hub
Best Prompts Directory
Curated prompts for every Claude Code job: code review, refactor, audit, plan, ship. Copy, paste, adapt.
The Directory
40 prompts that actually work.
Searchable. Copy-paste ready. Every prompt tested in real production sessions with Claude Code. 8 categories, from engineering to business. Filter by difficulty, search by keyword, copy in one click.
Engineering
8Interview-driven spec generation
Interview me about [feature] using AskUserQuestion. Ask about implementation, edge cases, tradeoffs. Don't ask obvious things. When done, write spec to SPEC.md
When to use: Starting any non-trivial feature. Forces you to think through requirements before writing code.
Multi-layer code review
Review all changed files. For each: 1) security issues (OWASP top 10), 2) N+1 queries, 3) missing input validation. Label severity.
When to use: Before merging any PR. Catches the three most common production issues in one pass.
Ruthless senior engineer review
You are a senior engineer with no patience for bad code. Review src/auth.ts. Start with the riskiest issue. No warm-up.
When to use: When you want unfiltered, prioritized feedback on a critical file.
Parallel hypothesis debugging with agents
Spawn 3 agents in parallel: Agent 1 investigates [bug] assuming race condition. Agent 2 assuming memory leak. Agent 3 assuming auth expiry. First to find root cause wins.
When to use: When you have a tricky bug and want to test multiple theories simultaneously.
Incremental API migration
Find all usages of [old_api] in src/. List them. Then migrate each file one by one, running tests after each change.
When to use: Migrating deprecated APIs safely. The one-by-one approach catches regressions early.
N+1 query detector
Analyze our codebase for N+1 queries. Search for patterns like loop + database call. Return file:line for each.
When to use: Performance audit. N+1 queries are the most common hidden performance killer.
Feature spec with rollback plan
Create a SPEC.md for [feature] that covers: data model, API contract, edge cases, test scenarios, rollback plan.
When to use: Before building any feature that touches the database or has external dependencies.
Auth flow walkthrough with ASCII diagram
Walk me through the auth flow step by step. Read the files as you go. Draw an ASCII diagram of the flow.
When to use: Onboarding to a new codebase or documenting an existing flow for the team.
Security
6Hardcoded secrets scanner
Scan the entire codebase for patterns matching API keys, tokens, passwords. Flag any that look hardcoded. Report file:line.
When to use: Before any public release, after onboarding new developers, or during periodic audits.
Auth vulnerability deep scan
Review src/auth/ for these vulnerabilities: JWT algorithm confusion, token expiry bypass, privilege escalation. Be specific.
When to use: When auth is business-critical. These three are the most exploited auth vulnerabilities.
Simulated penetration test
You are running a penetration test on our API. Describe the 5 most likely attack vectors based on the code you see.
When to use: Before launching a public API. Gets you attacker-mindset analysis without hiring a pen tester.
SQL injection audit
Check every SQL query in the codebase. Flag any that use string concatenation instead of parameterized queries.
When to use: Any project with raw SQL. Parameterized queries are the only reliable defense against SQLi.
Gitignore completeness check
Review our .gitignore. List any sensitive file types that are NOT ignored that should be.
When to use: Project setup, before first commit, or after adding new tools that create config files.
PR security checklist generator
Generate a security checklist for this PR. Cover: OWASP top 10, secrets exposure, rate limiting, auth checks.
When to use: Every PR that touches backend logic. Paste the checklist into the PR description.
Context Management
5Structured context gathering
Before we start: ask me one question at a time about the context I need you to know for this session. Stop when you have enough.
When to use: Start of any complex session. Prevents Claude from guessing and making wrong assumptions.
Session summary and context reset
Summarize what we've accomplished so far in this session in 5 bullet points. Then clear your context.
When to use: When your session is getting long and Claude starts losing track of earlier decisions.
Focused context compaction
/compact focus on the auth changes only. Forget everything else.
When to use: When context is bloated and you need Claude to focus on one specific area.
Spec-first session split
I want to build [X]. Interview me about the requirements, edge cases, and constraints. Write the full spec to SPEC.md. I'll start a fresh session to implement it.
When to use: Large features. Splitting spec and implementation into separate sessions keeps context clean.
CLAUDE.md knowledge extraction
What are the 3 most important things you've learned about this codebase that I should add to CLAUDE.md?
When to use: End of productive sessions. Captures institutional knowledge so future sessions start smarter.
Swarms & Agents
4Parallel client status dashboard
Spawn 5 agents in parallel. Each takes one client folder from [list]. Each reads Progress.md and Communication-Log.md. Returns: blockers + next action. Merge into one dashboard.
When to use: Weekly client review. Gets status on all clients simultaneously instead of reading files one by one.
Parallel security scan across routes
Spawn one agent per file in src/api/routes/. Each runs a security review. Output JSON: {file, issues:[]}. Collect all results and merge into security-report.md.When to use: Full codebase security audit. Parallelizes what would otherwise be a serial, hours-long review.
Writer-critic loop for quality
Create a writer-critic loop: Agent 1 implements [feature]. Agent 2 reviews it as a paranoid senior engineer. Agent 1 addresses the critique. Repeat until Critic has no critical issues.
When to use: High-stakes features where quality matters more than speed. Simulates a real code review cycle.
End-to-end feature build with parallel agents
I need [feature] built end to end. Spawn: Agent A for the database schema, Agent B for the API layer, Agent C for the tests. Coordinate and merge.
When to use: Building a complete feature fast. Each layer gets focused attention from a dedicated agent.
Debugging
4Slow-down diagnostic interview
The bug is [describe]. Slow down. Ask me one question at a time. Don't jump to solutions. First map every possible cause.
When to use: When you are stuck. Prevents Claude from jumping to the first solution and missing the real issue.
Five competing hypotheses
I can't reproduce this bug. Generate 5 competing hypotheses. For each: what evidence would confirm it, and where in the code to look.
When to use: Intermittent or hard-to-reproduce bugs. Gives you a structured investigation plan.
Non-determinism source finder
This test is failing intermittently. Find every possible source of non-determinism in this test. List them with likelihood.
When to use: Flaky tests. Common sources: time, random, network, shared state, async race conditions.
Function deep-dive with edge case testing
Explain this function like I've never seen it before. Then tell me what happens if [edge case]. Then show the test that would catch it.
When to use: Understanding unfamiliar code and immediately hardening it with tests.
Workflow & Automation
4Inbox processing automation
Every morning: read 00_Inbox/. Process each item: extract actions, route to right PARA folder, update relevant client files. Output: list of actions.
When to use: Daily inbox zero routine. Turns a 30-minute manual task into a 2-minute AI task.
Stale PR monitor
/loop 30m check all open GitHub PRs. Flag any open more than 48 hours with no activity. List with links and assignees.
When to use: Team management. Catches PRs that are silently stalling before they become blockers.
Pre-commit safety gate
Build a pre-commit hook that: 1) runs tests, 2) checks for secrets in changed files, 3) blocks commit if either fails.
When to use: Any production project. Prevents the two most common commit mistakes: broken tests and leaked secrets.
Deployment watchdog
Watch the deployment logs at [URL]. Message me on Slack if: response time > 2s, error rate > 1%, or any 5xx appears.
When to use: Post-deployment monitoring. Catches issues before users report them.
Content & Writing
4Voice-matched content generation
Write this content in my voice. Read @02_Areas/Marketing/Content_Strategy/walid-writing-style.md first. Then write [content]. No corporate language.
When to use: Any content creation. The @ reference loads your actual writing style so output matches your voice.
Raw notes to LinkedIn post
Turn these raw notes into a LinkedIn post. Direct, no fluff, specific. Under 1300 chars. Hook on line 1.
When to use: When you have ideas but not time to craft them. Converts rough thoughts into publish-ready posts.
Multi-angle content from one win
I have [result/win]. Write 5 different angles for LinkedIn posts about it. Each under 200 chars for the hook. Different formats.
When to use: Maximizing content from a single win. One result can fuel a week of different posts.
30-day content calendar
Create a content calendar for the next 30 days around [theme]. Each post: hook, angle, format, CTA. Save to content-calendar.md.
When to use: Monthly content planning. Gets you from zero to a full month of planned content in one session.
Business & Sales
5Client status deep-read
Read everything in 01_Projects/Clients/[client]/. Give me: current blockers, next 3 actions, anything I might be forgetting.
When to use: Before any client interaction. Surfaces forgotten tasks and keeps nothing from slipping through.
Client call preparation kit
I'm preparing for a call with [client]. Read their folder. Generate: 5 discovery questions, likely objections with responses, recommended next step.
When to use: 15 minutes before any client call. Shows up prepared without spending an hour re-reading files.
Discovery-to-proposal generator
Write a proposal for [client] based on the discovery notes in their folder. Use the 1-hour analysis framework. Include ROI calculation.
When to use: After a discovery call. Turns raw notes into a professional proposal in minutes.
Prospect ICP scoring and outreach
Research [company/person] as a prospect. Score them against our ICP. Find 3 specific pain points I can reference in outreach.
When to use: Before cold outreach. Personalized pain points convert 3-5x better than generic messages.
Win-to-case-study extractor
I just finished a project for [client]. Extract the win metrics from their Progress.md. Write a case study in: problem --> solution --> results format.
When to use: After every successful project. Case studies are the highest-converting sales asset.
Pro Tips
How to write your own power prompts
Be specific about role
Give Claude a persona with clear expertise and temperament. A paranoid security auditor reviews differently than a friendly code mentor.
State the output format
Tell Claude exactly what structure you want back. JSON, bullet list, markdown file, ASCII diagram. Never leave it ambiguous.
Add constraints
Constraints prevent Claude from going off track. "Ask one question at a time" and "Don't jump to solutions" are two of the most powerful.
Use the Interview Pattern
For complex tasks, don't dump requirements. Tell Claude to interview you. This surfaces edge cases you forgot and produces better specs.
Tell Claude what NOT to do
Negative constraints are as important as positive ones. Prevents over-engineering, fake enthusiasm, and vague suggestions.
Want this running in your stack?
AY Automate builds AI automation systems for production teams.