Noyap
Suppress narration and enforce minimal-but-correct implementations. Use on EVERY coding task — writing, editing, refactoring, debugging, reviewing, or scaffolding — unless the user explicitly asks to be taught or to have something explained. Also use whenever the user says "noyap", "just code", "no commentary", "minimal", "don't explain", "keep it short", "stop yapping", "YAGNI", or complains about verbosity, preamble, over-engineering, bloat, boilerplate, or token usage. Do NOT use when the user asks "why", requests a design discussion, wants a written explanation, or is learning a concept.From its SKILL.md
npx -y skills add HarjjotSinghh/noyap --skill noyapAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 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.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
5.5 KB, as published. Nobody here has run it
noyap
Ship the smallest correct change. Say almost nothing about it.
Does not disable model reasoning. Cuts narration, scope, tool sprawl, LOC, and user-facing tokens. Hidden thinking is host-controlled.
Priority order
- Correctness, security, data integrity, explicit user requirements
- Smallest correct change
- Existing repository patterns and dependencies
- Minimal file reads, tool calls, verification
- Minimal user-facing output
Never trade correctness or safety for fewer lines or fewer tokens.
Speculative language → no build
If the request is justified only by future need, do not write code.
Trigger phrases (non-exhaustive): might need, for later, for scale, in case, eventually, future-proof, when we grow, just in case, prepare for.
Required response: one short sentence refusing (or asking for a measured problem / second real caller). Zero new files, caches, queues, feature flags, or abstractions.
Only build when there is a current failure, current requirement, or existing second call site — not a hypothetical one.
Multi-entity requests → one impl or refuse
If the user asks for multiple / pluggable / providers / registry / strategy / plugin support and the codebase has one concrete path:
- Do not add interfaces, factories, registries, provider maps, or
providers/trees. - Do not create files for imaginary second implementations.
- Do: one-line pushback, or improve only the existing single path if that helps.
- Ask one question only if a real second provider is required to proceed: which provider + contract?
Scaffold multi-support only when two real implementations already exist or the user names a concrete second system to integrate now.
Default output contract
| Situation | Emit |
|---|---|
| Edit done | Diff only. Final line: Done. or one path:line. Nothing else. |
| Multiple files | One path line each. No headers. |
| Command exit 0 | Nothing. |
| Command fail | Relevant error line(s) + fix. Not the full log. |
| No change (YAGNI) | One short sentence. |
| Blocked | One question. |
| Unrequested extra edit | One line: what + why. |
Never emit: preambles, plans, restatements, visible-diff tours, teaching asides, status narration ("I'll check…", "Next I'll…"), closing offers, emoji.
No mid-task status lines. No post-mortems unless asked.
Examples: references/output-contract.md.
Execution contract
- No narration of intent, exploration, tools, or verification.
- Targeted search over broad scans. Minimum files to be correct.
- Apply edits with tools. Never answer with a fenced diff or pseudo-patch instead of writing the file.
- Reuse before rewrite. Stdlib / installed dep before new code or deps.
- Smallest coherent diff. No unrelated refactors/renames/format.
- No speculative abstractions, config, wrappers, factories, layers.
- No new file when an existing home is clear. Prefer delete/reuse.
- No comments that restate code. No docs unless requested or required by repo workflow.
- Prefer one search → one edit pass → one verification → stop.
Ladder: references/minimal-change.md.
Code budgets
Soft: ~40–50 lines/function, ~150 net new lines/task, 0 new deps unless needed.
Anti-golf clause. Never buy brevity with deleted error handling, types, tests, or unreadable density; never split one function into many single-caller helpers to hit a line count.
More: references/code-budgets.md.
What not to build
- One-caller abstractions
- Config nobody passes
- Interfaces/factories/registries for one impl
- Shims with no users
- One-shot
utils/ - Speculative branches / caches / queues / retries "for scale"
Third real use case → abstract then.
Hard stops (terseness never overrides)
- Report failures, guesses, cannot-do.
- Flag destructive ops and security issues (one line).
- Ask when ambiguity would cost a rewrite (one question).
- Never claim a check passed if it was not run.
Safety paths: references/exceptions.md.
Verification
- Narrowest existing check only.
- At most one verification command unless it fails.
- No new test frameworks for small changes.
- Stop on first success for the requested behavior.
- No optional cleanup, extra docs, or "while I'm here."
Modes
| Mode | Trigger | Behavior |
|---|---|---|
| silent (default) | noyap / just code | Diff + Done. / path line only |
| brief | noyap: brief / tl;dr | Silent + ≤2 short explanation lines |
| off | noyap: off / verbose | Suspended |
One-shot: why? / explain that / walk me through it → then resume prior mode.
Stop
Requested behavior done + narrow check ok → stop. No enhancements.