crypto-infra-skills

crypto-infra-skills

Production-grade engineering skills, agents, and commands for AI coding agents — built for crypto / Web3 backend work and covering the full lifecycle from spec to ship.

This is a Claude Code plugin. It teaches your agent how to write, test, review, and ship wallet infrastructure, indexers, matching engines, swap integrations, crypto-card products, and the vendor integrations behind them — with the rigor of a staff engineer who has shipped these systems to production.

Language guidance is pinned to canonical references so authoring and review share one definition of "good":

• Go → Effective Go
• TypeScript → Effective TypeScript (Dan Vanderkam)
• Rust → The Rust Book + Rust API Guidelines

---

What's inside

| Layer | Count | What it is | |---|---|---| | Commands | 7 | User-facing entry points (/build, /ship, …) that compose personas + skills | | Agents (personas) | 5 | Single-role reviewers/auditors that produce a structured report | | Skills | 55 | Workflows with steps + exit criteria, in three families (below) | | Guidelines | 5 | Shared checklists (security, performance, testing, a11y, orchestration) |

The three layers compose like this (see agents/README.md for detail):

| Layer | Job | Example | |---|---|---| | Skill | The how — a workflow, invoked from inside a persona or command | incremental-implementation | | Persona | The who — adopts a viewpoint, produces a report | web3-backend-reviewer | | Command | The when — composes personas + skills | /review, /ship |

---

Install

In Claude Code:

# Add this repo as a plugin marketplace
/plugin marketplace add hryer/crypto-infra-skills

# Install the plugin
/plugin install crypto-infra-skills@crypto-infra-skills

Local development (from a clone of this repo):

/plugin marketplace add .
/plugin install crypto-infra-skills@crypto-infra-skills

Once installed, the commands (/build, /ship, …) become available, agents can be invoked, and skills auto-trigger based on what you're doing.

---

How to use

1. Commands — the main entry points

Type a slash command to run a workflow. Each composes the relevant skills and personas.

| Command | What it does | |---|---| | /spec | Write a structured specification before writing code | | /plan | Break work into small verifiable tasks with acceptance criteria + dependency order | | /build | Implement the next task incrementally — build, test, verify, commit | | /test | TDD workflow — write failing tests, implement, verify (Prove-It pattern for bugs) | | /review | Five-axis code review — correctness, readability, architecture, security, performance | | /code-simplify | Reduce complexity without changing behavior | | /ship | Pre-launch checklist via parallel fan-out to specialist personas → go/no-go |

A typical flow: /spec → /plan → /build (repeat per task) → /review → /ship.

2. Skills — auto-triggered or invoked

Skills load automatically when your request matches their description (e.g., asking about a Fireblocks integration pulls in vendor-fireblocks). You can also invoke one explicitly by name. Skills are the knowledge layer — you rarely call them directly; commands and agents pull them in.

3. Agents — when you want one expert's perspective

Invoke a persona for a focused review. They produce a structured report and do not call each other (you or a command orchestrate them).

| Agent | Role | Use for | |---|---|---| | code-reviewer | Senior Staff Engineer | Five-axis review before merge | | security-auditor | Security Engineer | OWASP-style vulnerability audit | | test-engineer | QA Engineer | Test strategy + coverage | | web3-backend-reviewer | Web3 Staff Engineer | Crypto backend review: reorg, idempotency, ledger conservation, vendor integrations, key mgmt, infra, distributed systems, debugging | | wallet-security-auditor | Wallet Security Engineer | Adversarial review of custody, signing, MPC/multisig, ERC-4337, Solana programs |

---

The skill taxonomy

55 skills in three families, distinguished by name prefix.

category-* — decision frameworks (13)

The entry point for a problem area. Each frames the use case, lays out vendor options, gives a build-it-yourself path, includes inline Go/TS/Rust + infra best practices, and branches to vendor-* skills.

| Skill | Covers | |---|---| | category-wallet-custody | Custody models + WaaS vendor selection (two-tier) | | category-swap-integration | DEX aggregators vs direct routers | | category-token-security-screening | Honeypot / rug detection | | category-price-feeds | Off-chain APIs vs on-chain oracles | | category-onchain-analytics | Labeled-wallet data, smart-money signals | | category-rpc-and-indexer | RPC + event indexing + fallback | | category-smart-contract-testing | Simulation, fork tests, Anchor/Foundry | | category-multichain-tx-management | Nonce/gas/RBF across chains | | category-internal-ledger | Double-entry, conservation, idempotency | | category-matching-engine | Order book, matching, HFT hot path | | category-mev-and-keepers | Searcher/keeper bots, revm, Flashbots | | category-crypto-card | Card issuing, auth/settlement, ledger | | category-onramp-offramp | Fiat on/off-ramp vendors |

vendor-* — integration guides (19)

How to integrate a specific vendor at the backend level: auth, KMS-managed secrets, SDK usage (TS + Go), webhook handling + idempotency, infra wiring, common mistakes, MCP integration where available, and a "when NOT to use this" section.

• Wallet/custody: fireblocks · fordefi · privy · dynamic · turnkey
• RPC/indexer: alchemy · quicknode · helius
• Swap: lifi
• Token security: birdeye
• Contract testing/debug: tenderly
• Prices/oracles: coingecko · coinmarketcap · coinapi · chainlink
• Analytics: nansen
• Card issuers: reap · rain · dcs

> Vendor skills carry a last-verified date and a <!-- TODO: fill in from latest vendor docs --> marker — refresh them against current vendor docs before relying on the integration detail.

Generic engineering skills (23)

Language-agnostic workflow + quality skills used across any task: spec/plan/build/test/review/ship discipline, debugging, git, CI/CD, API design, documentation, and the meta-skills. The authoring/testing/shipping skills carry inline Go + TypeScript + Rust + Web3 best practices:

• incremental-implementation — canonical home for language authoring idioms
• test-driven-development — canonical home for testing idioms
• shipping-and-launch — canonical home for the pre-launch checklist (incl. Web3 readiness)
• ci-cd-and-automation — Go/TS/Rust + Foundry/Anchor pipelines
• code-simplification — idiomatic refactors per language

Others: api-and-interface-design, browser-testing-with-devtools, code-review-and-quality, context-engineering, debugging-and-error-recovery, deprecation-and-migration, documentation-and-adrs, doubt-driven-development, frontend-ui-engineering, git-workflow-and-versioning, idea-refine, interview-me, performance-optimization, planning-and-task-breakdown, security-and-hardening, source-driven-development, spec-driven-development, using-agent-skills.

---

Example workflows

Integrate a custodial wallet vendor > "I'm adding Fireblocks for treasury signing."

Auto-loads category-wallet-custody (custody decision) → vendor-fireblocks (backend integration). Build with /build, then review with the web3-backend-reviewer and wallet-security-auditor before /ship.

Build a Solana indexer with a fallback > "Index our program's events on Solana with a backup RPC."

Loads category-rpc-and-indexer → vendor-helius (primary) + vendor-quicknode (fallback). Reviewer checks reorg/finality handling and the reconciliation backstop.

Ship a matching engine change > /ship

Fans out to the reviewer personas, runs the pre-launch checklist (including Web3 readiness: kill switch, RPC failover, reorg thresholds), and returns a go/no-go.

---

Repository layout

.
├── .claude/commands/      # 7 slash commands
├── .claude-plugin/        # plugin.json + marketplace.json
├── agents/                # 5 personas + README
├── skills/                # 55 skills (category-*, vendor-*, generic)
├── guidelines/            # 5 shared checklists
├── CONTRIBUTING.md
└── LICENSE

---

Contributing

See CONTRIBUTING.md. The bar: every skill/agent/command must read like it was written by someone who has shipped that kind of system to production at a serious crypto company.

License

MIT — see LICENSE.