| name | code-comments |
| description | Write high-signal code comments for humans and coding agents. Use when adding inline comments, docstrings, API comments, or local rationale near code. Triggers on: "write better comments", "comment this", "document why", "agent-friendly comments", "explain this code with comments", "docstring". |
Code Comments
Write code so the structure explains the what. Write comments so future humans and agents understand the why, why not, what must stay true, and what already happened here.
Core Rule
Use this order:
1. Rename, extract, reorder, or delete code until the intent is obvious
2. Add a comment only when the missing context cannot live in code
3. Put the shortest useful comment as close as possible to the code
4. Keep the comment short, local, and durable
Good comments reduce wrong edits. Great comments stop humans and agents from repeating old mistakes.
The Comment Test
Before writing a comment, ask:
Can clearer code remove the need for this comment?
Yes -> refactor first
No -> comment the missing context
Will this comment still be true after a small refactor?
No -> move the detail into code or delete it
Yes -> keep it
Does this comment tell the reader something the code cannot?
No -> delete it
Yes -> keep it
What To Comment
Comments earn their keep when they capture one of these:
Intent
Why this code exists.
Constraint
A rule the code must respect because of product, legal, protocol, or platform limits.
Invariant
A property that must remain true across future changes.
Tradeoff
Why a less-obvious implementation beat the simpler-looking one.
History
What already happened that future editors should know.
Warning
What not to "simplify" and why.
Reference
Where the deeper story lives.
What Not To Comment
Do not comment things that should live in code.
Bad:
retryCount += 1
const user = await getUserById(userId)
Better:
retryCount += 1
const user = await getUserById(userId)
Bad:
const req = buildPaymentRequest(order)
Better:
const paymentRequest = buildPaymentRequest(order)
If a comment only translates weak names into better English, fix the names.
Write For Three Audiences
Every high-signal comment should help all three:
- Your future self scanning the file at speed
- A teammate without the original decision context
- A coding agent proposing edits from local evidence only
That means:
- Prefer explicit nouns over vague pronouns
- Name the thing that would break
- State the consequence, not just the preference
- Use issue or incident ids when they exist
- Keep comments local and durable
Inline Comment Templates
Use these templates directly. Replace brackets with concrete facts.
Decision Comment
Example:
Invariant Comment
Example:
Compatibility Comment
Example:
Incident Breadcrumb
Example:
Temporary Work Comment
Example:
Non-Obvious Example Comment
Example:
Docstrings And API Comments
Docstrings should describe contract, side effects, and sharp edges. Do not restate the implementation.
Bad:
def sync_users():
"""Sync users from the API."""
Better:
def sync_users():
"""Pull active users from the billing API and upsert local records.
Side effects:
- writes `users` and `subscriptions`
- emits `user_synced` events
Safe to retry. Not safe to run concurrently for the same account.
"""
Prefer this structure when the boundary matters:
What it guarantees
What it mutates or emits
What callers must provide
When it is unsafe or expensive
Anti-Patterns
Narration
for (const item of items) {
The code already says this.
Name Translation
const eml = user.email
Rename eml or delete the comment.
Vague Intent
Name the edge case and consequence.
Fake Temporariness
This never gets removed. Say when, after what, and by whom if needed.
Ghost History
Which bug? Under what condition? What breaks if removed?
Comment Drift
return users.sort((a, b) => b.createdAt - a.createdAt)
Stale comments are worse than no comments.
Essay Comments
Do not bury the point in five lines of setup. Lead with the rule or decision, then add one sentence of context if needed.
Review Checklist
Before keeping a comment, check all of these:
[] Does the code already say this?
[] Does the comment explain why, constraint, invariant, tradeoff, history, or warning?
[] Is the comment specific about what breaks or matters?
[] Will the comment likely survive a routine refactor?
[] Should any extra detail be cut because the local comment is already enough?
[] Did I include a reference if the decision came from an issue, incident, or RFC?
[] Would this comment stop a smart agent from making the wrong cleanup?
If the answer to the last question is no, the comment may not be pulling its weight.
Editing Workflow
When modifying code:
1. Delete comments made obsolete by your change
2. Rewrite comments whose scope changed
3. Add a short decision comment if the new code looks "weird" for a reason
4. Leave the file with fewer, better comments than you found it
Default Style
- Use short sentences
- Lead with the decision or warning
- Name concrete systems, fields, incidents, or documents
- Prefer
Do not ... because ... over soft phrasing
- Prefer one strong comment over three weak ones
- Avoid jokes, filler, and private context nobody else can recover
The goal is not more comments. The goal is better evidence for future readers and future agents.