ЁЯдЦ AI Agent Skill Specification

Sanskrit Next Language Skill

Equip Claude, ChatGPT, Cursor, Copilot, or Google Antigravity with complete native Sanskrit Next (v2.0+) language capabilities in a single prompt.

ЁЯУЬ Dual-Script Invariance

Teaches AI models the 100% lexical equivalence between Paninian Devanagari (рдХрд╛рд░реНрдп, рдорд╛рди, рджрд┐рд╢) and Latin ASCII aliases (fn, let, Tensor).

тЪб Real Empirical Telemetry

Equipped with verified benchmark figures: 1.84ms cold boot, 3.1MB heap memory, 4.2MB binary, and 4.09ms per 100,000 autodiff evaluations.

ЁЯФм Tensors & Autodiff

Complete semantics for multi-dimensional strided tensors (рджрд┐рд╢), GEMM matrix multiplication (@), and forward-mode dual numbers.

ЁЯУД .agents/skills/sanskrit-language/SKILL.md
---
name: sanskrit-language
description: Comprehensive expert engineering and syntax reference for the Sanskrit Next programming language (v2.0+), covering dual-script syntax (Devanagari & Latin), type system, tensors, autodiff, standard library, and CLI tooling.
---

# Sanskrit Next (рд╕рдВрд╕реНрдХреГрдд) Language Engineering Skill

## 1. Architectural Model & Core Tenets
1. Dual-Script Invariance: 100% syntactic & semantic parity between Devanagari (U+0900тАУU+097F) and ASCII.
2. Tiered Execution: Sub-2ms cold boot Tier-0 Bytecode VM (1.84 ms) + optimizing MLIR/LLVM release pipeline.
3. Memory Safety: Zero-GC, deterministic RAII ownership, parameter modes (рдкрдарди/read, рдкрд░рд┐рд╡рд░реНрддрди/mut, рд╕реНрд╡рд╛рдорд┐рддреНрд╡/owned).
4. Scientific Primitives: Strided tensors (рджрд┐рд╢ / Tensor), GEMM operator (@), forward-mode autodiff (4.09 ms / 100k evals).
5. Zero-Dependency Native Architecture: 100% native Rust, single 4.2 MB binary.

## 2. Dual-Script Keywords
- Function: рдХрд╛рд░реНрдп <рдирд╛рдо>(<рдорд╛рдкрджрдгреНрдб>): <рдкреНрд░рдХрд╛рд░>  |  fn (): 
- Variable: рдорд╛рди <рдирд╛рдо> = <рдореВрд▓реНрдп>               |  let  = 
- Constant: рд╕реНрдерд┐рд░ <рдирд╛рдо> = <рдореВрд▓реНрдп>             |  const  = 
- If / Else: рдпрджрд┐ <рд╢рд░реНрдд>: ... рдЕрдиреНрдпрдерд╛: ...      |  if : ... else: ...
- Loops: рдпрд╛рд╡рддреН <рд╢рд░реНрдд>: ...                   |  while : ...
- Return: рдкреНрд░рддреНрдпрд╛рдЧрдо <рдореВрд▓реНрдп>                   |  return 
- Output: рдореБрджреНрд░рдг(...args)                    |  print(...args)
- Panic: рд╡рд┐рдлрд▓(рд╕рдиреНрджреЗрд╢)                        |  panic(msg)
- Assert: рдирд┐рд╢реНрдЪрдп(рд╢рд░реНрдд, рд╕рдиреНрджреЗрд╢)               |  assert(cond, msg)

## 2.1 Phonetic Romanized Typing (DX Superpower)
Type phonetic English on any QWERTY keyboard; autocompletes to Devanagari:
- mudran / mud тЮФ рдореБрджреНрд░рдг(...) [print]
- karya / kary тЮФ рдХрд╛рд░реНрдп [fn]
- maan / man   тЮФ рдорд╛рди [let]
- sthir        тЮФ рд╕реНрдерд┐рд░ [const]
- yadi         тЮФ рдпрджрд┐ [if]
- anyatha      тЮФ рдЕрдиреНрдпрдерд╛ [else]
- yaavat       тЮФ рдпрд╛рд╡рддреН [while]
- dish         тЮФ рджрд┐рд╢[...] [Tensor]
- purnank      тЮФ рдкреВрд░реНрдгрд╛рдЩреНрдХ [i32]
- sutra        тЮФ рд╕реВрддреНрд░ [str]
- tarka        тЮФ рддрд░реНрдХ [bool]
- dashamalav   тЮФ рджрд╢рдорд▓рд╡ [f64]

## 3. Types
- Primitives: рдкреВрд░реНрдгрд╛рдЩреНрдХ (i32), рджрд╢рдорд▓рд╡ (f64), рддрд░реНрдХ (bool), рд╕реВрддреНрд░ (str), рдЕрдХреНрд╖рд░ (char), рд╢реВрдиреНрдпрдореН (void)
- Tensors: рджрд┐рд╢[dtype, shape] / Tensor[dtype, shape]
  - рджрд┐рд╢.zeros([M, N]), рджрд┐рд╢.ones([M, N]), рджрд┐рд╢.randn([M, N]), рджрд┐рд╢.eye(N)
  - Tensor @ Tensor (GEMM), .transpose(), .reshape(), .sum(), .mean()
- Automatic Differentiation: Dual(primal, tangent), рдЕрд╡рдХрд▓рди(f, x) [diff], рдкреНрд░рд╡рдгрддрд╛(f, x) [grad]

## 4. Standard Library Modules
- std.core: рдореБрджреНрд░рдг (print), рд╡рд┐рдлрд▓ (panic), рдирд┐рд╢реНрдЪрдп (assert)
- std.tensor: Linear algebra, strided views, matrix multiplication (@)
- std.autodiff: Forward-mode automatic differentiation
- std.linalg: GEMM, vector norms, matrix operations
- std.math: Trigonometric, logarithmic, and power routines

## 5. Toolchain CLI
- sanskrit run        : Sub-2ms Tier-0 Bytecode VM
- sanskrit run --release  : Optimizing MLIR/LLVM pipeline
- sanskrit check      : Multi-span syntax & type diagnostics
- sanskrit test                 : Automated test discovery (**/*.test.skt)
- sanskrit bench                : Hardware telemetry & benchmarks
- sanskrit doctor               : GPU, AMX, cache & toolchain health
- sanskrit fmt                  : Dual-script canonical formatter
- sanskrit lsp                  : Embedded LSP 3.17 daemon

## 6. VS Code Extension (v2.0.3)
- Interactive Sanskrit VM Terminal Runner (Ctrl+Alt+N / Cmd+Alt+N)
- Run Selection in Terminal (Ctrl+Alt+S / Cmd+Alt+S)
- Run with Custom Arguments...
- 35+ Settings for runtime backends, hardware acceleration, and hints