The Senior Engineer's AI Prompt Vault by SublimeCoding — 50 production-grade prompts for Claude Code and Cursor, built at staff level: code review, debugging, refactoring, system design, and docs. Plus 5 drop-in .cursorrules files, 3 CLAUDE.md starters (Go, Elixir, AI-agent projects), an AGENTS.md starter, and a Notion blueprint.
Free · no app · no login · lifetime updates.
Prompts were v1. The next layer is AIOS — an open-source, markdown-based memory system that gives coding agents persistent knowledge across sessions. Jump to AIOS ↓
→ Get the Vault, free · 15 sample prompts → · FAQ →
⭐ If these prompts save you a production bug or a wasted hour, star the repo — it's how other engineers find it.
- AIOS — Persistent Memory for Coding Agents
- Claude Code Prompts
- Cursor Prompts for Senior Engineers
- AI Code Review Prompts
- CLAUDE.md & AGENTS.md Templates
- .cursorrules for Senior Engineers
Well-crafted prompts get you one good session. AIOS is what replaced them for me: a git-versioned markdown vault that gives Claude Code (or any agent) memory that survives between sessions — editorial curation instead of vector retrieval.
- Knowledge Map + Skill Map — read-first navigation, no embedding search needed
- ~25 portable skills — markdown capabilities that work across Claude Code, Cursor, and Codex
- Cross-repo wiring — hooks that connect your code repos to the vault
- Write guards + leak auditors — scope isolation for privacy-sensitive work
→ github.com/sublimecoder/aios · AIOS: An AI Operating System in Plain Markdown · Building an AI Memory That Outlives Any Single Agent
15 free, production-tested Claude Code prompts for senior engineers — browse them in /prompts. Each encodes how a staff engineer actually uses AI: structured thinking over pattern-matched answers, named failure modes, and a model that commits instead of hedges.
- Debugging — production incident, race condition, hypothesis-driven
- Refactoring — god class, illegal states, long functions
- System design — idempotency, rate limiter, pick a queue
- Docs & PRs — ADR, postmortem, PR description
Every prompt here is tool-agnostic — they work in Cursor exactly as they do in Claude Code. For Cursor specifically, start with the senior-engineer .cursorrules, then layer in the task prompts above.
Staff-level AI code review prompts that fight the way models actually fail at review:
- Security-first review — auth, input handling, external I/O.
- Blast radius analysis — before merging changes to shared infrastructure.
- Database migration safety — DDL against a live, large production DB.
AGENTS.mdstarter — production-grade agent-instructions template (works asCLAUDE.mdtoo).CLAUDE.mdstarters — drop them into any repo and the agent understands your conventions on the first message:
Free, staff-level .cursorrules — each works in Cursor, Claude Code, Windsurf, or any rules-file tool:
- senior-engineer.cursorrules — the baseline posture: read-first, trace every change, handle the error path, verify before "done".
- go-microservices.cursorrules — Go services: context propagation, error wrapping, bounded concurrency, graceful shutdown, table tests.
- elixir.cursorrules — Elixir/Phoenix: OTP supervision,
withpipelines, Ecto changeset discipline, context boundaries, let-it-crash. - security.cursorrules — production security: untrusted-input handling, secrets hygiene, authz at every boundary, prompt-injection containment.
The Vault bundles all of these (plus the AI-agent ruleset), the 50 prompts, and the CLAUDE.md starters in one download.
→ Get all 50 prompts + every .cursorrules file, free
Production AI. Distributed systems. Security.
13+ years shipping software. 100K+ users on live AI products. $400M+ in fintech assets protected. 40K+ telematics devices streaming.
Director of Information Security at Lavender. Backend at BlockFi. Elixir at AAMP Global. Co-founded a startup, scaled engineering from 1 to 15.
Bias toward clarity, observability, and predictable failure. The prompts here came from the real problem of teaching 14 other engineers to use AI without making their code worse — which is where the failure-mode framing comes from.
- AIOS: An AI Operating System in Plain Markdown — the git-versioned vault, cross-repo wiring, and nightly ingest loop behind the AIOS repo.
- Building an AI Memory That Outlives Any Single Agent — a portable, git-versioned knowledge layer any agent can read.
- How I Prompt Claude as a Staff Engineer (50 Prompts I Actually Use) — the prompts behind the Vault and how I actually use them.
- Ruflo (formerly Claude Flow): An Honest Deep Dive — 45K+ stars, 700K+ downloads. What it does well, and where it falls down.
- The Claude Code Resource Bible: 46 Tools Worth Knowing in 2026 — official tools, MCP servers, agents, automation. Curated, not a link dump.
- Detection Got Cheap. Remediation Didn't.
- The One File Where Your AI Model Names Live
- What a 96% Security Benchmark Actually Measures
- 10 Ecto Query Patterns for N+1 and Beyond
- Isolating Parallel Claude Code Agents in Phoenix
→ More at sublimecoding.com/blog
Selective, senior engagements where a Staff/Principal-level brain moves the needle:
- Production AI delivery — ship LLM features that survive real users, with the evals, guardrails, and cost discipline to back them.
- Backend & distributed systems at scale — the architecture, reliability, and observability work that keeps growth from breaking you.
- Fractional security & compliance leadership (vCISO) — SOC 2, AWS hardening, and an AI data-handling posture your enterprise prospects will actually trust.
→ Book a 30-min intro call · What 90 days of a fractional security engagement actually looks like
Go, Elixir, Ruby. GCP, AWS, Azure. Postgres, Kafka, RabbitMQ. Claude Code, Cursor, Codex daily since 2023.
sublimecoding.com · @sublimecoder · LinkedIn · jared@sublimecoding.com