Agent Skills

A collection of skills for AI coding agents — packaged instructions and scripts that extend agent capabilities, in the Agent Skills format.

Installation

# Install all skills
npx skills add morzecrew/agent-skills

# Install a specific skill
npx skills add morzecrew/agent-skills@keep-a-changelog

Skills are available to the agent once installed, and it invokes them when a relevant task shows up. Each one is a SKILL.md plus optional scripts/ and references/.

Skills cross-reference each other in their "Related skills" sections. Every skill works standalone, but when installing a subset, consider including the referenced siblings so those links resolve.

Available Skills

Writing code

Skill	What it does
readable-code	Names, nesting, and comments as one pass — a routing table over three references, and the rules where they constrain each other.
composition-over-inheritance	Composition and interfaces by default, with the fragile-base-class trap and the Liskov behavioral is-a test.
less-code-same-behavior	Divergence-and-DRY audit in behavior-preserving steps, where NO ACTION is a first-class verdict.
error-taxonomy	A closed set of error kinds, each with its transport mapping, message exposure, and retryability decided once.
escape-hatch-policy	Decides when an abstraction earns a raw/bypass hatch, and designs one that is named, scoped, and fail-closed.
determinism-by-design	Time, randomness, IDs, iteration order, and schedule behind seams — single-seed replay and hermetic tests.

Testing and debugging

Skill	What it does
reproduce-then-fix	Red repro, minimize, explain the mechanism, fix the cause, watch the same red turn green, keep it as a regression test.
fewer-tests-more-proof	Each promise tested exactly once with its strongest assertion; deletions proven by hand-run sabotage.
reading-isnt-proof	No closing a test gap by reading code — implementations of one contract need a shared conformance battery, run.
failure-path-review	Sweeps the unhappy paths in async and background systems: poison, redelivery, drain-not-abandon, bounded growth.
measure-before-optimizing	Profile before optimizing, benchmark honestly (warmup, noise, percentiles), and know which design calls precede measurement.

Review and audit

Skill	What it does
self-audit	Adversarial pass over your own finished work, walking the ten places author blind spots concentrate.
flag-dont-flip	Executes a spec without silently changing it — plan gate first, halt on LOCKED decisions, every departure logged and checked.
pr-review-loop	Works AI and human PR feedback to convergence — deduped findings, evidence-backed verdicts, one push per round.
ratchet-what-you-build	Ranks every guard on an enforcement ladder; anything still sitting at "convention" is an open finding.
drift-to-gate	Turns a rule that keeps being broken into a program that refuses — then runs that control's whole lifecycle.
negative-result-taxonomy	Classifies every failed attempt as family-dead, design-dead, or undecidable, so a kill is a diagnosis and not a dead end.
authority-dissociation	Separates who does the work from who certifies it, so no actor writes the evidence that judges its own output.
decide-before-you-look	Pre-registers the call before the data exists — six lines, an interval that must be narrower than the decision band, and no moving goalposts.
distill-the-rule	Turns hard-won findings into one-line transferable rules, filed where future work will meet them.

Documentation

Skill	What it does
altitude-docs	Writes and polishes docs as a controlled descent through altitude bands, with a Diátaxis page contract and a ship rubric.
rfc-writer	Lightweight numbered RFCs in rfcs/, tracked by one INDEX.md, with append-only decision records.
keep-a-changelog	Keeps CHANGELOG.md in Keep a Changelog 1.1.0 format — user-focused entries, accurate Unreleased, clean version cuts.
python-google-docstrings	Google-style Args:/Returns:/Raises: docstrings that Napoleon renders and IDE tooltips scan.
python-rest-docstrings	reST docstrings — Sphinx-native :param:/:returns:/:raises: fields and cross-reference roles that really link.

Project hygiene

Skill	What it does
gitmoji-conventional	Commits and PR titles as <gitmoji> <type>[scope][!]: <description>, with all 75 official gitmojis mapped deterministically.
dependency-diligence	Evaluates a dependency before adoption — constraint test, capability-per-cost, health — ending in one recorded verdict.

Contributing

See AGENTS.md for repository layout, how to add a skill, and how to run validation and tests.