← Back

Open Source Tools

Claude Code skills I built for my own development workflow. Each one solves a specific problem I kept running into. They work as standalone skills you can install in one command.

Solana Development (8)Security (2)Developer Workflow (3)13 total

Solana Development

solana-master

Solana
216 lines

Full-stack Solana dApp development with strict coding standards. Covers UI, SDK, programs, and testing with enforced TypeScript style and TDD practices.

Enforces a complete development methodology: @solana/client + @solana/react-hooks for UI, @solana/kit for SDK, Anchor or Pinocchio for programs, LiteSVM/Mollusk/Surfpool for testing. Includes coding standards (Array<T> syntax, no abbreviations, no any), deletionist philosophy, and security checklists baked into every workflow step.

Install

carbon-decoder

Solana
651 lines

Build correct Carbon framework decoders for Solana program indexing. Covers codegen CLI, discriminators, account and instruction decoding, events, and pipeline wiring.

The Carbon framework indexes Solana transactions but building decoders correctly is tricky. This skill covers the full lifecycle: codegen CLI setup, trait implementation, discriminator computation, ArrangeAccounts patterns, event decoding, and wiring decoders into the Carbon pipeline. Prevents the most common failure mode: silent parse failures where transactions are dropped instead of erroring.

Install

borsh-schema-gen

Solana
185 lines

Generate TypeScript Borsh codecs from on-chain Rust struct definitions. Paste a Rust struct, get a working decode function.

Handles the tedious translation between Rust struct layouts and TypeScript Borsh schemas. Correctly maps Rust types (u64, Pubkey, Option<T>, Vec<T>, enums with data) to their Borsh codec equivalents. Produces working deserialize functions you can use immediately to read Solana account data.

Install

discriminator-verify

Solana
156 lines

Verify or compute Anchor instruction and account discriminators. Catches silent parse failures from discriminator mismatches.

Anchor uses SHA256-based discriminators to identify instructions and accounts. A mismatch means transactions are silently dropped rather than failing visibly. This skill computes the correct discriminator from the instruction or account name and cross-checks it against your parser code.

Install

pda-audit

Solana
158 lines

Audit PDA derivations for correctness. Wrong seeds produce wrong addresses silently, causing runtime failures or worse.

Traces every PDA derivation in the codebase and verifies seeds and program IDs match the on-chain program expectations. Catches the class of bugs where a PDA resolves to an account owned by a different program, or where seed ordering produces a valid but incorrect address.

Install

solana-dev

Solana
104 lines

End-to-end Solana development playbook. Framework-kit-first approach with @solana/kit, wallet-standard connections, and modern testing.

Opinionated playbook covering the 2026 Solana stack: @solana/client + @solana/react-hooks for React UI, @solana/kit for all client/RPC code, Anchor or Pinocchio for programs, Codama for client generation, and LiteSVM for fast local testing. Includes migration guidance from legacy web3.js.

Install

solana-security-checklist

Solana
207 lines

Client-side security checklist for TypeScript code that interacts with Solana programs. Covers swap execution, wallet ops, fees, and account data.

Adapts Solana program security best practices to the TypeScript client layer where the same bug classes appear. Checks for: unchecked account ownership, missing signature verification, incorrect PDA seeds, unsafe deserialization, fee calculation overflow, and stale account data. Run this before merging any client code that touches on-chain interactions.

Install

token2022-audit

Solana
138 lines

Audit code for Token 2022 vs original SPL Token handling gaps. Ensures both TOKEN_PROGRAM_ID and TOKEN_2022_PROGRAM_ID are handled correctly.

Token 2022 introduced transfer fees, interest-bearing tokens, and confidential transfers. Code that only handles the original SPL Token program will silently fail or produce wrong results for Token 2022 assets. This skill finds every place that reads token data and checks for dual-program handling.

Install

Security

vulnhunter

Security
324 lines

Vulnerability detection and variant analysis. Finds dangerous APIs, footgun patterns, and error-prone configurations across codebases.

Combines sharp-edges detection (dangerous APIs, footgun patterns, error-prone configurations) with variant hunting methodology (find all instances of a vulnerability pattern across the codebase). Inspired by Trail of Bits' approach to security research. Produces a prioritized list of findings with severity, location, and remediation guidance.

Install

code-recon

Security
504 lines

Deep architectural context building for security audits. Rapidly understand unfamiliar codebases before vulnerability analysis.

Builds comprehensive architectural understanding through ultra-granular code analysis. Maps trust boundaries, data flows, authentication patterns, and attack surfaces. Designed for security auditors who need to rapidly understand unfamiliar codebases before diving into vulnerability research. Inspired by Trail of Bits methodology.

Install

Developer Workflow

tla-precheck

Workflow
164 lines

Design and verify state machines using a TypeScript DSL. Catches impossible transitions, stuck states, and invariant violations before they ship.

Formal verification for state machines without learning TLA+. Write your machine in TypeScript, define invariants (e.g., 'at most one running item per user'), and the skill verifies all reachable states satisfy them. Use for billing flows, subscription lifecycles, agent orchestration, queue processing, or any state machine where a bug means corrupted data or stuck users.

Install

improve-claude-md

Workflow
275 lines

Optimize CLAUDE.md files for better AI instruction adherence. Adds conditional blocks and restructures for higher compliance rates.

Uses <important if> conditional blocks to make instructions context-aware rather than always-on. Restructures CLAUDE.md files so the AI agent follows instructions more reliably. Cuts instructions the agent can discover from existing code patterns, and strengthens the ones that actually need to be explicit.

Install

youtube-transcribe

Workflow
373 lines

Transcribe YouTube videos or full playlists into structured Obsidian notes. Free, uses yt-dlp, no external transcription service needed.

Extracts captions via yt-dlp, structures them into Obsidian-compatible markdown with YAML frontmatter, auto-generated tags, timestamps, and wikilinks. Handles single videos and full playlists. Infers vault path from channel name and content topic. Part of my Obsidian-based research workflow where video content becomes searchable, linkable knowledge.

Install

How to install

Each skill is a SKILL.md file that Claude Code reads as context when working on relevant tasks. Clone the repo once, then copy any skill into your project.

$ git clone https://github.com/Jerome2332/claude-skills.git
$ cp -r claude-skills/<skill-name> .claude/skills/<skill-name>