Agent skills and plugins

In one paragraph

Agent skills and plugins are installable packs that tell a coding agent how to do a job it would otherwise do generically. Some set a whole working method, some carry domain taste in design, writing or business, and some give the agent questions to think with. A few plugins add tools and integrations. You install the pack, and the agent follows it.

What counts here: installable skill packs, plugins and working methods that extend what a coding agent knows how to do.

The main approaches

Skill collections

A bundle of separate skills, hooks or plugins that you pick from. Nothing forces an order. skills keeps each skill small and composable (grilling sessions, TDD, bug diagnosis, spec writing). claude-code-playground collects hooks, CLAUDE.md files, settings, mods and skills, each in a self-contained folder with a README. hermes-official-plugins ships integrations such as Home Assistant, Spotify and Langfuse as separately versioned plugins for Hermes Agent.

Domain skills

Skills that carry expertise in one field so the agent stops producing generic output. designer-skills covers UX research through handoff. skills focuses on animation and motion taste. tastemaker grounds UI style in reference images and runnable scripts. diagram-design draws diagrams as HTML and SVG. skills turns a business book into ten commands. compound-writing covers prose, and iso-24495 applies plain language rules to agent output.

Development methodologies

Packs that prescribe how software gets built, not just what the agent knows. superpowers runs brainstorm, plan, subagent execution, review and TDD in sequence. ponytail is a single prompt that makes the agent reuse existing code and write the minimum. claude-kit plans a task, hands parts to tiered subagents and shows their progress in the UI.

Structured reasoning lenses

Tools that give the agent or the user prompts for thinking. ideonomy-skill and ideonomy-engine are based on Patrick Gunkel’s ideonomy. ideonomy-skill is a Claude Code skill that draws a random mix of operators, organons and dimension prompts. ideonomy-engine is a CLI and library with 28 lenses that outputs questions, not answers. ideonomy-legacy is an older set of Python scripts that call the OpenAI API for idea generation and evaluation.

Map of the theme

flowchart LR
  t["Agent skills and plugins"]
  t --> f1["Skill collections"]
  t --> f2["Domain skills"]
  t --> f3["Development methodologies"]
  t --> f4["Structured reasoning lenses"]
  f1 --> r1["mattpocock/skills"]
  f1 --> r2["anthropics/claude-code-playground"]
  f1 --> r3["NousResearch/hermes-official-plugins"]
  f2 --> r4["Owl-Listener/designer-skills"]
  f2 --> r5["emilkowalski/skills"]
  f2 --> r6["slavingia/skills"]
  f2 --> r7["cathrynlavery/diagram-design"]
  f2 --> r8["EveryInc/compound-writing"]
  f2 --> r9["codeswithroh/tastemaker"]
  f2 --> r10["GaZmagik/iso-24495"]
  f3 --> r11["obra/superpowers"]
  f3 --> r12["DietrichGebert/ponytail"]
  f3 --> r13["JohnnyVizz/claude-kit"]
  f4 --> r14["kindgracekind/ideonomy-legacy"]
  f4 --> r15["Morpheis/ideonomy-engine"]
  f4 --> r16["latentwill/ideonomy-skill"]

Where the new ideas are

Process as a skill chain: superpowers fires skills in order and reviews each subagent task for spec compliance, then code quality.
Restraint as a skill: ponytail asks whether the code needs to exist before writing any, and ends each reply with what was skipped.
Skills backed by scripts and state: tastemaker uses contrast and palette scripts and local lock files. iso-24495 pairs skills with a rule engine that audits Markdown.
Separating voice from standards: compound-writing keeps VOICE.md and STYLE.md apart and routes work through a front-door skill.
Checks on output: diagram-design places nodes on a 4px grid and runs CI overlap checks.
Plugins versioned alone: hermes-official-plugins installs each integration at a reviewed catalog pin, so a fix does not wait for an agent release.
Questions instead of answers: ideonomy-engine chains lenses and stores sessions as JSON.

Side by side

Repo	Approach	Installs into	What it ships	Language
kindgracekind/ideonomy-legacy	Structured reasoning lenses	Not stated	Python scripts calling the OpenAI chat API, plus text lists	Python
obra/superpowers	Development methodologies	Coding agent, as a plugin	Brainstorming, planning, subagent execution and TDD skills	Shell
mattpocock/skills	Skill collections	Claude Code, Codex and GitHub Copilot as a plugin, Gemini CLI by hand, other agents via npx	Engineering skills such as /tdd, /grill-with-docs and /diagnosing-bugs	Shell
Morpheis/ideonomy-engine	Structured reasoning lenses	CLI and TypeScript library	28 lenses with reason, drill, chain, synthesize and session operations	TypeScript
Owl-Listener/designer-skills	Domain skills	Claude Code and Gemini CLI	111 skills and 34 commands across 9 plugins	Markdown
emilkowalski/skills	Domain skills	Via `npx skills@latest add emilkowalski/skills`	Markdown skills for animation and UI design	Markdown
slavingia/skills	Domain skills	Claude Code	Ten slash commands based on The Minimalist Entrepreneur	Not stated
cathrynlavery/diagram-design	Domain skills	Agent Skills-compatible hosts such as Cursor, Claude Code and Codex	45 diagram types as self-contained HTML with inline SVG	HTML
latentwill/ideonomy-skill	Structured reasoning lenses	Claude Code	Two skill variants (plain and rich) with a local Bash picker	Shell
EveryInc/compound-writing	Domain skills	Claude Code or Codex, as a plugin	Writing skills under the `cw-` prefix	Python
DietrichGebert/ponytail	Development methodologies	Claude Code and Codex as a plugin, other agents via AGENTS.md or SKILL.md	One prompt plus /ponytail, /ponytail-review and /ponytail-audit commands	JavaScript
codeswithroh/tastemaker	Domain skills	Claude Code, Windsurf or Gemini CLI	A skill with palette, contrast and anti-slop Python scripts	Python
GaZmagik/iso-24495	Domain skills	Claude Code and Codex CLI, as a plugin	Seven skills and a rule engine for auditing text (needs Bun)	TypeScript
anthropics/claude-code-playground	Skill collections	Not stated (entries are run from their own folders)	Hooks, CLAUDE.md files, settings, mods, skills and demos	JavaScript
JohnnyVizz/claude-kit	Development methodologies	Claude Code, as two plugins	savvy-flow task runner and savvy-progress UI mod	TypeScript
NousResearch/hermes-official-plugins	Skill collections	Hermes Agent	Twelve official plugins, including Home Assistant, Spotify and Langfuse	Python

How the idea moved

flowchart LR
  n1["started Jul 2025<br/>kindgracekind/ideonomy-legacy"]
  n2["started Oct 2025<br/>obra/superpowers"]
  n3["started Feb 2026<br/>mattpocock/skills"]
  n4["started Feb 2026<br/>Morpheis/ideonomy-engine"]
  n5["started Mar 2026<br/>Owl-Listener/designer-skills"]
  n6["started Mar 2026<br/>emilkowalski/skills"]
  n7["started Mar 2026<br/>slavingia/skills"]
  n8["started Apr 2026<br/>cathrynlavery/diagram-design"]
  n9["started May 2026<br/>latentwill/ideonomy-skill"]
  n10["started Jun 2026<br/>EveryInc/compound-writing"]
  n11["started Jun 2026<br/>DietrichGebert/ponytail"]
  n12["started Jul 2026<br/>codeswithroh/tastemaker"]
  n13["started Aug 2026<br/>GaZmagik/iso-24495"]
  n14["started Sep 2026<br/>anthropics/claude-code-playground"]
  n15["started Oct 2026<br/>JohnnyVizz/claude-kit"]
  n16["started Oct 2026<br/>NousResearch/hermes-official-plugins"]
  n1 --> n2 --> n3 --> n4 --> n5 --> n6 --> n7 --> n8 --> n9 --> n10 --> n11 --> n12 --> n13 --> n14 --> n15 --> n16

started Jul 2025 · ideonomy-legacy · Adds: Python scripts that call the OpenAI chat API to generate and evaluate idea chains and neologisms, alongside curated text lists.
started Oct 2025 · superpowers · Adds: A full development workflow in which brainstorming, planning, subagent execution, two-stage review and enforced TDD fire automatically in sequence.
started Feb 2026 · skills · Adds: Small, composable skills for daily engineering work, each covering one job, so you install only the ones you want.
started Feb 2026 · ideonomy-engine · Adds: A CLI and library that runs a problem through 28 ideonomic lenses with chaining, synthesis and saved sessions, and returns questions instead of answers.
started Mar 2026 · designer-skills · Adds: UX and visual design judgment as agent-readable skills and commands, from research methods to design-system tokens and heuristic evaluation.
started Mar 2026 · skills · Adds: Installable skills that encode animation curves, motion performance and UI micro-decisions agents tend to get wrong.
started Mar 2026 · skills · Adds: Ten sequenced Claude Code commands that walk an agent through the Minimalist Entrepreneur playbook.
started Apr 2026 · diagram-design · Adds: Draws 45 diagram types as self-contained HTML and SVG in your brand’s palette and fonts, with nodes on a 4px grid and CI overlap checks.
started May 2026 · ideonomy-skill · Adds: Draws a fresh combination of 36 operators, 25 organons and 29 dimension prompts on each run through a local Bash picker, with no network call.
started Jun 2026 · compound-writing · Adds: Writing skills that keep voice rules (VOICE.md) apart from editorial standards (STYLE.md), reached through a context-aware front-door skill, cw-scribe.
started Jun 2026 · ponytail · Adds: A single prompt with a ladder the agent climbs before writing code, so it reuses what exists and writes the minimum, with one small test for logic.
started Jul 2026 · tastemaker · Adds: Grounds generated UI in runnable palette, contrast and anti-slop scripts, with decisions saved to local lock files across sessions.
started Aug 2026 · iso-24495 · Adds: Plain language skills for writing, legal, technical and code output, plus a rule engine that audits Markdown against measurable criteria.
started Sep 2026 · claude-code-playground · Adds: Working hooks, CLAUDE.md files, settings, mods and skills as self-contained folders, with READMEs that explain what failed before each setup worked.
started Oct 2026 · claude-kit · Adds: A planning-and-delegation skill that hands work to tiered subagents, paired with a UI mod that shows a progress bar and live agents panel.
started Oct 2026 · hermes-official-plugins · Adds: Official Hermes Agent integrations, each in its own directory and installed at a reviewed catalog pin, so one can be fixed without an agent release.

Easily confused

A prompt library. Some packs are close to one, but superpowers fires skills in a fixed sequence and tastemaker runs scripts. claude-code-playground is the nearest to a library of examples, and it is meant to be read and copied.
An agent framework. These packs do not replace the agent. They sit inside a host such as Claude Code, Codex or Hermes Agent and change what it does.
Tool or MCP integrations. Skill packs such as skills add instructions and method. Plugins like those in hermes-official-plugins add real tools, such as Spotify playback or Home Assistant control. The difference is knowledge versus access.
Reasoning engines that answer. The ideonomy projects produce questions and combinations to work through. ideonomy-engine says outright that it does not produce answers.

Gaps nobody has filled

An independent comparison of agent results with and without a pack across many tasks. The one benchmark, in DietrichGebert/ponytail, is the author’s own.
A way to detect and resolve conflicts when several packs are installed together, such as a fixed workflow next to a minimal-code prompt.
A link from a reasoning lens session to a result in a coding task, such as a plan or a diff that came from the session.

RationalStuff v3