Commenting Intent

Commenting Intent

Overview

Good comments explain WHY, not WHAT. Code already shows what it does. Comments should explain intent, decisions, and non-obvious reasoning.

Core principle: If the comment just restates the code, delete the comment or improve the code.

Baseline Violation

Agents comment mechanics (what code does):

❌ Over-commenting (baseline):

# Calculate subtotal by multiplying price and quantity for each item, then summing
subtotal = sum(item.price * item.quantity for item in items)

# Calculate tax amount based on the subtotal and tax rate
tax = subtotal * tax_rate

# Calculate final total by adding subtotal and tax
total = subtotal + tax

Problem: Comments just restate obvious code. No value added.

✅ Comment intent only:

def calculate_total_price(items, tax_rate):
    """Calculate order total including tax."""
    subtotal = sum(item.price * item.quantity for item in items)
    tax = subtotal * tax_rate
    return subtotal + tax  # No comments needed - code is clear

What to Comment

1. Non-Obvious Decisions

✅ WHY you chose this approach:

def is_rate_limited(user_id, redis_client):
    # Using Redis instead of in-memory to support distributed rate limiting
    # across multiple app servers. Limit: 100 req/min per business requirements.
    key = f"rate_limit:{user_id}"
    current = redis_client.get(key)

    if current is None:
        redis_client.setex(key, 60, 1)  # 60 sec TTL
        return False

    return int(current) >= 100  # Limit per product requirements doc

Explains: WHY Redis, WHY these limits, WHERE requirements came from.

2. Algorithms and Complex Logic

✅ WHY this algorithm:

def find_user(users, email):
    # Binary search requires sorted array. We sort by email on load
    # to enable O(log n) lookups. Worth the upfront sort cost because
    # lookups happen 100x more frequently than updates.
    return binary_search(users, email)

Explains: WHY binary search, WHY pre-sorted, trade-off reasoning.

3. Magic Numbers and Constants

✅ WHY these values:

MAX_RETRIES = 3  # Testing showed 3 retries handles 99.9% of transient failures
TIMEOUT_MS = 5000  # API SLA guarantees < 5sec response time
BATCH_SIZE = 100  # Larger batches caused memory issues in prod (incident #1234)

Explains: WHERE values came from (testing, SLA, incident).

4. Workarounds and Gotchas

✅ WHY unusual code:

# WORKAROUND: Library bug #456 - must call reset() twice
# Fixed in v2.0 but we're on v1.8
client.reset()
client.reset()

Warns future maintainer: Unusual code has reason, link to issue.

What NOT to Comment

Don't Comment Obvious Code

❌ Restates the code:

# Set user name to "John"
user.name = "John"

# Loop through items
for item in items:
    # Process the item
    process(item)

✅ Let code speak:

user.name = "John"

for item in items:
    process(item)

Don't Comment Mechanics of Standard Patterns

❌ Obvious pattern:

# Initialize search boundaries to cover entire array
left, right = 0, len(arr) - 1

# Continue searching while there's a valid range
while left <= right:

✅ Comment intent only:

def binary_search(arr, target):
    # Binary search for O(log n) performance on sorted array
    left, right = 0, len(arr) - 1
    while left <= right:
        # ... implementation (standard pattern, no comments needed)

The Comment Test

For each comment, ask:

1. Does it explain WHY, not WHAT?

   • WHY: "Redis for distributed limiting" ✅
   • WHAT: "Get value from Redis" ❌

2. Would I understand without it?

   • If yes: Delete comment
   • If no: Keep comment OR improve code clarity

3. Does it add information beyond the code?

   • Yes: Keep it
   • No: Delete it

Quick Reference

Real-World Impact

From baseline:

• Agents commented every line of simple price calculation (over-commenting)
• Comments restated code without adding value
• Missed opportunities to explain WHY (decisions, trade-offs)

With this skill: Comment intent, not mechanics.

Integration with Other Skills

For evergreen comments: See skills/writing-evergreen-comments - no temporal context in comments

For self-documenting code: See skills/naming-variables - good names reduce need for comments