AgentSidekick

Agent skills and plugins

3,144 agent skills and plugins found on GitHub: who publishes them, what they do, which platforms they name, and what static inspection does and does not tell you.

Executive summary

  • 2,056 agent skills and 1,088 plugins are indexed; 2,620 of them have profile pages.

  • 1,135 GitHub accounts publish them, and 190 publish more than one.

  • The most-starred repository is mattpocock/skills (329,350 stars); 3,144 of 3,144 repositories have fetched metrics.

  • 925 skills and plugins have a repository created in the last 12 months.

  • 791 (25.2%) name a specific agent platform in their description.

  • 691 have been statically inspected; signals describe what files reference, not whether they are safe.

Every figure above is computed from the database DER

Methodology

How this report was produced, and its limits

Generated on 2026-11-21 from the AgentSidekick database. The latest completed ingestion step finished at 2026-11-21 13:23 UTC. Tables queried: products, product_repos, repositories, product_taxonomy, taxonomy_nodes, product_platforms, product_signals. The catalogue holds 73,820 active products, of which 9,469 meet the data-completeness bar and have a profile page.

Rules packs (267 products, such as editor rule files) are a related but separate type and are excluded here.

Classification method: product type comes from the source (registry entries are MCP servers by definition) or is inferred from repository topics; categories, capabilities, integrations and industries come from keyword rules in the editable taxonomy, so they are derived, not checked by a person. Curated records were researched against the official sources listed on each record.

Known limits:

  • Skills and plugins are discovered through GitHub repository topics, so each product is one repository. A collection of 50 skills in one repository counts once; skills not tagged with a topic are missed.
  • Type (skill vs plugin) is inferred from topics and repository layout, not declared by a registry.
  • Static-inspection signals (where present) come from reading files as text: they flag references such as shell commands or network calls. They are not a security review, and no product on this site has been security reviewed.
  • Launch dates are the first publish date in the source registry (official MCP Registry) or the GitHub repository creation date. They are not product announcements, and a re-published entry can look like a new launch.
  • GitHub stars measure attention on a repository, not installs, active users or revenue. Several products can share one repository.
  • Categories, capabilities and integrations are assigned by keyword rules over names and descriptions (derived). Products that do not name a capability are missed, and generic words can produce false matches.

Note. Claim labels: source = stated by the publisher or registry; curated = checked by us against cited sources; derived = computed by a documented rule; heuristic = a rule of thumb with no measurement behind it. Full method: methodology.

Size of the catalogue

Agent skills (folders with a SKILL.md file and optional scripts) and plugins (bundles of skills, commands, hooks and MCP servers for a specific agent) are a small catalogue next to MCP servers: 3,144 products against tens of thousands of registry entries. That gap reflects discovery more than activity. There is no central skill registry to ingest, so only repositories that tag themselves with a topic appear.

Skills and plugins indexedDER

Skills and plugins indexed
TypeProductsWith profile page
Agent skills2,0561,898
Plugins1,088722

Skills and plugins by repository creation monthSRC

Unit: products per month, 2025-12 to 2026-11.

  • Agent skills
  • Plugins
Skills and plugins by repository creation monthUnit: products per month, 2025-12 to 2026-11.. Agent skills: 12 points, latest 13 products in 2026-11. Plugins: 12 points, latest 9 products in 2026-11.
Show the data as a table
Skills and plugins by repository creation month. Unit: products per month, 2025-12 to 2026-11.
Agent skills (products)Plugins (products)
2025-122512
2026-017229
2026-026327
2026-038542
2026-049536
2026-054117
2026-064919
2026-073917
2026-084111
2026-094024
2026-105861
2026-11139

Who publishes

1,135 accounts publish the 3,144 repositories. Most publish one; 190 publish several. 2,721 repositories (86.5%) were pushed to in the 90 days before this build and 0 are archived. Because skills are cheap to write, many repositories are personal collections; stars are the quickest way to find the ones others have looked at, but not evidence that a skill works.

Attention is extremely concentrated. The 10 most-starred repositories hold 3,239,023 of 78,189,903 stars across all skill and plugin repositories (4.1%), while 15 repositories have no stars at all. Many of the top repositories are collections maintained by model vendors or well-known developers, so a newcomer's skill is far more likely to be found through a collection than as a standalone repository.

Accounts publishing the most skills and pluginsSRC

Accounts publishing the most skills and plugins
GitHub accountSkillsPluginsCombined stars
ccplugins (external)0150151,800
anthropics (external)21566,843,571
grcengclub (external)07131,240
nexu-io (external)6306,649,900
phuryn (external)6201,720,834
composio-community (external)6201,062,245
composiohq (external)6104,878,475
k-dense-ai (external)6103,536,780
wshobson (external)6102,608,360
github (external)6102,526,132
davila7 (external)6102,076,867
googleworkspace (external)6101,977,620
nanocoai (external)6101,899,113
elementalsouls (external)610916,769
orchestra-research (external)610897,615

Most-starred skill and plugin repositoriesSRC

Most-starred skill and plugin repositories
ProductTypeRepositoryStarsLast push
mattpocock-skillsPluginmattpocock/skills329,3502026-11-19
finishing-a-development-branchAgent skillobra/superpowers323,2972026-11-03
systematic-debuggingAgent skillobra/superpowers323,2972026-11-03
dispatching-parallel-agentsAgent skillobra/superpowers323,2972026-11-03
receiving-code-reviewAgent skillobra/superpowers323,2972026-11-03
executing-plansAgent skillobra/superpowers323,2972026-11-03
subagent-driven-developmentAgent skillobra/superpowers323,2972026-11-03
brainstormingAgent skillobra/superpowers323,2972026-11-03
diagnosing-superpowersAgent skillobra/superpowers323,2972026-11-03
test-driven-developmentAgent skillobra/superpowers323,2972026-11-03
requesting-code-reviewAgent skillobra/superpowers323,2972026-11-03
writing-plansAgent skillobra/superpowers323,2972026-11-03
verification-before-completionAgent skillobra/superpowers323,2972026-11-03
using-superpowersAgent skillobra/superpowers323,2972026-11-03
using-git-worktreesAgent skillobra/superpowers323,2972026-11-03

What they do

Matched by keyword to specific capabilities, skills and plugins lean heavily toward coding work. 1,139 (36.2%) match no specific capability at all, typically because the repository description says only "my skills" or similar.

Most common capabilities (keyword-derived)DER

Unit: products. Top 15 shown.

Most common capabilities (keyword-derived)Unit: products. Top 15 shown.. Highest: Unified APIs, 114 products.Unified APIs114Agent Skills & Plugins102Prompt Optimization98Multi-agent Coordination48SEO Research47Design Systems45Digital Marketplaces44Personal Knowledge Management43AWS Integration43Authentication43Instruction Management41Code Quality40Video Editing37Azure Integration37Headless Browsing36
Show the data as a table
Most common capabilities (keyword-derived). Unit: products. Top 15 shown.
products
Unified APIs114
Agent Skills & Plugins102
Prompt Optimization98
Multi-agent Coordination48
SEO Research47
Design Systems45
Digital Marketplaces44
Personal Knowledge Management43
AWS Integration43
Authentication43
Instruction Management41
Code Quality40
Video Editing37
Azure Integration37
Headless Browsing36

791 products name a platform in their description. The most named is Claude Code (680). Since Agent Skills is a shared format, a skill written for one coding agent often loads in others; the name in the description reflects where the author tested it, not a limit.

Platforms named by skills and plugins (vendor-claimed)SRC

Unit: products. Top 12 shown.

Platforms named by skills and plugins (vendor-claimed)Unit: products. Top 12 shown.. Highest: Claude Code, 680 products.Claude Code680Codex282Cursor126Gemini CLI44OpenClaw35OpenCode34Google Antigravity22ChatGPT19GitHub Copilot15Devin Desktop (formerly Windsurf)14Claude (claude.ai apps)8Claude Desktop8
Show the data as a table
Platforms named by skills and plugins (vendor-claimed). Unit: products. Top 12 shown.
products
Claude Code680
Codex282
Cursor126
Gemini CLI44
OpenClaw35
OpenCode34
Google Antigravity22
ChatGPT19
GitHub Copilot15
Devin Desktop (formerly Windsurf)14
Claude (claude.ai apps)8
Claude Desktop8

Static-inspection signals

Note. Signals are not safety verdicts. A signal says a file references something (for example a shell command, a network request or a credential); it does not say whether that is harmful. Security review: not reviewed, for every product.

Static-inspection signals among skills and pluginsDER

Static-inspection signals among skills and plugins
SignalProductsShare of inspected
shell-execution58985.2%
bundled-scripts23033.3%
credential-access16123.3%
network-access15622.6%
destructive-commands223.2%
external-scripts91.3%

691 skills and plugins (22.0%) have been inspected. The most common signal is "shell-execution", seen in 589 of them (85.2%). High prevalence is expected: skills exist to let an agent do things, and doing things means running commands and reaching services. The useful question is not whether a signal is present but whether it matches what the skill says it does.

Why this matters: a skill can include scripts the agent will run with the user's permissions. Reading a skill before installing it is the only reliable check today; signals, when available, only help decide which files to read first.

Recommendations

Editorial: the judgement of AgentSidekick, drawn from the figures above. Not a measurement, not investment advice.

  1. Editorial

    skill authors should tag repositories with a discoverable topic and write a one-line description naming the capability; 36.2% of skills and plugins match no capability today.

  2. Editorial

    read bundled scripts before installing any skill, whatever its stars; 691 have automated signals, and none has been security reviewed.

  3. Editorial

    test skills on more than one agent. Only 25.2% name a platform, and the Agent Skills format is shared, so wider claims are often possible.

Sources

Data behind this report

Cite this report

Agent skills and plugins. AgentSidekick, 2026-11-21. https://www.agentsidekick.dev/reports/agent-skills-and-plugins/

Figures change when the database is refreshed; cite the date shown.

Page data last updated .