Pratique du Shadowing: How to actually force Claude Code to use the right CLI (don't use CLAUDE.md) - Apprendre l'anglais à l'oral avec la vidéo
Création de la leçon...
1
One of the most common questions I get about agents
2
and about claw code is how do I force it to use the right CLI command?
3
How do I force it to adopt my workflow instead of the workflow that it knows best?
4
For instance, how do I get it to use PNPM instead of NPM?
5
How do I get it to call this wrapper script that we wrote instead of calling the thing directly with NPX?
6
How do I prevent it from using certain commands and only allow it to use other commands?
7
If you're interested in these kind of questions, this is the stuff that I answer on my newsletter.
8
I'm working on a Claude code course right now for real engineers, not Vibecoders.
9
So if you dig that stuff, the link is down below.
10
The first thing that you probably think about
11
when you think about steering Claude in one direction but not another is to put information in a Claude.md file.
12
In other words, to proactively tell it to use PNPM, not NPM.
13
Now, this will be pretty effective most of the time, especially if like this, you have a Claude.md file that is nearly empty apart from that one instruction.
14
However, in various places, including in the previous video I just posted, I've advocated almost always deleting your claw.md.
15
Because if you think about it, this instruction here is only relevant for certain tasks, and putting it in claw.md makes it global for every action that you're going to do in the repo.
16
LLMs have an instruction budget
17
that you can really only put about 500 instructions in before it starts getting less
18
and less powerful and more and more confused.
19
And we want our LLM thinking about the complicated plans and the implementation that it's going to put together, not having irrelevant instructions about using NPM or PNPM.
20
And the final thing that makes me uncomfortable about this is the fact
21
that it's not a deterministic way of preventing a command.
22
For instance, we may want to disallow it from running destructive commands.
23
For instance, we never want it to git push.
24
These kind of negative instructions to the LLM are really bad,
25
I think, because they burn instruction budget in something that we're not actually using the instruction budget to prevent deterministically.
26
By adding this, we're reducing our chances of running git push or reducing our chances of running NPM, but we're not preventing it.
27
So instead, I'm going to show you an approach that prevents it deterministically and guides the LLM towards the correct solution.
28
The way to do this is via Claude Code hooks.
29
Hooks allow you to run deterministic code at certain points during Claude Code's execution cycle.
30
When one of these events fire, all matching hooks run in parallel.
31
So for instance, you can run a hook when the session starts.
32
You can run a hook when a permission dialogue appears to automatically approve things.
33
The one that we're interested in is pre-tool use before a tool call executes can block it.
34
Our plan here is to use this pretool use hook to before it calls npm to use pnpm instead.
35
And the cool thing about this is that Claude code already knows how to do this.
36
And I've actually prepared a prompt inside this file to take the Claude.md, turn it into deterministic hooks, and then put those hooks into your project.
37
Here's what it looks like.
38
Take the instructions in your Claude.md file, turn them into deterministic Claude code hooks in this project directory.
39
Not Once all the instructions will be deterministic, only do the ones you can, such as instructions to use one CLI command over another or disallowing certain CLI commands.
40
Use separate bash scripts for running the hooks, in other words this is its own .sh file and it gives some syntax here for how to construct that.
41
This is directly lifted from the Claude code docs.
42
And then finally below here, first confirm with the user which hooks will be created,
43
second implement the hooks, third provide the user with instructions on how to test the newly created hooks.
44
I'll include this in an article which is linked below for you.
45
So I'm going to copy and paste this, I'm going to go into a fresh Claude Code session here, I'm going to zoom up the terminal a bit, I'm going to paste it in and see what happens.
46
Alright, it's given me two options here.
47
It says there are two rules that can be enforced deterministically, block NPM commands or block git push.
48
Now I've actually it already got some git push blockers in my global setup so I'm just going to say, yeah, do the npm one only.
49
So that sounds good.
50
Let's run this and see what happens.
51
So it's now given us something to review.
52
It's given us a nice version of the bash command that we saw earlier, except it just checks if it's npm.
53
Crucially it's checking if the command starts with npm as a command.
54
It echoes use pnpm not npm and then it exits with an exit code of 2.
55
So that is looking good to me.
56
Let's accept that.
57
It's then going to make it executable by running chmod plus x, so I'm happy with that.
58
It's then created the configuration inside settings.json here, where we've got now hooks, which is a property on the settings, pretool use, which is the one we saw earlier.
59
It's matching any bash tools here.
60
And then when it matches any bash tool, it runs this command, block npm.sh.
61
So that looks good to me.
62
We're in a good spot.
63
And it's now appearing to verify that the hook works in line.
64
I'm just going to try and run this to see what happens.
65
And you can see here just up there it said blocked use PNPM not NPM.
66
So we're definitely getting somewhere.
67
And finally it's given us some instructions here to test it.
68
To restart claw code in this project try asking it to run something like NPM install foo.
69
Alright the first thing I'm going to do is actually remove the instruction from claw.md here.
70
And then we're going to pull open a fresh clawed code instance
71
and we're going to say npm install let's say React query.
72
Okay, let's give it something kind of juicy.
73
And we can see that the pre-tooled use here,
74
it said bash hook error blocked use pnpm not npm and now it's running pnpm install tanstat react query instead.
75
So that's it.
76
We have prevented one command and we've automatically steered it to use another command.
77
And crucially we've managed to kill an instruction inside our global clawed.md file.
78
This is what I love so much about this approach.
79
You get to take instructions out of your instruction budget and enforce them deterministically.
80
This means that the instructions that we have in clawed.md, they are available just in time.
81
They get to hover in the background until clawed actually progressively kind of discovers them.
82
And this idea of these rules just being enforced just in time, you can take this and apply it to lots of different things.
83
For instance, I've been turning my style rules, like I really hate positional parameters in functions, and turning them into ESLint rules so
84
that I can lint Claude into the right thing without needing to warn it or instruct it.
85
So creating these feedback loops for Claude is so, so powerful.
86
Now if you dig this, then this and all of my ideas are going to be in my newsletter, so you never miss a trick.
87
And the link for that is below if you want to click it and sign up.
88
But thank you so much for watching.
89
I've been having a blast making these videos and discovering lots of new ideas about Clawed.
90
What else do you think you can do with Hooks?
91
What else do you want to see me cover with Clawed Code?
92
And now of course Clawed is not the be all and end all, right?
93
Like I'm pretty keen to try out other coding agents too.
94
And the Hooks idea is not exclusive to Clawed too, so you can take this knowledge and probably take it to Codex or whatever you're using.
95
But anyway, that's enough rambling.
96
Thanks for watching and I'll see you in the next one.
Why practice speaking with this video?
This video focuses on how to effectively guide Claude Code, a coding assistant, to enhance productivity in software development tasks. By understanding the intricacies of command-line interface (CLI) commands and technical workflows, you can practice speaking in a context that involves specific jargon and concepts. Engaging with technical content like this not only helps to expand your vocabulary but also increases your confidence when discussing complex ideas. Practicing speaking about these concepts helps you articulate your thoughts clearly, a vital skill for anyone looking to improve their English pronunciation and overall communication in professional settings.
Grammar & Expressions in Context
While analyzing the video, several key structures can be identified that you can incorporate into your own speaking practice:
- “How do I prevent it from using certain commands?” – This structure helps when forming questions about restrictions, useful in discussions about choices and limitations.
- “I’ve advocated almost always deleting your claw.md.” – Here, the use of the present perfect tense emphasizes continuous advocacy, which is common in discussions about best practices.
- “Our plan here is to use…” – This phrase indicates a future intention, helpful in project discussions where planning and strategy are key.
- “You can run a hook when the session starts.” – The use of modal verbs (“can”) expresses possibility and permission, essential in collaborative environments.
- “I’m going to copy and paste this…” – The future continuous tense shows a planned action, a useful structure when explaining processes or steps involved in tasks.
Common Pronunciation Traps
As you practice speaking, pay attention to a few challenging words and phrases that may arise from this video. Mastering these will significantly enhance your English pronunciation:
- “deterministic” – This technical term can be tricky. Practice breaking it into syllables: de-ter-mi-nis-tic.
- “commands” – The pronunciation can change based on the context. Ensure you stress the second syllable: com-MANDS.
- “execution” – This word is often mispronounced. Focus on articulating each part clearly: ex-e-cu-tion.
- “permissions” – A common word in software discussions, pay attention to the “sh” sound in the middle: per-MIS-sions.
To improve your English pronunciation, consider using a shadowing app or engaging in shadowing techniques, where you mimic the speaker closely. This method can help refine your articulation while you work on shadowspeaks, making your speaking practice more effective and enjoyable.
Qu'est-ce que la technique du Shadowing ?
Le Shadowing est une technique d'apprentissage des langues fondée sur la science, développée à l'origine pour la formation des interprètes professionnels. Le principe est simple mais puissant : vous écoutez de l'anglais natif et le répétez immédiatement à voix haute — comme une ombre suivant le locuteur avec un décalage de 1 à 2 secondes. Les recherches montrent une amélioration significative de la précision de la prononciation, de l'intonation, du rythme, des liaisons, de la compréhension orale et de la fluidité.