I Tested Every Claude Code Feature, These 12 Are the Best
A content creator with 500+ hours in Claude's ecosystem ranks Claude Code features from D to S tier based on personal productivity impact. The video highlights 12 top features, with Skills ranked #1 for enabling consistent, reusable agent workflows. The ranking prioritizes knowledge work and automation use cases over traditional software development.
Summary
The creator opens by explaining that rankings are based on personal impact to day-to-day knowledge work and automation workflows — not global popularity or technical impressiveness — and acknowledges that software engineers may rank features differently.
D-tier features include foundational but unglamorous tools like CLAUDE.md setup, context lifecycle commands (/clear, /compact), web search, file uploads, fast mode, permissions, and custom themes. C-tier includes voice mode, Claude Co-work (which the creator skipped in favor of Claude Code directly), interactive connectors, and local file access — all useful but infrequently used personally.
B-tier highlights dynamic workflows (parallel processing but token-intensive and still new), deep research (a workflow-based slash command), git worktrees (for isolated codebase branches), ultra review (code review in the cloud), interactive charts, and the session recap feature.
The A-tier honorable mentions include the Google Workspace CLI (for Sheets, Docs, Drive, Gmail, Calendar access), Dispatch for mobile conversation continuity, the Claude Desktop App for managing routines and sessions visually, Claude Design for team design systems, auto mode, hooks, effort levels, agent view, and prompt caching.
The top 12 features are ranked in ascending order: #12 /goal (persistent task execution until objective criteria are met), #11 ultra plan (offloads planning to cloud-based planning agents), #10 /insights (generates a 30-day usage report with workflow recommendations), #9 auto memory (automatic memory improvement between sessions), #8 agent teams (multiple personas that debate and analyze topics in rounds), #7 /rewind (rolls back code and conversation to earlier checkpoints), #6 sub agents (parallel background agents that work independently and can be defined as markdown files), #5 /loop (sets up recurring cron-based reminders or tasks tied to a session), #4 remote control (drive local Claude Code from a phone or web browser), #3 routines (schedule actual agent sessions, not just scripts, via the desktop app), #2 the status line (displays model, effort level, and context window usage in the terminal), and #1 Skills (reusable markdown-based recipe files that give the agent consistent, repeatable instructions for complex or simple tasks).
The creator emphasizes that skills are the single most impactful feature because they enable consistency, chaining, team sharing, and can range from a simple 4-sentence prompt shortcut to multi-step workflows with tool routing.
Key Insights
- The creator argues that Skills — markdown-based recipe files stored in .claude — are the single most impactful Claude Code feature because they enable consistent, repeatable agent behavior without re-prompting, and can be chained, shared with teams, or kept as simple as a 4-sentence shortcut.
- The creator distinguishes agent teams from sub agents by explaining that agent teams can communicate and debate with each other across rounds, while sub agents work in parallel but only report back to the main terminal session that invoked them.
- The creator identifies the terminal status line as one of the most underrated features, stating it was the primary reason they switched from the VS Code extension to the terminal — because seeing model, effort level, and exact context window usage (e.g., 274,000 of 1,000,000 tokens) is critical for knowing when to run session handoffs or compact.
- The creator uses /goal not for multi-day marathon runs (as demos often show) but for bounded tasks with objective completion criteria — citing a real example where it ran for 90 minutes optimizing a globe animation on a website until it loaded instantly on refresh.
- The creator explains that routines in Claude Code schedule an actual agent session — not a deterministic Python script or simple API call — meaning the scheduled task has the full reasoning and tool-use capabilities of a live Claude Code terminal.
Topics
Full transcript available for MurmurCast members
Sign Up to Access