on iOS | Minimum version |
|---|---|---|
/compact, /clear, /context | Bare command, text output | 2.1.202 |
/usage, /recap, /exit | Bare command, text output | 2.1.202 |
/usage-credits | Prints billing URL, no browser | 2.1.211 for the confirmation rule |
/model, /effort | Argument replaces the picker | 2.1.202 |
/fast, /color, /rename | Argument form | 2.1.202 |
/mcp | Text status summary; subcommands work | 2.1.166 |
/config | key=value, or bare to list keys | 2.1.181 |
/autocompact | Takes a size, e.g. 500k | 2.1.221 |
/advisor | Text form for Remote Control | 2.1.260 |
/rewind, /resume, /plugin | Local terminal only | — |
Counting the coverage
Nineteen slash commands are named across the supported and unsupported lists. Sixteen of them run from a phone and three do not, which puts documented mobile coverage at roughly 84% of the slash commands Anthropic explicitly addresses.
The Slash Commands That Still Need a Local Terminal
Three are named as local-only, and the reasoning is consistent once you see it.
/rewind, /resume and /plugin
Anthropic’s documentation says commands that only run in the terminal interface “work only from the local CLI, whether or not you pass an argument.” A user reported in July 2026 that /rewind over Remote Control returns a message saying it “isn’t available over Remote Control” — a clean refusal rather than a silent failure, which is the correct behaviour but still a wall.
Why a picker cannot travel
Each local-only slash command is built around an interactive terminal surface. /resume renders a scrollable conversation picker. /plugin runs a marketplace browser. /rewind needs checkpoint selection against files on disk. None of them reduce to a single line of text plus an argument, which is precisely the property that let /model and /effort make the trip.
The one that stings
/rewind is the painful absence, and the July report explains why better than a changelog could: rewinding a bad turn is one of the main reasons you would reach into a session mid-flight, and being away from your desk is exactly when you need that recovery. A session you can steer from a phone but cannot roll back from a phone is a session you still have to babysit.
| Command | Terminal surface it needs | Workaround from iOS |
|---|---|---|
/rewind | Checkpoint picker over local files | None; ask Claude to undo in prose |
/resume | Scrollable conversation list | Pick the session from the app’s Code list |
/plugin | Marketplace browser | /reload-plugins works remotely |
The Missing Autocomplete Is the Real Friction
This is the part the announcement does not solve, and it is worth being blunt about.
Typing a slash command from memory on a phone keyboard
A report filed in March 2026 laid out the problem precisely: on desktop, typing / opens a dropdown of every available slash command, built-in and custom. On mobile, nothing appears. You type the full name from memory. The person filing it noted their project carried more than 145 custom slash commands and described mobile use as impractical without a cheat sheet open in another app.
An earlier report caught the same gap in February
The original autocomplete request, filed the day after Remote Control launched, paired it with a second complaint: /clear appeared to work on the CLI side while the connected app’s chat window kept showing the old conversation. Another commenter narrowed it further, reporting that /compact worked correctly over Remote Control from an Android device on the Claude Max plan while /clear was silently ignored.
The contrast that makes it obvious
The Remote Control documentation notes that typing @ in the mobile composer autocompletes file paths from your local project. So one autocomplete surface already crosses the connection and renders on a phone. The / picker does not. That asymmetry is what makes the gap read as unfinished work rather than a deliberate design line.
A user found this out before the docs said so
In May 2026, months before the official confirmation, a user running v2.1.126 reported that slash commands were in fact being processed as commands. Their description of the experience is the whole story in two sentences: no UI help, so you need to know the slash commands, but Claude is definitely processing them. Support arrived before discoverability did, and before anyone announced it.
How the Request Became One of the Repository's Loudest
The community record is unusually legible here, because every step of it is public.
Eighty-nine reactions on a single issue
The canonical request, opened on 25 February 2026, accumulated 56 thumbs-up, 31 hearts and 2 eyes reactions — 89 in total — alongside 13 comments. For a routing limitation in a developer tool, that is a lot of people arriving at the same wall independently.
Four duplicates filed within seven weeks
Three separate reports were closed as duplicates of it or of each other between 25 February and 15 April 2026, each describing the same behaviour from a different angle: one about /compact in long-running sessions, one about autocomplete, one about project skills not registering.
The June regression muddied the picture
Between the first partial support and the August confirmation, a June 2026 report described slash commands intermittently routing to the model as plain text on CLI version 2.1.173 — with the app sometimes showing a success toast for a command that never ran. /context executed once in a session and then stopped. That report is still open as of 4 September 2026, and a July comment confirmed the same behaviour on Android, so it is not an iOS-specific fault.
Custom Slash Commands and Skills Are the Unfinished Half
Built-in commands travel. Your own do not, reliably, and this is the gap most likely to affect a team.
Project skills that the mobile harness cannot see
An April 2026 report, labelled by Anthropic with both platform:ios and area:skills, described project-level skills under .claude/skills/*/SKILL.md being readable on disk and visible to the model, yet not registered as invocable slash commands on mobile. Asking the agent which commands were available returned only about a dozen built-ins.
The failure has two separate faces
The user-facing half is that operators must describe a skill in natural language instead of invoking it. The deeper half is that the agent’s own Skill tool returns Unknown skill for the same entries, so any chained workflow that runs one skill after another breaks on mobile even when no human types a slash command at all.
The workaround has a real cost
The filed workaround is to instruct the agent to read the SKILL.md file directly and follow it inline. That reproduces the behaviour, but it lands the full skill file in the conversation context on every invocation instead of a short tool call and result — a token-cost asymmetry that grows with each step in a chain. For teams building autonomous AI agents on top of Claude Code, that is a design constraint rather than an inconvenience.
It was reported again on Android in May
The report was auto-closed for inactivity on 24 May 2026 and a second user asked the same day for it to be reopened, confirming the symptom on Android and describing how it reads from the user’s side: skills simply appear broken on mobile.
What Claude Code on iOS Actually Is
Some of the confusion around this feature comes from what people assume Remote Control does.
Your machine still does all the work
Remote Control connects claude.ai or the Claude app to a Claude Code session running on your own computer. Claude keeps running locally the entire time, so code execution and filesystem access stay on your machine. Your MCP servers, tools and project configuration remain available. The phone is a control surface, not a runtime.
The February launch and what has changed since
Remote Control launched on 24 February 2026, requiring Claude Code v2.1.52, started as a research preview limited to Max subscribers, and was pitched by Claude Code product manager Noah Zweben as a way to “take a walk, see the sun, walk your dog without losing your flow.” VentureBeat’s coverage put Claude Code at a $2.5 billion annualised run rate at the time. The plan restriction is gone: the documentation now says Remote Control is available on all plans.
The one setting an administrator controls
On Team and Enterprise plans it is off by default until an Owner enables the Remote Control toggle in the Claude Code admin settings. Letting a phone drive a machine that holds source code is a cybersecurity decision rather than a convenience one, so that switch belongs in the same conversation as the rest of your security and trust practices.
Sessions stop when the process stops
Remote Control runs as a local process. Close the terminal or quit VS Code and the session goes offline until you bring it back. To keep one alive on a remote machine after an SSH disconnect, the documentation recommends starting it inside tmux or screen.
How to Get Slash Commands Working From Your iPhone Today
Four practical steps, in the order that saves the most time.
Check your version first
The routing fix landed in 2.1.202 and the current release is 2.1.260. Individual commands have their own floors — 2.1.166 for /mcp, 2.1.181 for /config, 2.1.221 for /autocompact, 2.1.260 for /advisor — so a machine that has not updated in months will refuse commands the documentation lists as supported.
Start the session and connect the app
Run claude remote-control, claude --remote-control, or /remote-control — abbreviated /rc — inside an existing session. If you do not have the app yet, /mobile inside Claude Code displays a download QR code. Remote Control sessions appear under Code in the app’s navigation, marked with a computer icon and a green status dot when online.
Memorise six, not sixty
Without a picker, the practical set is small. /context to see where the window is, /compact with instructions when it is filling, /model <name> and /effort <level> when the task changes shape, /usage before a limit bites, and /clear between unrelated jobs. Everything else can wait for a keyboard.
Turn on notifications so you know when to reach for it
Accept the operating system notification prompt, then run /config in the terminal and enable “Push when Claude decides”, “Push when actions required”, or both. On iOS, Focus modes and notification summaries can suppress or delay these, so check Settings, Notifications, Claude if nothing arrives.
What This Means for Development Teams
The change is small in engineering terms and larger in operational ones.
Mobile approval is not the same as mobile administration
A phone can now change the model, adjust effort, compact context and read usage on a live session. It still cannot roll back a bad turn or install a plugin. That is a sensible split, but it should be a deliberate one in your working agreements rather than a surprise discovered mid-incident by someone standing on a train platform.
The autocomplete gap has a governance shape
A team that has invested in custom commands under .claude/commands — the 145-command project in the March report is not an outlier in mature repositories — currently gets none of that leverage on mobile, and none of the discoverability either. If your delivery process assumes those commands, mobile is a monitoring surface, not an execution one. This matters most where a slash command encodes a step in a DevOps pipeline.
Cost control genuinely improved
/usage, /context, /compact and /autocompact from a phone are the four that change spending behaviour. Manual compaction with custom preservation instructions produces better summaries than waiting for auto-compact, and being able to trigger it from anywhere means fewer sessions degrading quietly while nobody is at the desk.
Write down which commands your team relies on
Because there is no picker on mobile, the cheat sheet is now a real artefact. A short list in your repository README of the slash command set your team actually uses, with the argument forms spelled out, costs ten minutes and removes the single biggest friction in the workflow.
What We Could Not Verify
Three things remain genuinely open at the time of writing.
Whether autocomplete is planned
No Anthropic statement addresses a / picker for the mobile composer. The August confirmation covers execution only. Treat the missing autocomplete as current behaviour with no announced timeline.
Whether custom commands and skills now register
The skills report was closed for inactivity rather than fixed, and no changelog entry since describes registering project skills in the mobile harness. We could not confirm the current behaviour on a device.
Whether the intermittent routing fault persists
The June report describing commands occasionally arriving as plain text is still open. It predates 2.1.202, so it may well have been resolved by the same fix without the issue being closed, but nothing states that.
Claude Code Slash Command FAQ
Can I run /compact from the Claude iOS app?
Yes. /compact is one of the text-output commands that runs from the mobile app or claude.ai on Claude Code 2.1.202 and later, executing in the local CLI process against your real session context.
Why does typing / not show a list on my phone?
Because the mobile composer has no slash command picker. Autocomplete for / exists on desktop only; the @ file-path autocomplete does work on mobile, which is why the omission is noticeable.
How do I change the model from my phone?
Pass the value as an argument — /model sonnet rather than opening a picker. The same applies to /effort <level>. A pick made from the device applies to the running session on your machine.
Which commands will never work remotely?
/rewind, /resume and /plugin are documented as local-only because each depends on an interactive terminal surface. /rewind returns an explicit message saying it is not available over Remote Control.
Do my custom project commands work from iOS?
Not reliably. Reports through April and May 2026 describe project skills under .claude/skills/ failing to register as invocable commands in the mobile harness, with the agent’s own Skill tool returning Unknown skill.
Does a slash command run on my phone or my computer?
On your computer. Remote Control forwards it to the claude process on your machine, where it executes against that session’s context, MCP servers and settings.
References
Continue local sessions from any device with Remote Control
Slash commands not supported in /remote-control UI
Remote Control: slash command autocomplete missing and /clear doesn’t clear app chat window
Support slash commands (/compact) in Remote Control mobile sessions
Slash commands don’t autocomplete in Claude Code mobile app
Mobile harness does not register project skills as slash commands
Remote Control: slash commands route to model as plain text instead of executing
Anthropic just released a mobile version of Claude Code called Remote Control
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.