Cpp logo

Cpp

Community
ericrisco
cpp

Use when writing, reviewing, modernizing, building, or debugging C++ - RAII and resource lifetime, smart-pointer ownership, move semantics and the Rule of Zero/Five, target-based CMake with FetchContent, and killing undefined behavior with ASan/UBSan/TSan plus clang-tidy. NOT borrow-checker / Result-Option / cargo memory safety (that is rust).

Overview

Publisherericrisco
Repositoryrsc-harness
Skill namecpp
Stars
239
Forks
24
Bundled files
5
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 5 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by ericrisco on GitHub. Read the source before you install it.

Installation

Install the Cpp AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/ericrisco/rsc-harness/tree/main/skills/cpp
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/ericrisco/rsc-harness.git /tmp/rsc-harness
mkdir -p .claude/skills
cp -r /tmp/rsc-harness/skills/cpp .claude/skills/cpp
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Cpp in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Cpp on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Cpp is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Modern C++

Write, review, modernize, build, and debug C++ the way the C++ Core Guidelines intend: RAII for every resource, ownership made explicit through smart pointers and values, no undefined behavior by construction, and a target-based CMake build proven clean under sanitizers.

Targets C++20/23 for production today. C++23 is ISO/IEC 14882:2024; WG21 froze C++26's technical content on 2026-03-28 (ISO publication follows) — adopt C++26 features only behind confirmed compiler support. Compiler matrix:

CompilerC++23C++26Flag
GCCsince 11since 14 (GCC 16.1 covers most of C++26)-std=c++23 / -std=c++26
Clang13–18 progressivelyin progress (Clang 23 dev)-std=c++23 / -std=c++2c
MSVClatestpartial/std:c++23 / /std:c++latest

Delegate: borrow-checker, Result/Option, cargo, ownership-via-compiler -> rust — C++ buys safety with discipline (RAII + smart pointers + sanitizers); do not conflate the mechanisms. Language-agnostic threat modeling, authz, OWASP-class review -> secure-coding; the C++-specific memory/UB controls (bounds, lifetime, integer overflow, format-string, sanitizers) stay here. Containerizing and shipping the binary -> deployment; this skill stops at the CMake build + a sanitizer-CI note.

Decision rules

Apply these on every C++ edit:

  1. Rule of Zero first. Manage resources with members that already do it (vector, string, unique_ptr); write no destructor/copy/move at all. Why: hand-written special members are the #1 source of leaks and double-frees.
  2. Value by default. Pass and return by value for small/copyable types; reach for the heap only when you need polymorphism, shared lifetime, or a large/stable address. Why: values can't dangle.
  3. Name the owner. Exactly one type owns each resource; everyone else borrows. Why: ambiguous ownership is how use-after-free is born.
  4. make_unique/make_shared, never new. So no naked owning pointer ever exists.
  5. Never an owning raw pointer. Raw pointers/references are non-owning borrows only.
  6. Borrow with span / string_view / const T&. Pass a view, not a copy or an owner, for read access. Why: zero-copy, and the callee provably can't free what it doesn't own.
  7. const and constexpr by default. Why: the compiler enforces what you don't mutate and moves work off the hot path.
  8. No UB by construction. No use-after-move, OOB index, signed overflow, uninitialized read, or data race.
  9. Sanitizers + warnings-as-errors in CI. Build and test under -fsanitize=address,undefined with -Werror.
  10. Target-based CMake only. target_link_libraries / target_compile_features, never directory-level include_directories/link_libraries. Why: directory commands leak flags globally and break composition.

Ownership & smart pointers

Pick the type from the need, not from habit:

NeedUse
Exclusive owner, one place frees itstd::unique_ptr<T>
Genuinely shared lifetime (multiple owners, last one frees)std::shared_ptr<T>
Observe / break a shared_ptr cycle, no ownershipstd::weak_ptr<T> (.lock() to use)
Read-only borrow of contiguous range / stringstd::span<const T> / std::string_view
Borrow a single object, non-owningconst T& / T& / T* (never owning)
Small, copyable, value-likethe value itself — no heap

Default to unique_ptr; only escalate to shared_ptr when ownership is actually shared, and prove the shared case isn't a disguised single owner first — shared_ptr is not "the safe default."

cpp
// Bad: naked owning pointer; leaks on the throw, double-frees if you copy the handle.
Widget* w = new Widget(cfg);
configure(w);            // if this throws, w leaks
delete w;

// Good: ownership is the type; freed exactly once, exception-safe, no delete to forget.
auto w = std::make_unique<Widget>(cfg);
configure(*w);
cpp
// Bad: parent <-> child shared_ptr cycle -> neither refcount hits zero -> leak forever.
struct Node { std::shared_ptr<Node> parent, child; };

// Good: child owns down, parent observes up. Cycle broken; lock() before use.
struct Node {
    std::shared_ptr<Node> child;   // owns
    std::weak_ptr<Node>   parent;  // observes
};
if (auto p = node.parent.lock()) { /* p is a valid shared_ptr here */ }

When an object must hand out a shared_ptr to itself, derive from std::enable_shared_from_this<T> and call shared_from_this() — never wrap this in a fresh shared_ptr (that creates a second, independent refcount and a guaranteed double-free).

Deeper ownership/move reasoning -> references/move-and-templates.md.

RAII

Tie every resource — heap memory, file, socket, mutex, OS handle — to an object's lifetime; the destructor releases it. Why: cleanup then happens on every exit path (return, exception, break) for free, with no GC and no finally.

Use the standard guards before writing your own:

cpp
std::lock_guard  lock(mtx_);            // locks now, unlocks at scope end (C++17 CTAD)
std::scoped_lock locks(a_mtx, b_mtx);   // multiple mutexes, deadlock-free acquisition
std::unique_lock lk(mtx_);              // movable / deferrable, for condition_variable
std::ifstream    in("data.txt");        // closes in its destructor

When you wrap a C resource yourself, make the destructor release and disable copies (Rule of Five or unique_ptr with a custom deleter):

cpp
// RAII wrapper for a FILE*: closes once, can't leak, can't double-close.
class File {
public:
    explicit File(const char* path, const char* mode) : f_(std::fopen(path, mode)) {
        if (!f_) throw std::runtime_error("open failed");
    }
    ~File() { if (f_) std::fclose(f_); }
    File(const File&) = delete;                 // not copyable
    File& operator=(const File&) = delete;
    File(File&& o) noexcept : f_(std::exchange(o.f_, nullptr)) {}        // move = steal
    File& operator=(File&& o) noexcept { std::swap(f_, o.f_); return *this; }
    FILE* get() const noexcept { return f_; }
private:
    FILE* f_{};
};
// Even simpler when a deleter suffices — let unique_ptr own it (Rule of Zero):
auto fp = std::unique_ptr<FILE, decltype(&std::fclose)>(std::fopen("d", "r"), &std::fclose);

Move semantics & Rule of Zero/Five

Every expression is an lvalue (has a name, persists) or an rvalue (a temporary, about to die). std::move does not move anything — it casts an lvalue to an rvalue so a move constructor/assignment can steal its guts instead of copying. After you move from an object, it is valid but unspecified: only assign to it or destroy it; reading it is use-after-move (a real bug ASan/UBSan won't catch — clang-tidy will).

  • Rule of Zero (default): manage nothing by hand; let the compiler generate all five special members. This is correct for the vast majority of types.
  • Rule of Five: the moment you write one of destructor / copy-ctor / copy-assign / move-ctor / move-assign, you must reason about all five. If you're writing them, you probably should have used a unique_ptr/vector member and gone back to Rule of Zero.
  • Move ops must be noexcept. Why: std::vector reallocation only moves elements instead of copying them when the move is noexcept — otherwise it silently falls back to copies for the strong exception guarantee.
cpp
std::vector<std::string> v;
v.push_back(std::move(name));   // transfers the buffer; `name` is now empty-but-valid
// Bad: use-after-move — `name` holds an unspecified state here.
log(name);                      // don't. Reassign name first, or just don't read it.

Return local objects by value and let RVO / copy elision remove the copy — do not return std::move(local), which pessimizes by blocking elision. Take a forwarding reference T&& plus std::forward<T>(x) only in generic code that must preserve value category.

Worked Rule-of-Five, perfect forwarding, CTAD, and C++20 concepts -> references/move-and-templates.md.

Avoiding UB (essentials)

Undefined behavior is the compiler's permission to assume the bug can't happen and optimize on that assumption — so the symptom is often a distant crash or a "works in Debug, breaks in Release." Pair static analysis (clang-tidy, cppcheck) with dynamic sanitizers; they catch disjoint bug classes.

SanitizerFlagCatches
AddressSanitizer-fsanitize=addressuse-after-free, heap/stack buffer overflow, double-free
UndefinedBehaviorSanitizer-fsanitize=undefinedsigned overflow, null/misaligned deref, bad shifts, invalid enum
ThreadSanitizer-fsanitize=threaddata races

Combine ASan + UBSan in one build (-fsanitize=address,undefined); run TSan alone (it's incompatible with ASan). Always add -fno-omit-frame-pointer -g for readable reports.

cpp
// Bad: returns a dangling reference into a destroyed temporary -> use-after-free, ASan fires.
const std::string& name() { std::string s = build(); return s; }   // s dies at return

// Good: return by value; RVO makes it free.
std::string name() { return build(); }

The full catalog (lifetime, OOB, signed overflow, strict-aliasing, uninitialized, data races, use-after-move), which sanitizer surfaces each, the canonical fix, and a "reading an ASan report" walkthrough -> references/undefined-behavior.md.

Modern CMake (essentials)

Target-based only. State requirements on the target, never globally:

cmake
cmake_minimum_required(VERSION 3.21)
project(app LANGUAGES CXX)

set(CMAKE_EXPORT_COMPILE_COMMANDS ON)   # feeds clang-tidy / clangd

include(FetchContent)                   # FetchContent ships with CMake since 3.11
FetchContent_Declare(Catch2
    GIT_REPOSITORY https://github.com/catchorg/Catch2.git
    GIT_TAG        v3.7.1)
FetchContent_MakeAvailable(Catch2)      # its targets work just like find_package targets

add_executable(app src/main.cpp)
target_compile_features(app PRIVATE cxx_std_23)     # request the standard on the target
target_compile_options(app PRIVATE -Wall -Wextra -Wpedantic -Werror)
target_link_libraries(app PRIVATE Catch2::Catch2WithMain)

Full template (src/include/tests layout, CMakePresets.json with debug/asan/release presets, fmt + GoogleTest via FetchContent, per-compiler warning + sanitizer flags, install/export) -> references/cmake.md.

Standard-library idioms

Reach for the library before hand-rolling:

cpp
#include <algorithm>
#include <ranges>
#include <expected>   // C++23
#include <format>     // C++20

// Ranges over raw index loops — no off-by-one, no manual bounds.
auto evens = nums | std::views::filter([](int n){ return n % 2 == 0; });
std::ranges::sort(v);

// std::expected (C++23) over out-params / sentinel returns / exceptions for expected failure.
std::expected<Config, std::string> load(std::string_view path);
if (auto cfg = load(p)) use(*cfg); else log(cfg.error());

std::optional<User> find(int id);            // "maybe absent", not a magic -1 / nullptr
std::span<const int> view(v);                // borrow a contiguous range, no copy, no owner
auto [it, inserted] = m.try_emplace(k, val); // structured bindings
enum class Color { Red, Green };             // scoped, no implicit int conversions
std::string msg = std::format("{} of {}", i, n);  // type-safe, no printf format-string UB

Prefer at() or a range-checked view when the index isn't provably in bounds; operator[] on a bad index is UB, not an exception.

Testing & tooling

  • Tests: Catch2 or GoogleTest pulled via FetchContent (above); run them with ctest.
  • Static analysis: clang-tidy -p build (reads compile_commands.json) and cppcheck — they catch use-after-move, missing noexcept, and lifetime bugs the compiler won't.
  • Dynamic analysis: run the test target under ASan+UBSan so tests prove no UB on covered paths.
  • Format: clang-format -i with a checked-in .clang-format.
  • Local gate: ./scripts/verify.sh from the project root runs format + an ASan/UBSan, warnings-as-errors build + ctest + optional tidy/cppcheck. Missing tools are skipped, not failed.

Anti-patterns

RationalizationReality / Do instead
"I'll just new/delete carefully"One early return or throw and you leak/double-free. make_unique, always.
"A raw owning pointer is faster"unique_ptr is zero-overhead; the cost is imaginary, the leak is real.
"shared_ptr everywhere is the safe default"Shared ownership invites cycles + atomic refcount cost. Default unique_ptr; share only when truly shared.
"The C-style cast is fine, I know the type"Use static_cast/dynamic_cast; C casts silently reinterpret and hide bugs.
"Skip noexcept on the move ctor"vector then copies instead of moving on realloc. Mark moves noexcept.
"UB won't happen on my compiler"UB lets the optimizer delete your checks; "works in Debug" proves nothing.
"No sanitizers, it ran fine"It ran; it wasn't correct. Build+test under ASan+UBSan.
"A hand-rolled Makefile is simpler"It rots and leaks flags. Target-based CMake is the contract.
"v[i] is in range, I checked"If it's not provable, use .at() or a checked view; OOB is UB.
"return std::move(local) to be fast"It blocks RVO and is slower. Return the local by value.
"using namespace std; in a header"Pollutes every includer; ODR/ambiguity bugs. Never in a header.
"An out-param instead of returning the value"Return by value (RVO) or optional/expected; out-params hide aliasing and UB.
"A global / singleton is simpler"It's hidden shared mutable state -> data races + untestable. Inject it.

Quick reference

TaskCommand / idiom
Configure + buildcmake -S . -B build && cmake --build build
Configure with sanitizerscmake -S . -B build -DCMAKE_CXX_FLAGS="-fsanitize=address,undefined -g"
Testctest --test-dir build --output-on-failure
ASan + UBSan-fsanitize=address,undefined -fno-omit-frame-pointer -g
ThreadSanitizer (alone)-fsanitize=thread -g
Formatclang-format -i src/*.cpp
Static analysisclang-tidy -p build src/*.cpp · cppcheck --enable=warning src/
Std flagGCC/Clang -std=c++23 (C++26: GCC -std=c++26, Clang -std=c++2c), MSVC /std:c++23
Local gate./scripts/verify.sh (run in your project root)

Project grounding (02-DOCS)

In a project with a 02-DOCS/ layer (the harness Karpathy wiki), read 02-DOCS/wiki/stack/cpp.md first and stay consistent with it. If it is missing or stale, write this project's real choices there — std version and compiler matrix, CMake layout and presets, the sanitizer/warning policy, the ownership/error conventions — index it in 02-DOCS/wiki/index.md (the Knowledge map; root CLAUDE.md keeps only a pointer to it), and bump its Updated date in the same change as any convention change. No 02-DOCS/ layer? Skip silently (optionally suggest harness). Conventions are recorded, not gated — never block the task on this.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Cpp AI skill do?

Use when writing, reviewing, modernizing, building, or debugging C++ - RAII and resource lifetime, smart-pointer ownership, move semantics and the Rule of Zero/Five, target-based CMake with FetchContent, and killing undefined behavior with ASan/UBSan/TSan plus clang-tidy. NOT borrow-checker / Result-Option / cargo memory safety (that is rust).

Why use Cpp on TypingMind?

Because you install it once and use it with any model. Cpp is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Cpp in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/ericrisco/rsc-harness/tree/main/skills/cpp. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Cpp?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Cpp?

As many as you like. As long as a model supports skills, you can use Cpp with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Cpp AI skill free?

Yes. It is published on GitHub by ericrisco under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇