README

A curated repository of high-signal skills for coding agents. The catalog favors behavior-changing engineering guidance over framework encyclopedias, version snapshots, and material a capable model or current owner documentation already provides.

Skills follow the Agent Skills specification and can be installed by clients supported by the skills CLI.

See CHANGELOG.md for recent additions, removals, and ownership decisions.

Installation

Install the catalog:

npx skills add voku/agent-skills

Install one skill:

npx skills add voku/agent-skills --skill operational-prompting
npx skills add voku/agent-skills --skill coding-simplicity
npx skills add voku/agent-skills --skill php-static-analysis
npx skills add voku/agent-skills --skill linux-strace

Install globally or for a specific agent:

npx skills add voku/agent-skills -g
npx skills add voku/agent-skills -a claude-code

List the current catalog instead of relying on a manually duplicated README inventory:

npx skills add voku/agent-skills --list

Manual installation remains ordinary Git:

git clone https://github.com/voku/agent-skills.git
cp -r agent-skills/skills/* .claude/skills/

Composer (optional PHP distribution)

After publishing this repository on Packagist and tagging a release, PHP projects can install the same portable catalog without a separate Git checkout:

composer require --dev voku/agent-skills

For voku/agent-loop, configure the existing .agent-loop/init.json skill root:

{
  "extra_skills_roots": ["vendor/voku/agent-skills/skills"]
}

Then run vendor/bin/agent-loop init install-assets --agent=all --dry-run and apply the projection only after reviewing the proposed changes. Composer only distributes the catalog; it does not execute or activate skills.

voku engineering skills

`operational-prompting`

Reusable, tool-neutral operational prompting for coding agents. It covers repo-owned instructions, instruction hierarchy, bounded task contracts, validation/evidence contracts, machine-readable workflows, stopping conditions, and portable skill manifests.

The canonical skill deliberately does not own the governed voku/agent-* operating-prompt recipe catalog. A machine-readable recipe catalog whose correctness depends on one compiler belongs with that compiler. For the current workflow, voku/agent-recall-compiler owns recipe definitions, typed arguments, rendering/template identity, and recipe applicability metadata; voku/agent-loop owns lifecycle and execution authority.

A project-specific execution contract still benefits from the familiar five concerns:

Goal
Context
Constraints
Verification
Done When

Verification defines how reality is measured. Done When defines the observed result that permits the task to stop. The skill teaches those reusable principles without duplicating Recall-owned recipes or Loop-owned lifecycle policy.

`coding-simplicity`

Implementation-time simplicity without deleting correctness. Before adding new surface area, search in this order:

no code change;
existing repository owner/pattern;
language standard library;
native platform capability;
already-installed dependency;
one shared root-cause fix;
minimum new code.

The skill keeps explicit safety and verification floors. A shorter patch that moves behavior into the wrong layer, drops a trust-boundary check, or leaves sibling callers broken is not simpler.

`engineering-codelight`

Technology-neutral reasoning for non-trivial engineering work. It keeps evidence separate from authority, preserves uncertainty, checks instruction provenance and owner boundaries, favors falsifiable evidence and small causal changes, re-grounds stale plans, keeps failure recoverable, and turns durable lessons into structure.

It is deliberately workflow-neutral: an existing project workflow remains the authority for lifecycle state and approvals. engineering-codelight complements the concrete implementation search in coding-simplicity; it does not replace it or any language- or review-specific skill.

`code-review-*`

The review skills are independent targeted lenses, not a mandatory review swarm. Start with the dominant concern and keep the review evidence-first. If another concern becomes primary, hand off to at most one smaller follow-up lens.

Available lenses include architecture, error handling, performance, security, simplicity, and type safety.

`php-static-analysis`

Focused PHP static-analysis guidance for preserving strict contracts, precise array shapes and generics, root-cause typing, and scoped analyzer exceptions. It deliberately avoids duplicating general PHP syntax and framework knowledge.

Full catalog

The repository intentionally keeps a small catalog under skills/. A skill should earn its context cost by changing agent behavior: evidence discipline, review focus, scoped execution, static-analysis rigor, runtime diagnosis, or another reusable reasoning constraint.

Framework and library cookbooks do not belong here when current owner documentation, package-aware tooling, or the model's baseline knowledge is the better source. In particular, version-sensitive Laravel, React, Tailwind, Playwright, state-management, SEO, and similar reference material should be looked up from its semantic owner when needed.

Use the repository itself as the source of truth:

npx skills add voku/agent-skills --list

Each skill's own SKILL.md, rules, metadata, and supporting files define its actual contract. README summaries are intentionally not a second hand-maintained manifest.

Using with agent-loop

voku/agent-loop can install this catalog as an explicit extra skill root for portable engineering guidance. Governed workflow prompt recipes are not sourced from this catalog: voku/agent-recall-compiler owns and releases that catalog, while voku/agent-loop owns durable Contract/Run orchestration and the gates that may authorize execution.

The intended ownership shape is:

tool-neutral engineering principles  -> voku/agent-skills
bounded project context + recipes     -> voku/agent-recall-compiler
Contract / Run / lifecycle authority  -> voku/agent-loop

Working-memory packages may carry resumable Session state, but they do not become the durable authority for approved task policy. Likewise, installing a portable skill does not grant workflow or mutation authority.

Changelog

See CHANGELOG.md for voku-specific catalog changes.

Provenance

This repository is derived from the public AsyrafHussin/agent-skills catalog but retains upstream material only where it still earns its context cost. voku-specific changes focus on evidence-driven engineering, operational prompting, simplicity, review routing, static analysis, runtime diagnosis, and integration with the voku/agent-* packages.

License

MIT

voku / agent-skills

Package info

Security

Fund package maintenance!

Statistics