State of the skills market¶
Observation date: October 27, 2026. Install counts, repository metrics, releases, and compatibility are snapshots, not live status.
Agent skills differ in what they supply, how they are installed, and who controls updates. This page separates those roles so you can choose the right type.
What is changing¶
Reusable skills increasingly ship as versioned, updatable packages through two channels. Cross-agent installer CLIs, such as skills.sh's npx skills, copy editable files into the target agent's skill directory. Native plugin marketplaces for Claude Code, GitHub Copilot, and Codex install managed bundles inside one runtime, with marketplace-specific update policies.
Some libraries also bundle an orchestrating mode: a dispatcher that reads a request and selects a playbook or skill. This adds workflow decisions to a library of capabilities.
Taxonomy¶
| Category | Example | What it provides | What it does not guarantee |
|---|---|---|---|
| Specification | Agent Skills | SKILL.md structure, metadata, progressive disclosure |
Identical behavior in every runtime |
| Composable library | mattpocock/skills |
Small, independently selectable skills | A mandatory lifecycle or governance model |
| Lifecycle library | addyosmani/agent-skills |
Skills mapped to development stages via slash commands | A single mandatory path through those stages |
| Orchestrating mode | pstack/poteto-mode |
A dispatcher that routes requests into a library's other skills | A standalone capability — it depends entirely on the library it dispatches into |
| Methodology | Superpowers | Mandatory skills, lifecycle sequencing, TDD, review, enforcement | Team orchestration or enterprise governance |
| Reference implementation | anthropics/skills |
Example skills, templates, Claude Code plugin distribution | Uniform open-source licensing across every included skill |
| Installer and registry | skills.sh / vercel-labs/skills |
Discovery, vendored install, update, ranking | Curation, signing, attestation, or semantic parity |
| Native marketplace | Claude Code, GitHub Copilot, Codex | Managed, auto-updating plugin bundles inside one runtime | Cross-runtime portability or a local, editable copy |
| Curated channel | github/awesome-copilot |
Copilot-focused community content | That the content is official guidance or portable beyond Copilot |
| Runtime | Claude Code, Copilot, Codex, Cline, Goose, and others | Skill discovery and execution | Support for every optional skill feature |
These roles overlap but are not interchangeable. A specification defines a format; a library supplies content; an orchestrating mode selects skills; an installer or marketplace distributes them; a runtime executes them; a methodology prescribes their use.
Why it matters¶
A runtime may parse a SKILL.md file without supporting every feature it uses.
A prior cross-runtime pass found basic skill content worked across all 18 agents checked. Hooks worked in only 4 of them. context: fork worked in just 1.
No installer, marketplace, or library we examined signs, attests, or lockfile-pins a skill on any install path. Most evaluation evidence comes from vendors, not independent audits. Check provenance and runtime behavior before standardizing on a library.
Installation also needs checking. Per-skill installers can omit shared dependencies, and a marketplace bundle may differ from an older vendored copy. No unified signal identifies version drift across channels. Choose an update policy rather than relying on defaults.
mattpocock/skills and addyosmani/agent-skills illustrate two library approaches: composable and lifecycle-mapped. Both ship through native marketplaces and installer CLIs. Neither defines an end-to-end SDLC.
Relationship to the model, runtime, and methodology layers¶
Skills support the runtime and methodology layers of the three-layer model. Runtimes load and execute them; some methodologies make them mandatory. OpenSpec generates skills, HVE has moved most reusable task templates to skills, and BMAD's V7 preview installs its method as skills.
Star counts and install figures indicate attention, not production adoption. See Research Methodology for the evidence labels.
Where to go next¶
- Skill Libraries — compare
mattpocock/skills,addyosmani/agent-skills,anthropics/skills,vercel-labs/agent-skills, and watchlist collections. - Orchestrating Modes — how
pstack/poteto-moderoutes requests, how ports differ, and why it is unrelated topotato-mode. - Distribution and Governance — installation, updates, provenance, permissions, and rollout.
- Choosing a Skills Approach — match a library, mode, methodology, or internal catalog to your needs.
- Superpowers — a skill-driven methodology with mandatory TDD and review.