Essential comments

Skill yakoshiq/skills/skills/essential-comments

Add, keep, or remove code comments so only remarks that help humans remain - why, invariants, tradeoffs, and external constraints - not narration or restated code. Use when cleaning noisy comments, writing new code, reviewing comment quality, or when the user wants fewer/better comments.From its SKILL.md

Install

npx -y skills add yakoshiq/skills --skill essential-comments

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

22 days oldThe repository was created 22 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.0 KB, 380 tokens by cl100k_base, as published. Nobody here has run it

Essential Comments

A comment earns its place only if a human would lose real context without it.

Match the repo's comment/doc idioms. Never require JSDoc/docstrings or invent a doc system.

Choose scope

Review: findings only. Noise to delete, missing whys to add; cite symbol and human risk.
Edit: comments only. Same behavior, structure, names, and APIs.

Keep or add

Only when names and code do not already say it:

Why / policy - intentional tradeoff, partial success, deliberate swallow
Invariant / ordering - concurrency, sequence, non-obvious control flow
External / non-local - vendor quirk, ticket, runbook, magic value whose meaning lives elsewhere

When tightening a good comment, preserve concrete facts (thresholds, ticket IDs, runbook names).

After you write or edit logic, scan once for why / invariant / external facts you actually know (task, ticket, surrounding code) that a later reader cannot recover from names alone. If you know one, add one short comment. Do not guess. If unsure, omit.

Remove

Narration or English restatement of the next line, name, type, or clear flag
Docs that only echo identifiers and parameters
AI / changelog residue and commented-out code (unless the user wants it kept)
Meta banners; policies or caller roles not evidenced in visible code
Duplicate decoding of the same magic (at most one short legend at the definition)

Discipline

Why, not what. If unsure, omit.
A comment papering over a bad name is a naming issue: say so in review; in comments-only edit, do not rename and do not keep pure narration as a stand-in.
Ship the diff or findings - no essay.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.