← Back to library

9 Hacks to Use Claude Code Better Than 90% of People (In 9 Minutes)

Watch videoView transcript

Chase AI8m 41sTranscript ✅Added May 6, 3:52 pm GMT+8

Actionable Insights

  1. Prefer a CLI over MCP when the task is terminal-native and non-interactive If a mature CLI exists (gh, rg, jq, docker, kubectl), teach Claude Code the command shape in a project skill/doc instead of adding an MCP server. Evaluation: fewer tool round trips, lower context overhead, easier local reproduction. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: The video is a rapid list of Claude Code workflow optimizations: choose CLIs over unnecessary MCPs, use side conversations/commands/hooks, manage context aggressively, and encode repeatable tool knowledge as skills or docs. But something odd I found is codex seems to pick up on skills more than Claude code. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

  2. Use MCP when resources need structured access, auth mediation, or long-lived server state The MCP spec is valuable for databases, SaaS APIs, browser/session tools, and shared resources. Caution: MCP adds permission and trust boundaries; review tool schemas and outputs. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: - Supporting evidence: The transcript provides direct evidence for what the creator demonstrated or recommended; source links in Actionable Insights identify the projects/docs/tools that should be inspected before adoption. The video is a rapid list of Claude Code workflow optimizations: choose CLIs over unnecessary MCPs, use side conversations/commands/hooks, manage context aggressively, and encode repeatable tool knowledge as skills or docs. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

  3. Create a “CLI skill card” for each tool Include install command, auth prerequisite, 5 safe read-only commands, 3 write commands requiring approval, output-size limits, and rollback. Example: rg for search, jq for JSON, gh pr diff --name-only for PR review. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: The video is a rapid list of Claude Code workflow optimizations: choose CLIs over unnecessary MCPs, use side conversations/commands/hooks, manage context aggressively, and encode repeatable tool knowledge as skills or docs. The missing evidence to look for is reproducible install steps, current official docs, security model, pricing/limits, recent maintenance, and before/after metrics on real tasks. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

  4. Add completion notifications with hooks Use Claude Code hooks for “task finished / needs input” signals. Start with OS notification or sound only; do not add auto-commit/deploy hooks until reviewed. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: The video is a rapid list of Claude Code workflow optimizations: choose CLIs over unnecessary MCPs, use side conversations/commands/hooks, manage context aggressively, and encode repeatable tool knowledge as skills or docs. - Top audience signal: @btwsaid (8 likes) said: “Claude Code usage got worse a few days back. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

  5. Keep long explorations out of one giant session Use /clear/fresh sessions/subagents where appropriate, and summarize durable findings into CLAUDE.md or project docs. Evaluation: fewer “lost in context” failures and lower token burn. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: The video is a rapid list of Claude Code workflow optimizations: choose CLIs over unnecessary MCPs, use side conversations/commands/hooks, manage context aggressively, and encode repeatable tool knowledge as skills or docs. - Supporting evidence: The transcript provides direct evidence for what the creator demonstrated or recommended; source links in Actionable Insights identify the projects/docs/tools that should be inspected before adoption. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

Core thesis

The video is a rapid list of Claude Code workflow optimizations: choose CLIs over unnecessary MCPs, use side conversations/commands/hooks, manage context aggressively, and encode repeatable tool knowledge as skills or docs.

Creator’s main claims and verdicts

1. “Stop using MCPs for everything; prefer CLIs”

Verdict: Mostly agree, medium-high confidence. For terminal-native tools, CLIs are inspectable, scriptable, and usually lower overhead. This matches practical agent harness design: if the model can run rg, git, jq, or docker directly, an MCP wrapper may be unnecessary.

Underclaimed: MCP is still the right abstraction for authenticated APIs, shared resources, structured prompts/resources/tools, and controlled server-side capabilities. The MCP spec exists because raw shell is not a universal safe interface.

2. Hooks/notifications save real operator time

Verdict: Agree, high confidence. Official hooks support event-driven behavior. The comment suggesting Windows toast notifications is a useful practitioner improvement over simple sounds.

3. Clearing/splitting context improves efficiency

Verdict: Agree, high confidence. Token-counting docs confirm repeated context costs; long sessions also accumulate irrelevant instructions. The nuance: clearing too aggressively can lose important state, so write a summary first.

4. Skills/docs make CLIs usable by agents

Verdict: Strong agree, high confidence. A CLI alone is just an affordance; a short skill/doc with safe patterns, examples, and cautions makes it reliable.

Best timestamped moments

  • 0:00 — opens with the CLI-over-MCP recommendation. Evidence: “All of my favorite cloud code techniques have two things in common. One, they’re extremely high leverage, and two, they’re dead simple to implement. And today, I’m sharing nine of them in 9 minutes with you. So, let’s get started. Now, tip number one is about …”
  • 0:30 — explains terminal-native CLIs as lower-overhead tools. Evidence: “an MCP to a CLI, we want to. Now, why is that? It’s because they’re more efficient. They use less tokens. And oftentimes, we also get more functionality because, think about it, command line interfaces live in the terminal. Where does Cloud Code live? In the t…”
  • 1:01 — introduces /btw sidebar conversation claim. Evidence: “two things always go in tandem. So the second tip is related to a new feature that only came out a few weeks ago, and that is the buy the way feature or /btw. Now, what this does, it allows me to hold a sidebar conversation with Cloud Code inside of the same s…”
  • 1:34 — positions /btw and hooks as efficiency hacks. Evidence: “you know, it was just a normal combo. The good thing about this is not that I can just talk to it while it’s doing something else. It’s the fact that this conversation does not add to the context window. So it’s almost like free. So BTW becomes a great efficie…”
  • 2:05 — recommends a completion sound hook. Evidence: “and you’ll forget about it for 20, 30 minutes, an hour, right? However, if you have it played just like a small subtle ding or something, I’m telling you, you are much more likely to come back and see what happened, right? And this ends up saving you like hour…”
  • 2:05 — moves into /clear and context-window management. Evidence: “and you’ll forget about it for 20, 30 minutes, an hour, right? However, if you have it played just like a small subtle ding or something, I’m telling you, you are much more likely to come back and see what happened, right? And this ends up saving you like hour…”

Comment-derived insights

  • 8 likes @btwsaid: Claude Code usage got worse a few days back. Do you plan to make videos on Codex?
  • 5 likes @sssubarusss: I use both Claude code and codex. From my own experience I find codex to be a better and more robust coder. Claude is a little better at design. But something odd I found is codex seems to pick up on skills more than Claude code.
  • 3 likes @rasm0225: could you do an explainer on the difference between CLI and Desktop App for Claude Code?
  • 3 likes @truyenhinhsuckhoe3077: Claude code is better. But Codex offers more tokens
  • 1 likes @WillyDrucker: Have an improved hook to share tip. You have it play a sound when it’s finished, I used to do this as well, found a better version. Install the powershell toast cmdlet, Claude will guide you. And similarly create a hook so that you get Windows toast notificati
  • 1 likes @TiriRiri_PL: King Claude
  • 0 likes @Chase-H-AI: Get the Claude Code Masterclass 👉 https://www.skool.com/chase-ai
  • 0 likes @ed_nico: Tip 5 - love it. Thank you very much
  • 0 likes @NickRomanek: the context management tips alone are worth it. most people treat it like fancy autocomplete and wonder why results drift
  • 0 likes @deborahlynna.astrology: Thank you!

Commenters add useful field notes: some compare Claude Code with Codex, one reports Codex picking up skills well, and one suggests Windows toast notifications via PowerShell instead of a sound-only hook. That makes the best immediate improvement to the video’s hook tip: prefer native OS notifications that appear in a tray and can be disabled centrally.

Screen-level insights

  • 0:00 — youtube-extract/XkSBO-CZDFs/frames/000_000000.jpg: visible key frame extracted for this workflow step; nearby transcript says “All of my favorite cloud code techniques have two things in common. One, they’re extremely high leverage, and two, they’re dead simple to implement. And today, I’m sharing nine of …”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 0:30 — youtube-extract/XkSBO-CZDFs/frames/001_000030.jpg: visible key frame extracted for this workflow step; nearby transcript says “an MCP to a CLI, we want to. Now, why is that? It’s because they’re more efficient. They use less tokens. And oftentimes, we also get more functionality because, think about it, co…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 1:01 — youtube-extract/XkSBO-CZDFs/frames/002_000061.jpg: visible key frame extracted for this workflow step; nearby transcript says “two things always go in tandem. So the second tip is related to a new feature that only came out a few weeks ago, and that is the buy the way feature or /btw. Now, what this does, …”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 1:34 — youtube-extract/XkSBO-CZDFs/frames/003_000094.jpg: visible key frame extracted for this workflow step; nearby transcript says “you know, it was just a normal combo. The good thing about this is not that I can just talk to it while it’s doing something else. It’s the fact that this conversation does not add…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 3:07 — youtube-extract/XkSBO-CZDFs/frames/006_000187.jpg: visible key frame extracted for this workflow step; nearby transcript says “tokens, that drops to 78.3 on Opus and 65.1. So, with this in mind, that is telling us that anytime we get to 20% of the context window, 25% we want to clear because if you don’t h…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 4:09 — youtube-extract/XkSBO-CZDFs/frames/008_000249.jpg: visible key frame extracted for this workflow step; nearby transcript says “despite its name, it does a lot more than create skills. It also modifies skills, improves them, and measures skills performance. And so this means if we ever want to change a skil…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 4:39 — youtube-extract/XkSBO-CZDFs/frames/009_000279.jpg: visible key frame extracted for this workflow step; nearby transcript says “experimental feature that you need to enable to have access to, but it allows us to go well beyond simple sub aents. That’s because normal sub aents operate in a vacuum, right? I c…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 5:09 — youtube-extract/XkSBO-CZDFs/frames/010_000309.jpg: visible key frame extracted for this workflow step; nearby transcript says “just something you have to balance out. And I found that the outputs are totally worth it. And like I said at the start, you have to enable agent teams in the settings. Easiest way…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 6:40 — youtube-extract/XkSBO-CZDFs/frames/013_000400.jpg: visible key frame extracted for this workflow step; nearby transcript says “download Obsidian. You create a vault, aka just some sort of folder on your computer that is designated the vault, and then you open up Cloud Code inside it. From there, you can ha…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.
  • 7:10 — youtube-extract/XkSBO-CZDFs/frames/014_000430.jpg: visible key frame extracted for this workflow step; nearby transcript says “up using the terminal, it’s extremely trivial. All you need to do is navigate to whatever folder you have designated the vault. In my case, it’s literally called the vault. And the…”. Use this as screen evidence, not just narration: verify the UI/tool named at that point and whether the demo actually shows execution or only a slide/product page.

The visual pass matters because this is a tips video: the screen can show whether a tip is a real command, a settings UI, a terminal workflow, or just presenter narration. For /btw and slash-command claims, frame evidence should be paired with official Claude Code docs or local version tests.

CLI-vs-MCP decision table

  • Use CLI when: local repo/files, mature command exists, output is text, human can rerun command, auth already handled locally.
  • Use MCP when: server-side auth/resources, structured tool schema, shared remote state, non-shell UX, or safety boundary is needed.
  • Use neither when: deterministic library/API call inside your app is simpler.

Sources / evidence checked

Verification notes

Four verification roles were applied before publishing: source/evidence audit, transcript/comment/frame fidelity audit, hallucination/overclaim audit, and Actionable Insights audit. Corrections made: softened “stop using MCPs” into a decision rule; added direct MCP/Claude hooks docs; converted tips into executable evaluation criteria; added comment-derived Windows notification improvement. Residual uncertainty: /btw and some slash commands may be version-specific, so verify locally before documenting as team standard.

  • Actionable Insights audit: expanded to the newer detailed format with fuller implementation notes, evaluation checks, and cautions where the existing evidence supports elaboration.

Comment insights

  • Top audience signal: @btwsaid (8 likes) said: “Claude Code usage got worse a few days back. Do you plan to make videos on Codex?”. This is the highest-salience community reaction and should be weighted as audience evidence, not proof.
  • practitioner addition: @sssubarusss (5 likes) — I use both Claude code and codex. From my own experience I find codex to be a better and more robust coder. Claude is a little better at design. But something odd I found is codex seems to pick up on skills more than Claude code.
  • practitioner addition: @rasm0225 (3 likes) — could you do an explainer on the difference between CLI and Desktop App for Claude Code?
  • practitioner addition: @truyenhinhsuckhoe3077 (3 likes) — Claude code is better. But Codex offers more tokens
  • pushback / caveat: @WillyDrucker (1 likes) — Have an improved hook to share tip. You have it play a sound when it’s finished, I used to do this as well, found a better version. Install the powershell toast cmdlet, Claude will guide you. And similarly create a hook so that you get Windows toast notifications when it finishes instead. They route
  • practitioner addition: @TiriRiri_PL (1 likes) — King Claude
  • Synthesis: Treat the comments as an adoption-risk check: if commenters ask for proof, cost controls, setup details, or safety boundaries, the workflow should include those checks before production use.

Deep research

  • Research scope: This pass cross-checks the creator’s claims in “9 Hacks to Use Claude Code Better Than 90% of People (In 9 Minutes)” against the extraction transcript, available linked/tool names in the analysis, and general public documentation/search evidence already cited elsewhere in this page where present.
  • Supporting evidence: The transcript provides direct evidence for what the creator demonstrated or recommended; source links in Actionable Insights identify the projects/docs/tools that should be inspected before adoption.
  • Contradicting/limiting evidence: Video demos and tool lists rarely prove production reliability. The missing evidence to look for is reproducible install steps, current official docs, security model, pricing/limits, recent maintenance, and before/after metrics on real tasks.
  • Verification method: Before using this in production, rerun the workflow on a small representative repo/task, save logs and outputs, compare against a non-agent baseline, and require human review for any external write/deploy/payment action.