agentleFS
k3ybladewielderk3ybladewielder/skills/bend-formalizer/SKILL.md

bend-formalizer

Skill7 starsChanged 53 days agoApache-2.0View on GitHub

Use this skill exclusively when the user explicitly asks to use Bend to mathematically prove a project or repository and requests the rewriting of logic in Bend, the declaration of properties in LAWS.bend, and the proof in PROOF.bend. Do not use it for general explanations, Bend generation without this formal objective, or generic validation.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the request merely mentions "using Bend" but does not ask for a mathematical formalization of a project/repository involving LAWS.bend …

    line 19
  2. Distinctive:

    Use exclusively the installed CLI to query the version, language, and library:

    line 57
  3. Very specific:

    If ~/.bend/bend (or bend) does not exist or is not executable, report that the installed Bend is unavailable.

    line 66
  4. Very specific:

    Do not install, update, or make network requests silently.

    line 74
  5. Very specific:

    Do not use concepts from Bend 1/HVM as current references: u24, i24, f24, run-rs, run-c, run-cu, gen-c, gen-cu, HVM2/interaction nets, …

    line 95
  6. Distinctive:

    Before creating any Bend tests or proofs, understand how the repository organizes them.

    line 99

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Bend Formalizer — Mathematical Formalization of Projects

Mandatory Activation Guard

Apply this skill only when the request explicitly aims to use Bend to mathematically prove a project or repository and includes the following workflow:

  1. rewrite the project's relevant logic in Bend;
  2. declare properties or invariants in LAWS.bend;
  3. prove the laws in PROOF.bend.

If the request merely mentions "using Bend" but does not ask for a mathematical formalization of a project/repository involving LAWS.bend and PROOF.bend, do not use this skill. This includes:

  • general questions or explanations about Bend;
  • Bend code generation without the goal of mathematical proof;
  • generic code validation without an explicit request to formalize the project in Bend;
  • using Bend solely as a runtime, implementation language, or parallel backend;
  • attempts to directly prove arbitrary code written in Python, TypeScript, Rust, C, or other languages.

This guard must be respected even if the skill is loaded manually. If any condition is missing, respond normally without applying this skill's workflow.

Proof Objective and Scope

The goal is to construct a verifiable formalization of the project's relevant behavior:

original project logic
        ↓ rewrite/modeling
Bend modules and functions
        ↓ specification
LAWS.bend
        ↓ checker-based proof
PROOF.bend

Bend does not automatically import external implementations, nor does it directly prove Python, TypeScript, Rust, or C code. If the original project is in another language:

  • select the logic and behavior that need to be formalized;
  • rewrite that behavior in Bend;
  • declare the properties of the formal version in LAWS.bend;
  • prove them in PROOF.bend;
  • explicitly state that the proof covers the Bend model.

Equivalence between the original implementation and the Bend model requires an additional demonstration of refinement or equivalence. Do not claim that the original code has been proven simply because the Bend rewrite passed the checker.

Sources of Truth

The operational and semantic source of truth is the Bend installation at ~/.bend/ — typically ~/.bend/bend — rather than a checkout or copy of the repository in another directory.

Use exclusively the installed CLI to query the version, language, and library:

~/.bend/bend --version
~/.bend/bend guide
~/.bend/bend base
~/.bend/bend --help

If ~/.bend/bend (or bend) does not exist or is not executable, report that the installed Bend is unavailable. If the user has already installed Bend in the default path, suggest:

export PATH="$HOME/.bend:$PATH"
rehash
bend --version

Do not install, update, or make network requests silently. When bend is in the PATH, confirm that command -v bend points to ~/.bend/bend or bend. Do not use lab/bend study/Bend, util-repos/bend, or any other checkout as the source of truth for the language, CLI, Base, or runtime.

The documentation and files of the target repository continue to be used solely to understand the logic to be formalized, locate tests, and identify the link between the original code and the Bend model. They do not replace the documentation provided by the installation at ~/.bend/.

Current Bend Model

The implementation must adhere to the Bend specifications found in the ~/.bend/ installation:

  • Python-like syntax, with explicit annotations and minimal inference;
  • pure language, with effects encapsulated in IO;
  • affine values by default (used at most once);
  • Data, Type, kinds, and quantities (-, affine, and +);
  • core types: Nat, U32, and F32;
  • match for branching; do not introduce if or switch;
  • terminating recursion, unless explicitly justified via @unsafe;
  • single-owner arrays with a[i] <- value updates;
  • independent calls using a b = f(x) g(y);
  • f!(x) to dispatch parallel calls to the GPU backend in native builds;
  • one module per file and aliased imports;
  • do blocks for IO, Maybe, Result, and other monads.

Do not use concepts from Bend 1/HVM as current references: u24, i24, f24, run-rs, run-c, run-cu, gen-c, gen-cu, HVM2/interaction nets, fold, bend, fork, open, if, or switch.

Mandatory Testing and Proof Structure

Before creating any Bend tests or proofs, understand how the repository organizes them. Examine directories, conventions, AGENTS.md, package manager configurations, and documentation.

Core Examples and Guides (in .agents/skills/bend/examples/)

  • Laws and Proofs: Specification and theorem proving with LAWS.bend and PROOF.bend.
  • Syntax Reference: Grammar rules, AST terms, and assignments.
  • Common Errors: Anti-patterns, compiler diagnostics, and idiomatic fixes.
  • Types and Functions: Data types, kinds, and affine variables.

Test Subskills Reference (in .agents/skills/bend/tests/)

  • Proof Tests: Theorem proving, equality types, reflexivity ({==}), and rewrites.
  • Check Tests: Type checking, kind inference, and affine linearity verification.
  • Halt Tests: Structural recursion and termination checking.
  • Show Tests: Goal hole diagnostics (?name, ?TODO) and proof inspection.
  • Base Tests: Standard library algebraic types and operations.
  • Compile Tests: Backend lowering and native C/CUDA generation.
  • Import Tests: Multi-file resolution, namespaces, and relative imports.
  • State Tests: Pure state threading and monadic state transitions.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.