Build agentic systems that run your business: https://skool.com/scrapes
Don't miss the next build - / @simonscrapes
Grab the full resources for this video here (free): https://scrapeshq.notion.site/10-tips...

If you’re using the Claude Code like most people, you're missing out on its true power. Most tutorials show isolated tricks, but in this video, we walk through 10 advanced techniques that compound on each other, from simple tweaks to complex workflows that even experienced users overlook. By the end, you'll be able to use Claude Code in ways you probably haven't even imagined yet.

00:00 - Intro: The Problem with Most Claude Workflows
00:40 - When NOT to use Agent Teams
04:45 - Hooks, Skills & Slash Commands Simplified
07:10 - Point, Don’t Dump
08:05 - Plugins - Why?
09:07 - Restore History
10:07 - Getting past paywalls
11:50 - 30 days of research in 30 seconds
16:25 - Highly used built-in commands
17:13 - No more manual tasks
18:44 - Hallucinations 101

#claudecodetips #claudecode #claudecodetutorial

Create a detailed outline of this video, broken down by sections and subjects that are given in the video description.

Here is a detailed outline of the video, using the sections and subjects from the description and chapter list.​

Intro – The Problem with Most Claude Workflows (00:00)

• Explains that most people use Claude Code as a “fancy autocomplete” and rely on isolated tricks instead of compound workflows.​

• States goal of the video: 10 techniques that build on each other, from small tweaks to complex systems (agent teams, plugins, skills, research loops).​

• Positions the tips as relevant both for new and daily users who want to move beyond simple one‑off prompts.​

Tip 1 – When NOT to Use Agent Teams (00:40)

• Compares old sub‑agent architecture vs new agent teams.

  • Sub‑agents: one main parent agent delegating to specialists (researcher, reviewer, writer) who only report back to the parent.​

  • Agent teams: multiple agents share a task list and can talk to each other as well as the main orchestrator, reducing the bottleneck.​

• Explains downsides: more conversations between agents mean higher token usage and costs, so agent teams are not always justified.​

• Gives a “don’t use” example: content workflow for 5–10 social posts with three roles (researcher, writer, reviewer).​

  • Scope is small and tasks are mostly linear; researcher and reviewer only need to talk to the main writer agent, not each other.​

  • Agent teams here are “overkill” and add cost without real benefit.​

• Gives “good use” examples:

  • Multi‑asset content project from a single YouTube transcript: blog article, LinkedIn carousel, newsletter version.​

    • Teammates must coordinate hooks and angles so they don’t contradict or duplicate each other.​

  • Complex app build (e.g., Next.js app with Stripe) where API agent, front‑end agent, and testing agent all need to collaborate.​

    • API and front‑end agents frequently talk to the testing agent to spin up tests; shared task list prevents the main agent becoming a delegation bottleneck.​

• Practical guidance:

  • Start with a single agent, then sub‑agents, and only move to agent teams when cross‑collaboration is truly required.​

  • Notes you can navigate directly to a teammate and assign tasks without going through the main agent using keyboard shortcuts.​

Tip 2 – Hooks, Skills & Slash Commands Simplified (04:45)

• Introduces the three core customization concepts and how they work together in a content example.​

1. Slash commands

   • Stored in .claude/commands as single markdown files.​

   • Mental model: press a button to do a specific action on demand.​

   • Example: /lin_posts (or /linkedin-post) that always writes three LinkedIn posts about a given topic using $arguments inside the command file.​

2. Skills

   • Stored in .claude/skills as folders containing skill.md plus supporting files.​

   • Invoked automatically when Claude recognizes that a task matches the skill description.​

   • Example: a brand‑voice skill with:

     • Platform‑specific examples (LinkedIn, X, Instagram).

     • Banned words list.

     • A skill.md that instructs Claude to apply these guidelines whenever writing social content.​

   • The description alone is enough for Claude to choose the skill, though you can reinforce usage via claude.md.​

3. Hooks

   • Defined in settings.json, do not use LLM tokens, and run programmatically as post‑processing.​

   • Example: banned‑word checker that watches an output/drafts folder and scans new drafts for forbidden phrases, reporting whether any are present.​

• Shows how all three combine in a pipeline:

  • User runs /lin_posts to create drafts.​

  • Claude automatically loads brand‑voice and humanizer skills.​

  • Hooks then scan outputs for banned words in the background.​

Tip 3 – Point, Don’t Dump (07:10)

• Identifies common mistake: dumping entire brand guides, content pillars, audience research, and example posts directly into claude.md.​

• Explains why this is bad:

  • claude.md is loaded into context every conversation; massive files waste context and hurt clarity.​

• Recommended pattern:

  • Keep claude.md lean (roughly 20–30 lines) and point to detailed docs instead.​

  • Example: instruct Claude to look in /skills/humanizer or brand‑voice files when writing content, instead of embedding them.​

• Result: a clean, scannable claude.md that preserves context window and still gives Claude access to rich detail as needed.​

Tip 4 – Plugins – Why? (08:05)

• Clarifies that plugins are not a separate “capability type” but a distribution mechanism that bundles skills, hooks, commands, and agents into one installable package.​

• Shows the Claude plugin marketplace and basic install flow: run two commands to clone or configure a plugin, gaining a full workflow at once.​

• Example: Reddit research plugin for content creation:

  • Includes a skill teaching Claude how to analyze Reddit threads.

  • A command that triggers deep Reddit research.

  • A hook that summarises findings into a markdown file.​

• Benefit: install one plugin instead of configuring multiple skills/commands/hooks manually.​

• Caveat: most current plugins are dev‑focused, but the pattern is powerful and the ecosystem is evolving.​

Tip 5 – Restore History (09:07)

• Shows how to search and summarize past Claude Code projects via the global projects file.​

• Example workflow:

  • Ask Claude: “What have we spoken about in the last week in all my projects? Give me a summary.”​

  • Claude reads the global projects log and returns:

    • List of projects with activity.

    • Number of sessions per project.

    • Summary of what each project covered and outputs produced.​

• You can then ask follow‑ups like “Tell me more about the agent teams we created in the Claude Code demo folder” and Claude restores the specific history.​

• Purpose: quickly regain lost context and resume older work without manually grepping logs or reopening old sessions by hand.​

Tip 6 – Getting Past Paywalls (10:07)

• Problem: Claude Code cannot access some sites (Reddit, certain news sites, paywalled content) directly from the terminal.​

• Shows typical failure pattern: Claude reports it cannot fetch from Reddit and falls back to generic web search.​

• Presents workaround inspired by YK’s “45 Claude Code tips”:

  • Install Gemini’s command‑line interface as a fallback tool.​

  • Create a “Reddit fetch” skill that:

    • Starts a teamwork session.

    • Uses Gemini CLI to fetch and summarize a given URL or query.

    • Captures output and returns results to Claude.​

• Demonstrates using the skill on an AI subreddit thread to retrieve top posts with points, comments, and content summary.​

• Notes limitations: sometimes still only returns generalized info, but forms basis for deeper research workflows.​

Tip 7 – 30 Days of Research in 30 Seconds (11:50)

• Introduces “last 30 days” research skill that scans Reddit, X, and the web for discussions over the last month on a topic.​

• Setup requirements: add OpenAI and xAI API keys; install from the referenced GitHub repo.​

• Usage via command:

  • /last-30-days [topic or topic for tool] (e.g., “prompting techniques for chat for legal questions,” “iOS app mockups for Figma AI,” “LinkedIn hooks that are working right now”).​

• Shows internal logic:

  • Breaks down “topic,” “target tool,” and “query type” (prompting, general, or trending topics).​

  • Research typically takes 2–8 minutes.​

• Example 1: SaaS landing page mockups for Nano Banana.

  • Finds that JSON‑structured prompts dominate, with nested properties like visual style and constraints.​

  • Reports real metrics (upvotes, posts, pages) and highlights top voices.​

  • Returns ready‑to‑use JSON prompts that generate consistent Nano Banana images.​

• Example 2: latest Claude Code news.

  • Reports recent launches, adoption stats (e.g., share of GitHub commits), viral posts, and new features like Claude Co‑work.​

• Example 3: AI automation YouTube content that’s working now.

  • Extracts patterns (e.g., AI + history as a breakout niche, hyper‑real visuals, long‑form storytelling).​

• Takeaway: you get summarized, source‑backed trend research instead of generic search results.​

Tip 8 – Highly Used Built‑In Commands (16:25)

• Highlights built‑in slash commands many users overlook but use frequently.​

1. /clear

   • Compresses and clears conversation history to free context without fully resetting the project.​

2. /usage

   • Shows current session usage vs plan limits (e.g., percentage of Opus 4.6 allocation used in this session and this week).​

3. /stats

   • Reports usage over time: number of sessions, longest session duration, and fun comparisons like “equivalent to the tokens in Moby Dick.”​

• Emphasizes using these commands to avoid context rot and to monitor costs.​

Tip 9 – No More Manual Tasks (Clipboard & Formatting) (17:13)

• Focuses on moving output into other apps (especially LinkedIn) without breaking formatting.​

• Problem: copy‑pasting from chat often breaks bullets, emojis, and bold formatting.​

• Solution pattern:

  • Ask Claude to write in “LinkedIn‑native format” with no markdown headers and emoji bullets.​

  • Use the built‑in functionality to copy directly to clipboard instead of manually selecting and copying.​

• Demonstrates pasting the clipboard content into LinkedIn with preserved bullets and arrows.​

• Additional tip:

  • Paste into a blank Notion page first to preserve markdown, then copy from Notion to other platforms where appropriate.​

  • Notes this works better for many platforms than direct paste, though LinkedIn often still benefits from the native‑format approach.​

Tip 10 – Hallucinations 101 (18:44)

• Addresses Claude confidently inventing stats and facts.​

• Simple safeguard pattern:

  • After generating content with factual claims (e.g., posts with specific numbers), instruct Claude:

    • “Double‑check every claim and statistic, and make a table of what you could and couldn’t verify, including the source.”​

• Demonstration:

  • Claude produces a markdown fact‑check table with each claim, its verification status, and sources.​

  • Example: “MCP launched November 2024” marked as verified; “downloads 100K to 8 million in five months” flagged as unverified because it only appears in a single blog without methodology.​

  • Claude recommends actions like “remove claims 5, 6, 7” or “soften claims 2 and 3 or swap for official numbers.”​

• Takeaway: build a quick verification pass into your workflow to filter out hallucinated numbers before publishing.​

Closing – Putting the 10 Tips Together (around 20:00)

• Recaps that the 10 techniques compound: from structuring claude.md and commands, to plugins, research skills, and fact checking.​

• Emphasizes these patterns as foundational for the creator’s own workflows and encourages viewers to adopt them end‑to‑end.​

• Points viewers to a deeper dive Claude Code video and links to the free resources/Notion page for this episode.​