[Resource]: Claude Code System Prompts

[bl-ue]

Display Name

Claude Code System Prompts

Category

Workflows & Knowledge Guides

Sub-Category

General

Primary Link

https://github.com/Piebald-AI/claude-code-system-prompts

Secondary Link

No response

Author Name

Piebald AI

Author Link

https://github.com/Piebald-AI

License

MIT

Other License

No response

Description

All parts of Claude Code's system prompt, including builtin tool descriptions, sub agent prompts (Plan/Explore/Task), utility prompts (CLAUDE.md, compact, Bash cmd, security review, agent creation, etc.). Updated for each Claude Code version.

Validate Claims

No response

Additional Comments

No response

Recommendation Checklist

• I have checked that this resource hasn't already been submitted
• My resource provides genuine value to Claude Code users, and any risks are clearly stated
• All provided links are working and publicly accessible
• I am submitting only ONE resource in this issue
• I understand that low-quality or duplicate submissions may be rejected

[hesreallyhim]

thanks @bl-ue ... gotta think this on for a sec, i think it's prob worth it and low-liability (for me, that is). can i ask what method you use to get these and do you plan to keep it up to date with new releases?

Tip

You can encrypt your response with my PGP key if you want.

EDIT: This is totally up to you, but it might be nice include some tiny bit of context about when it might be a good/bad idea to edit the system prompt. also maybe mention there's also the Claude system prompt in addition, AFAIK, right?

[bl-ue]

Hi @hesreallyhim! We use a script to extract the prompts, and yes, we certainly do plan to keep it up to date, as we use them in tweakcc. The resource being constantly updated provides a lot of value.

The Claude system prompt is actually not included. It's not inside Claude Code—our collection only contains CC's system prompts.

[github-actions]

🤖 Validation Results

✅ All validation checks passed!

Your submission is ready for review by a maintainer.

Validated Data:

{
  "display_name": "Claude Code System Prompts",
  "category": "Workflows & Knowledge Guides",
  "subcategory": "General",
  "primary_link": "https://github.com/Piebald-AI/claude-code-system-prompts",
  "secondary_link": "",
  "author_name": "Piebald AI",
  "author_link": "https://github.com/Piebald-AI",
  "license": "MIT",
  "description": "All parts of Claude Code's system prompt, including builtin tool descriptions, sub agent prompts (Plan/Explore/Task), utility prompts (CLAUDE.md, compact, Bash cmd, security review, agent creation, etc.). Updated for each Claude Code version.",
  "active": "TRUE",
  "last_checked": "2025-12-18:05-36-09"
}

---

_{This comment is automatically updated when you edit the issue.}

[hesreallyhim]

The Claude system prompt is actually not included. It's not inside Claude Code—our collection only contains CC's system prompts.

No that makes sense, I just was suggesting that for people who don't know that, it might be good to know that when you say "all the CC system prompts" it's not preciely correct... anyway that's fine.

I think it would be good to link the script and the repo so that people can validate where the information is coming from - how do you feel about me including that in the resource description? (unless it's available in the repo and I didn't see it)

anyway, I don't want to nit-pick your code or do a review or anything, but it is a remarkably long script, and it seems like you're using pattern matching instead of some kind of JSON parsing, which is kind of surprising to me...

I've changed the category because, well, how to put it, "Official" means like, stuff that Anthropic has released.

Last thing, I decided to truncate the description a little so that things like "16 prompts" e.g. don't become stale in the future when they add or remove stuff, hope that's all right with you.

/approve

[github-actions]

✅ Resource Approved!

❌ There was an error creating the pull request.

Please check the workflow logs for details.

[bl-ue]

No that makes sense, I just was suggesting that for people who don't know that, it might be good to know that when you say "all the CC system prompts" it's not preciely correct... anyway that's fine.

That's a good idea; I didn't think about users not knowing the difference. I'll put a note in per your suggestion.

anyway, I don't want to nit-pick your code or do a review or anything, but it is a remarkably long script, and it seems like you're using pattern matching instead of some kind of JSON parsing, which is kind of surprising to me...

To get the system prompts, we search for long strings in cli.js, which is a huge minified JS file, so AST parsing and pattern matching is the only option.

I've changed the category because, well, how to put it, "Official" means like, stuff that Anthropic has released.

Makes sense. "Knowledge Guides" fits better than "Official Documentation."

Last thing, I decided to truncate the description a little so that things like "16 prompts" e.g. don't become stale in the future when they add or remove stuff, hope that's all right with you.

Great thinking! As a matter of fact, it was already stale—the number of builtin tools has increased to 21 over time.

[hesreallyhim]

@bl-ue

To get the system prompts, we search for long strings in cli.js, which is a huge minified JS file, so AST parsing and pattern matching is the only option.

OK that's interesting. What about using proxy tools like (better-)ccflare or mitm-proxy, can't you just read them straight off of the message channel?