noyap

No yap. Ship the smallest correct change. Shut up about it.

Portable Agent Skills package: minimal narration + minimal correct code in one skill.

| Axis | What it controls | Similar to | |---|---|---| | Mouth | Near-zero narration, preambles, tours, closing offers | caveman | | Hands | YAGNI, soft LOC budgets, anti-golf, surgical diffs | ponytail | | Dial | silent (default) · brief · off + one-shot why? | overrides |

noyap does not disable model reasoning. It reduces chatter, scope, tool sprawl, LOC, and tokens. Correctness and safety always win.

Install

# One command (skills + always-on pin into the current project)
npx @harjjotsinghh/noyap --project .

# Full skill body as always-on (more tokens)
npx @harjjotsinghh/noyap --project . --full

# Or skills CLI + pin
npx skills add HarjjotSinghh/noyap
npx @harjjotsinghh/noyap --agents-only --project .

Claude Code:

/plugin marketplace add HarjjotSinghh/noyap
/plugin install noyap@noyap

Full matrix: INSTALL.md · Help: npx @harjjotsinghh/noyap --help

Modes

| Mode | Say | Behavior | |---|---|---| | silent (default) | noyap / just code | Diff + path line or Done. | | brief | noyap: brief | Silent + ≤2 short lines | | off | noyap: off / verbose | Suspended | | one-shot | why? | Full explanation of last change, then resume |

vs ponytail / caveman

| | caveman | ponytail | noyap | |---|---|---|---| | Shrinks prose | ✅ | — | ✅ | | Shrinks code / scope | — | ✅ | ✅ | | Soft LOC + anti-golf | — | partial | ✅ | | Speculative “for later” refuse | — | ladder | ✅ hard rule | | Multi-provider scaffold refuse | — | partial | ✅ hard rule | | silent / brief / off | levels | lite/full/ultra | ✅ | | Always-on auto-pin | varies | AGENTS.md | ✅ npx / installer | | Hooks / heavy plugin stack | heavy | heavy | none |

Use caveman+ponytail if you want those brands. Use noyap for one portable skill covering both halves.

More: docs/compare.md

Benchmarks

Primary (n=3, evals 1/7/10, Grok 4.5):
benchmarks/results/n3-20260811T055638Z/COMPARISON.md

| arm (median) | success | out_tok vs base | total_tok vs base | |---|--:|--:|--:| | noyap | 89% | −61% | −49% | | ponytail | 67% | −39%* | −10%* | | caveman | 44% | −22%* | +16%* |

*approx from medians vs baseline in that stamp. noyap alone hit 100% on multi-auth YAGNI (n=3).

Also: single-shot full suite 20260811T053548Z · real package sindresorhus/is real-20260811T055638Z

| Published elsewhere (different engines/repos — not interchangeable) | Source | |---|---| | Ponytail −54% LOC / −22% tokens (Claude Code · Haiku · FastAPI template) | ponytail agentic | | Caveman −65% chat output · −8.5% agentic (JetBrains) | caveman |

Do not treat our Grok % as equal to ponytail’s Claude %. Same metric names only.

Reproduce: benchmarks/README.md

License

MIT