19 Ways to Use Claude So Well It Feels Illegal (Tutorial)
This tutorial covers 19 advanced Claude Code commands and techniques designed to improve productivity and manage token consumption. The presenter walks through features ranging from multitasking during long-running tasks (/btw) to scheduling automation (/schedule, /loop) and remote session control (/remote-control). Many of the commands discussed are described as new or little-known features within Claude Code's terminal environment.
Summary
The video tutorial presents 19 Claude Code commands organized into thematic clusters. The first command, /btw, allows users to ask Claude quick questions while it is already running a long task in the same session, eliminating the need to open a new tab or session.
Commands 2 through 7 focus on token consumption management. /model lets users switch between Claude's model tiers (Haiku, Sonnet, Opus) depending on task complexity. /effort adjusts reasoning depth across four levels (low, medium, high, max). /advisor is described as a brand-new feature that automatically escalates difficult problems to a more powerful model only when needed, allowing users to keep a cheaper default while accessing Opus-level reasoning selectively. The keyword 'ultrathink' can be appended to any prompt to trigger maximum reasoning effort on a per-prompt basis. Two open-source plugins — 'caveman' and 'context mode' — are recommended for reducing output and MCP tool call tokens respectively. /context visualizes token usage as a color-coded grid with per-category breakdowns and optimization suggestions. /clear removes previous conversation context to reduce token overhead and prevent quality degradation from irrelevant history.
Commands 9 and 10 address skill development. /insights analyzes 30 days of session history to produce a report card identifying strengths, weaknesses, and underused features tailored to the user's patterns. Skill Creator is a plugin-based meta-skill that builds, tests, and benchmarks other skills using structured evaluations (evals), variance benchmarking, and an improve mode that iteratively refines skills based on benchmark results.
/team onboarding (command 11) generates a project onboarding guide from session history that a new team member can paste as their first message to get oriented on tools, commands, and workflows.
Commands 12 and 13 cover scheduling. /loop creates a session-scoped repeating task with a minimum 1-minute interval that expires after 3 days. /schedule creates a cloud-hosted cron trigger independent of the local machine, with a minimum 1-hour interval and daily run caps varying by plan tier (5 for Pro, 15 for Max, ~25 for Team/Enterprise).
Commands 14 through 16 address session continuity. /remote-control (or /rc) syncs Claude Code running on a local computer with the Claude mobile app, allowing remote prompting and permission approval while away from the desk. /rename lets users label sessions for easier identification during remote control. /resume allows continuing a previous session across different devices.
Command 17 is the Claude Chrome Extension, which gives Claude browser access for visual interaction and automation on sites without APIs. The presenter notes that Playwright MCP is a more powerful alternative for technical users. A key use case described is creating a feedback loop where Claude builds a web app, launches it in the browser, tests it visually, and iterates until the feature works correctly.
/rewind (command 18) lets users select a past message and branch the conversation from that point, effectively creating a new timeline. /status line (command 19) customizes the information bar below the Claude Code input field to display project folder, git branch, context window usage percentage, session cost, and elapsed time.
Key Insights
- The presenter describes /advisor as a brand-new feature that automatically escalates tasks to the most advanced model only when Claude determines it needs stronger judgment, allowing users to maintain a cheaper default model (e.g., Sonnet at medium effort) while accessing Opus-level reasoning selectively without manually switching models.
- The presenter argues that /schedule runs cloud-hosted cron jobs independently of the user's local machine, but notes it has meaningful limitations: a minimum 1-hour interval, no local file system access (it clones the project remotely), and a daily run cap of 5 for Pro, 15 for Max, and ~25 for Team/Enterprise plans.
- The Skill Creator plugin is described as a meta-skill that not only generates skill files but also creates structured evaluations (evals) that grade skill quality over time, runs benchmarks measuring output variance across multiple executions, and has an improve mode that analyzes failures and refines the skill automatically.
- The presenter claims that /insights analyzes 30 days of Claude Code session history to produce a detailed report card with charts identifying where users are performing well, what is holding them back, and which Claude Code features they are not using but should be based on their specific usage patterns.
- The presenter describes using the Claude Chrome Extension to create an autonomous feedback loop for web app development, where Claude codes a feature, launches the app in the browser, visually tests each user flow step, and iterates until the feature works as intended — without requiring manual testing by the developer.
Topics
Full transcript available for MurmurCast members
Sign Up to Access