In one paragraph
Agent skills and plugins give a coding agent new behaviors by installing Markdown or code files that describe procedures, constraints, or domain knowledge. Some packs ship dozens of small skills you mix and match. Others enforce a complete workflow sequence. A few draw on external standards or frameworks: a business methodology, principles inspired by a plain language ISO standard, a UX practice body of knowledge, or a structured reasoning system with named lenses. The underlying format is similar across all of them, but the scope varies from a single UI taste constraint to a full software development lifecycle.
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 set of small, single-job skills you install individually and invoke on demand. Each skill covers one task: run a TDD cycle, diagnose a bug, interview the developer before starting work. skills is the clearest example: each skill is a separate file you pick from a menu. Some skills are user-invoked (you type /grill-with-docs), and others are model-invoked (the agent reaches for /tdd when it fits).
Domain skills
Skills that encode judgment from a specific field rather than general engineering. The agent learns when to apply a particular research method, how to sequence a business decision, or what a heuristic evaluation should cover. designer-skills packages 111 UX skills across 9 plugins. skills translates Sahil Lavingia’s Minimalist Entrepreneur framework into ten sequenced commands.
Development methodologies
A single opinionated workflow that fires automatically in sequence: brainstorm, plan, implement via subagents, review, ship. superpowers is the main example, enforcing RED-GREEN-REFACTOR and deleting code written before tests.
Structured reasoning lenses
Rather than producing answers, these apply a named set of thinking frames to a problem and return structured questions. ideonomy-engine runs a problem through up to 28 named reasoning divisions drawn from Patrick Gunkel’s Ideonomy framework. ideonomy-legacy contains Python scripts that use the OpenAI API for idea generation, chaining, evaluation, and branching, alongside curated topic lists. It has no README or install instructions.
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"] f2 --> r2["Owl-Listener/designer-skills"] f2 --> r3["slavingia/skills"] f2 --> r4["codeswithroh/tastemaker"] f2 --> r5["GaZmagik/iso-24495"] f3 --> r6["obra/superpowers"] f4 --> r7["kindgracekind/ideonomy-legacy"] f4 --> r8["Morpheis/ideonomy-engine"]
Where the new ideas are
- superpowers can spawn a fresh subagent per task and review each one before proceeding; the alternative runs every task inline in one session with a single final review.
- ideonomy-engine treats the agent skill as a question generator rather than an answer generator, with lens chaining and cross-lens synthesis as first-class operations.
- tastemaker uses runnable Python scripts and local lock files to make style constraints self-enforcing across sessions, rather than relying on prompt reminders.
- iso-24495 applies principles inspired by ISO 24495, a published plain language standard series, as a rule engine with measurable criteria that are the project’s own proxies, not clauses of the standard.
- designer-skills structures skills as nouns and commands as verbs, separating knowledge units from workflow chains.
Side by side
| Repo | Approach | Auto-triggers | Subagent support | External standard or framework | Audit or verification step |
|---|---|---|---|---|---|
| kindgracekind/ideonomy-legacy | Structured reasoning lenses | Not stated | Not stated | Not stated | Not stated |
| obra/superpowers | Development methodologies | Yes, fires on session start | Yes, one subagent per task (inline execution is the alternative) | Not stated | Yes, spec compliance then code quality review per task |
| mattpocock/skills | Skill collections | Yes, model-invoked skills fire when the agent decides they fit | Yes, /code-review uses parallel sub-agents | Not stated | Yes, /diagnosing-bugs enforces reproduce-minimise-fix loop |
| Morpheis/ideonomy-engine | Structured reasoning lenses | Not stated | Not stated | Yes, Patrick Gunkel’s Ideonomy framework | Yes, synthesize scans for tensions and convergences |
| Owl-Listener/designer-skills | Domain skills | Not stated | Not stated | Not stated | Yes, /visual-critique:critique-screen runs seven visual critiques and returns a prioritized fix list |
| slavingia/skills | Domain skills | Not stated | Not stated | Yes, The Minimalist Entrepreneur by Sahil Lavingia | Not stated |
| codeswithroh/tastemaker | Domain skills | Yes, triggers automatically on UI build or style requests | Not stated | Not stated | Yes, contrast checker and anti-slop scanner scripts |
| GaZmagik/iso-24495 | Domain skills | Yes, specialist skills trigger automatically from core skill | Not stated | Inspired by the ISO 24495 plain language series; unofficial | Yes, rule engine audits Markdown against measurable criteria |
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/>slavingia/skills"] n7["started Jul 2026<br/>codeswithroh/tastemaker"] n8["started Aug 2026<br/>GaZmagik/iso-24495"] n1 --> n2 --> n3 --> n4 --> n5 --> n6 --> n7 --> n8
- started Jul 2025 · ideonomy-legacy · Adds: Python scripts that call the OpenAI API to generate idea chains, evaluate hypotheticals, and branch reasoning, alongside curated text lists on topics like flowers, fractals, and cooperation.
- started Oct 2025 · superpowers · Adds: A complete, auto-firing SDLC methodology that can spawn isolated subagents per task, enforces TDD, and reviews each task against spec before continuing, covering the full path from brainstorm to ship.
- started Feb 2026 · skills · Adds: A menu of small, composable engineering skills covering TDD, bug diagnosis, spec writing, code review, and architecture scanning, installable one at a time through the npx installer, and split into user-invoked and model-invoked groups.
- started Feb 2026 · ideonomy-engine · Adds: A CLI and TypeScript library that applies 28 named ideonomic reasoning lenses to a problem, generates structured questions rather than answers, supports lens chaining and cross-lens synthesis, and persists sessions to disk.
- started Mar 2026 · designer-skills · Adds: 111 UX and visual design skills and 34 commands across 9 independently installable plugins, encoding design research, systems, interaction, and delivery judgment for Claude Code and Gemini CLI.
- started Mar 2026 · skills · Adds: Ten sequenced Claude Code commands that translate the Minimalist Entrepreneur framework into agent-runnable steps from community discovery through pricing, marketing, and final methodology review.
- started Jul 2026 · tastemaker · Adds: A self-triggering skill that grounds AI UI generation in real pixel-extracted palettes or harmony-rule-derived palettes, running contrast checks and anti-slop scans and writing decisions to local lock files for cross-session persistence.
- started Aug 2026 · iso-24495 · Adds: Seven skills that apply principles inspired by the ISO 24495 plain language series to agent output, with a rule engine that audits Markdown files against measurable criteria including sentence length, legalese, heading depth, and undefined acronyms.
Easily confused
Not a prompt library. A prompt library stores text snippets you paste into a chat. Skills and plugins here install into an agent’s runtime and fire procedurally, sometimes automatically, sometimes as slash commands. The agent follows a defined procedure rather than simply receiving a better prompt. The exception to the install step is kindgracekind/ideonomy-legacy, a set of standalone Python scripts.
Not an agent framework. Agent frameworks define how agents are built and how they call tools. Skills and plugins here extend an already-running agent (Claude Code, Gemini CLI, Windsurf) with new behaviors, the way a browser extension extends a browser.
Not fine-tuning. None of these repos modify model weights. They shape behavior through installed instruction files and runnable scripts.
Gaps nobody has filled
- No skill pack covers security review or threat modeling as a domain, despite covering UX, business methodology, and plain language.
- Lens chaining and skill sequencing are each handled differently across repos, with no shared composition protocol between packs.
- Packs test their own skills (obra/superpowers runs skill-behavior tests with a separate eval harness), but no benchmark compares skills across packs or tracks output quality across skill versions.
- No mechanism exists for a project to pin a skill to a specific version and detect breaking changes across updates.