production-awareness

Production Awareness

What This Skill Enables

An agent that considers how code will behave in production - under real load, with real data, facing real failures - while writing it, not after shipping it. Without this skill, agents write code that works in development but fails under production conditions: high concurrency, network failures, malformed data, resource exhaustion, and hostile input.

Core Competencies

1. Production Mental Model

While writing any code, ask these questions:

Question	Why It Matters
What happens at 100x current load?	Linear algorithms become bottlenecks; database connections exhaust
What happens when this external call fails?	Network timeouts, retries, and circuit breakers become essential
What happens with unexpected input?	User-generated content includes Unicode, injection attempts, and extreme lengths
What happens when disk/memory fills up?	Unbounded logs, caches, and queues cause cascading failures
What happens during a deploy?	In-flight requests, database migrations, and cache invalidation need handling

2. Failure Mode Awareness

Every external dependency will eventually fail. Design for it:

Network calls: Always have a timeout, retry policy, and fallback behavior

GOOD: fetch(url, { signal: AbortSignal.timeout(5000) })
       .catch(err => fallbackResponse)

BAD:  fetch(url)  // hangs forever if the server doesn't respond

Database queries: Handle connection pool exhaustion, slow queries, and deadlocks

• Set query timeouts
• Use connection pool limits with appropriate max/min
• Log slow queries above a threshold

File system: Handle disk full, permission denied, and race conditions

• Check available space before large writes
• Use atomic write patterns (write to temp, rename)
• Handle ENOENT and EACCES explicitly

3. Data Safety

Treat production data with care at every level:

• Never log sensitive data: passwords, tokens, PII, credit card numbers
• Never trust input size: enforce limits on request bodies, file uploads, query parameters
• Never expose internals: stack traces, database schemas, internal paths in error responses
• Always validate at boundaries: data from users, APIs, message queues, and even your own database (data can be corrupted)