Automation and Web Commands

Choose the Right Surface

PR Automation Pattern

/autofix-pr detects the open PR from your checked-out branch with gh pr view. To watch a different PR, check out its branch first. By default the remote session is told to fix every CI failure and review comment; pass a prompt to give it narrower instructions.

1. Ensure the branch has an open PR.
2. Run /web-setup if the web GitHub connection is not already configured.
3. Start with a bounded prompt:

/autofix-pr only fix lint, typecheck, and reviewer comments that do not change architecture

4. Review the resulting commits and CI status before merging.

/pr-comments was removed in v2.1.91. To view pull request comments inside a session, ask Claude directly to read them instead.

Routine Pattern

Use /schedule when a workflow should run without your laptop staying open. Routines execute on Anthropic-managed cloud infrastructure, and Claude walks you through the setup conversationally:

/schedule every weekday morning triage new issues and post a Slack summary

If GitHub is not yet connected, /schedule prompts you to run /web-setup first.

Use /loop instead when the loop depends on the current local shell, running server, or local filesystem state. /loop [interval] [prompt] runs a prompt repeatedly while the session stays open; omit the interval to let Claude self-pace between iterations (alias: /proactive).

Use /goal when the task is not scheduled but has a concrete stop condition:

/goal tests pass and the migration notes are updated

Clear an active goal early with /goal clear (also stop, off, reset, none, or cancel).

Subagent definitions and /agents

Since v2.1.198, /agents prints guidance to ask Claude to create or edit a subagent file. The Running/Library wizard belongs to v2.1.197 and earlier. Use .claude/agents/ for project definitions or ~/.claude/agents/ for personal definitions; claude agents is the separate background-session view.

Changes in existing project/user agents directories are detected automatically. A newly created agents directory, one loaded through an added directory, or a session launched with --disable-slash-commands still needs a restart. See Subagents.

Remote Work Pattern

Use Remote Control (/remote-control or /rc) when you want the work to keep running on your machine while interacting from a browser or phone. Use Claude Code on the web when the work should run in Anthropic-managed cloud infrastructure. Use /teleport (or /tp) when a web session needs to come back to your terminal.

Use Agent View for local background work you may attach to later. Open it with claude agents, or dispatch a session straight to the background from the shell:

claude --bg --name "flaky-test-fix" "investigate the failing SettingsChangeDetector test"
claude --bg --exec 'pnpm test -- --runInBand'
claude agents

--bg starts a session that goes straight to the background; --name sets its display name in Agent View; --exec runs a shell command as a background job in place of a Claude session, so no model is invoked. After backgrounding, Claude prints the session's short ID. Current top-level management commands are claude attach <id>, claude logs <id>, claude stop <id>, claude respawn <id>, and claude rm <id> (v2.1.251+).

Before editing files, a background session moves into an isolated git worktree under .claude/worktrees/ so parallel sessions can read the same checkout while each writes to its own.

Use /workflows when Claude has generated a dynamic workflow that coordinates many background agents. It opens the workflow progress view to watch, pause, resume, or save runs. It is a control surface, not a replacement for merge review.

Related Commands and Naming Notes

A few commands relevant to automation and review have been renamed or aliased recently:

• /code-review [low|medium|high|xhigh|max|ultra] reviews the current diff for correctness bugs, with cleanup coverage depending on model and effort; pass --fix to apply findings or --comment to post them as inline GitHub PR comments. /code-review ultra runs a deep cloud-based multi-agent review; this is the preferred invocation, and /ultrareview remains as an alias.
• /code-review and /verify are user-invoked only since v2.1.215. v2.1.218 introduced background review execution. In v2.1.274+, models without tuned review settings use a leaner inline prompt, so the execution strategy is model-dependent.
• /fork copies the current conversation into a separate background session; /subtask delegates inherited-context work inside the current conversation (v2.1.212+).
• /usage shows session cost, plan usage limits, and activity stats. /cost and /stats are aliases (/stats opens on the Stats tab).
• /usage-credits configures usage credits to keep working when you hit a limit; it was previously named /extra-usage.
• /config opens the Settings interface; /settings is an alias. The legacy /vim command was removed in v2.1.92 — toggle Vim editing mode through /config → Editor mode.
• /init initializes a project CLAUDE.md guide. Set CLAUDE_CODE_NEW_INIT=1 for an interactive flow that also walks through skills, hooks, and personal memory files; when a CLAUDE.md already exists, /init suggests improvements rather than overwriting it.
• /fewer-permission-prompts is the bundled skill that scans your transcripts for common read-only Bash and MCP calls and adds a prioritized allowlist to .claude/settings.json to reduce permission prompts.