CheatCoders

Claude Code Hooks: Gate Risky Shell Commands Before CI Runs

Wire Claude Code hooks that block destructive git and cloud CLI calls, require human approval for IAM mutations, redact secrets from tool logs, and keep a durable audit trail of every agent shell invocation.

AI AWS

Bedrock Prompt Management: Versioned System Prompts for Production

Treat Bedrock system prompts like versioned code: publish through Prompt Management, canary with aliases, compare quality and latency, and roll back instantly when a prompt change regresses assistants.

AI AWS

Cursor Agent Mode: Multi-File Refactors That Survive CI Gates

Senior patterns for Cursor agent mode across TypeScript monorepos: constrain multi-file edits with project rules, verify with targeted CI jobs, quarantine partial refactors, and roll back before anything lands on main.

Multi-Model Routing: Cheap Draft Models Plus Expensive Verifiers

Route trivial edits to small cheap models and reserve frontier models for architecture review, security diffs, and flaky-test root-cause analysis—cut spend without sacrificing judgments that protect production.

AI AWS

AI Coding in VPC: Private Bedrock Endpoints and Secret Hygiene

Run coding agents over VPC endpoints to Bedrock with no public egress, CMK-encrypted prompt storage, redacted CloudTrail payloads, and private ECR pulls so regulated code never leaves your network.

AI AWS Machine Learning

RAG Evaluation on AWS: Hit Rate, Faithfulness, and Cost Curves

Build an offline evaluation harness on S3 and Batch for code RAG that measures retrieval hit rate, answer faithfulness, and per-query Bedrock spend before promoting prompt changes.

AI AWS

Bedrock Converse API: Tool Choice Modes for Production Coding Agents

Pick Bedrock Converse tool-choice modes—auto, any, or specific—so coding agents stop chatting when they should call apply_patch, run_tests, or search_repo in automated PR workflows. Production checklist included.

Structured Outputs for Codegen: JSON Schemas That Actually Compile

Force LLM codegen into typed JSON that maps cleanly to ASTs and templates, reject invalid shapes before any filesystem write, and keep mechanical transforms reproducible across CI and local agents.

AI AWS

Cursor Rules for AWS CDK: Stop AI From Inventing IAM Wildcards

Encode least-privilege AWS CDK patterns as Cursor rules so generated stacks never ship Resource or Action wildcards, open security groups, or public buckets into production by accident.

AI AWS Debugging

LLM Incident Runbooks: Ground On-Call Answers in CloudWatch Signals

Ground on-call LLM assistants in CloudWatch Live Tail, Logs Insights queries, and X-Ray traces so suggested remediations cite real production signals instead of generic runbook guesswork during sev-1s.

SSRF Hardening in Node: Block Metadata and DNS Rebinding on Fetches

Distributed Deadlock Detection: Cycles Across Postgres Rows and Redis Locks

CloudWatch Logs Anomaly Detection: Ignore Deploy Noise, Catch Novel Errors

Node Core Dump Analysis: Symbolicate Native Crashes Inside Containers

API Gateway WebSocket Fan-Out: Coalesce Pushes After Reconnect Storms

Feature Stores vs Inline Compute: Real-Time Decisions Without Training Skew

X-Ray Service Maps Lie: Fix Missing Spans in Node Trace Edges

Exactly-Once Illusions: Design At-Least-Once Plus Truly Idempotent Handlers

SLA Error Budgets: Drive Automated Load Shedding Before Hard Outages

Consistent Hashing for WebSockets: Sticky Gateways During Scale Events

CheatCoders

Claude Code Hooks: Gate Risky Shell Commands Before CI Runs

Bedrock Prompt Management: Versioned System Prompts for Production

Cursor Agent Mode: Multi-File Refactors That Survive CI Gates

Multi-Model Routing: Cheap Draft Models Plus Expensive Verifiers

AI Coding in VPC: Private Bedrock Endpoints and Secret Hygiene

RAG Evaluation on AWS: Hit Rate, Faithfulness, and Cost Curves

Bedrock Converse API: Tool Choice Modes for Production Coding Agents

Structured Outputs for Codegen: JSON Schemas That Actually Compile

Cursor Rules for AWS CDK: Stop AI From Inventing IAM Wildcards

LLM Incident Runbooks: Ground On-Call Answers in CloudWatch Signals

SSRF Hardening in Node: Block Metadata and DNS Rebinding on Fetches

Distributed Deadlock Detection: Cycles Across Postgres Rows and Redis Locks

CloudWatch Logs Anomaly Detection: Ignore Deploy Noise, Catch Novel Errors

Node Core Dump Analysis: Symbolicate Native Crashes Inside Containers