Best AI Coding Agent Skills in 2026
May 23, 2026
AI coding agents are only as good as the skills you give them. A raw LLM can write a function. A skilled agent can plan a sprint, write tests first, audit its own work for security holes, and remember what it did yesterday. Agent skills are the instruction packs that bridge that gap. We reviewed every major skill collection on GearScope and pulled out the seven worth your time.
1. Agent Skills Specification
What it is: The open standard that defines how agent skills work. SKILL.md format, directory conventions, progressive disclosure, and a Python reference validator.
Strengths: This is the foundation everything else builds on. Quality 5/5, documentation 5/5. The spec is originally from Anthropic but now community-governed under Apache-2.0 and CC-BY-4.0. It defines the YAML frontmatter schema (name, description, license, compatibility, allowed-tools) that every well-behaved skill follows. The Python validator catches malformed skill files before they reach your agent.
Weaknesses: It is a spec, not a skill pack. You do not install this to make your agent smarter. You install it to understand the format and validate your own skills. Pure infrastructure.
Gear Rating: 5/5
Who should use it: Anyone building or publishing agent skills. If you are writing SKILL.md files without reading this spec first, you are doing it wrong.
2. Engram
What it is: A persistent memory system for AI coding agents. Single Go binary, SQLite backend with FTS5 search, exposed as an MCP server, HTTP API, CLI, and TUI.
Strengths: Perfect scores across the board: quality 5, docs 5, setup 5, value 5, ecosystem fit 5. That is rare. Engram solves the real problem of agents forgetting everything between sessions. It ships four interfaces (MCP, HTTP, CLI, TUI) so it works with any agent setup. SQLite with full-text search means fast retrieval. One binary install, no external dependencies.
Weaknesses: Memory systems add complexity. You need to decide what to store and when to prune. The agent has to be prompted to use it, which means your skill files need to reference Engram explicitly.
Gear Rating: 5/5
Who should use it: Any developer running multi-session agent workflows. If your agent starts fresh every time and re-reads the entire codebase, Engram fixes that.
3. last30days
What it is: A multi-source research skill that searches Reddit, X, YouTube, TikTok, Hacker News, Polymarket, and 15+ other platforms within a configurable time window, then synthesizes a structured brief.
Strengths: Another 5/5 sweep. This skill goes where Google cannot. It searches platforms with real engagement metrics, scores results, and produces a structured output your agent can actually reason about. For coding agents doing technology evaluation, dependency research, or competitive analysis, this is the research engine.
Weaknesses: Depends on external platform availability. Rate limits and API changes can break individual sources. The skill needs network access, which some sandboxed agent environments restrict.
Gear Rating: 5/5
Who should use it: Agents doing technology selection, dependency audits, or any task that requires current information from social and discussion platforms.
4. Addy Osmani Agent Skills
What it is: A 23-skill engineering methodology framework covering the full software development lifecycle from spec to ship. Built by Addy Osmani (Google Chrome team).
Strengths: The craft shows. Quality 5/5, documentation 5/5. This is not a random pile of prompts. It is a structured methodology: brainstorm, refine idea, write spec, plan implementation, execute, review, ship. Each skill builds on the previous one. The MIT license and clear documentation make it straightforward to adopt incrementally.
Weaknesses: Opinionated. If your workflow does not match Osmani's SDLC phases, you will fight it. Ecosystem fit is 5/5 but some skills assume Claude Code conventions that need adjustment for other agents.
Gear Rating: 4/5
Who should use it: Teams wanting a disciplined, phase-gated development process baked into their agent. Particularly strong for solo developers who want their agent to act like a proper engineering team.
5. Superpowers
What it is: A cross-platform framework with 14 interlocking skills enforcing TDD-first development across 8 agent harnesses. 193K GitHub stars.
Strengths: Brainstorm before building. Write specs. Create implementation plans. Then code. Superpowers forces agents through a disciplined workflow, and it works across Claude Code, Codex CLI, Cursor, Gemini CLI, OpenCode, and others. The "value" dimension scores 5/5, which is the important one: it actually makes agents produce better code.
Weaknesses: The framework is rigid. If you just want an agent to fix a typo, you do not want to brainstorm first. The 14-skill chain adds overhead to simple tasks. Setup is 4/5, which is fine but not plug-and-play.
Gear Rating: 4/5
Who should use it: Developers working on multi-file features who want their agent to think before it types. Skip this for quick fixes; install it for serious feature work.
6. Antigravity Awesome Skills
What it is: The biggest skill pack on GitHub. 1,464 SKILL.md playbooks installable via npm, targeting Claude Code, Cursor, Codex CLI, and Gemini CLI.
Strengths: Breadth is the entire point. 1,464 skills means coverage for nearly any task. Value scores 5/5 and ecosystem fit 5/5. The npm installer and editorial bundles make it easy to grab what you need without taking everything. MIT and CC BY 4.0 licensing is clean.
Weaknesses: Quality averages 3/5. This is the tradeoff with massive collections: many skills are thin or duplicative. You will spend time curating which ones to actually use. Not every skill has been tested against real-world tasks.
Gear Rating: 4/5
Who should use it: Developers who want a large catalog to cherry-pick from. Browse the collection, grab the 20 skills that fit your stack, ignore the rest.
7. Snyk Agent Scan
What it is: A CLI security scanner that crawls your local machine for agent configs, MCP server definitions, and skill files, then checks them for prompt injection, tool poisoning, and supply-chain threats.
Strengths: This is the first serious security tool built for the agent ecosystem. It auto-discovers agent configurations, connects to MCP servers, retrieves tool manifests, and checks against 15+ issue types. Setup scores 5/5: install, run, read the output. Quality 4/5 and ecosystem fit 5/5 because every agent user needs this.
Weaknesses: It is a scanner, not a fixer. It tells you what is wrong; you still have to fix it. The catalog of issue types is growing but not exhaustive. New attack vectors for agents emerge weekly.
Gear Rating: 4/5
Who should use it: Everyone running agent skills locally. Before you install a skill pack from GitHub, run Snyk Agent Scan on it. This should be part of your agent setup checklist.
Summary Table
| Skill | Rating | Quality | Docs | Setup | Value | Verdict |
|---|---|---|---|---|---|---|
| Agent Skills Spec | 5/5 | 5 | 5 | 4 | 5 | KEEP IT |
| Engram | 5/5 | 5 | 5 | 5 | 5 | KEEP IT |
| last30days | 5/5 | 5 | 5 | 4 | 5 | KEEP IT |
| Addy Osmani Skills | 4/5 | 5 | 5 | 4 | 4 | KEEP IT |
| Superpowers | 4/5 | 4 | 4 | 4 | 5 | KEEP IT |
| Antigravity Pack | 4/5 | 3 | 4 | 4 | 5 | TRY IT |
| Snyk Agent Scan | 4/5 | 4 | 4 | 5 | 4 | KEEP IT |
Quick Verdict
Start with the spec. Read the Agent Skills Specification to understand the format, then install Engram so your agent can remember context between sessions. Add Addy Osmani's framework or Superpowers depending on whether you want SDLC coverage or TDD enforcement. Grab skills from Antigravity for edge cases. Install last30days if your agent does research. And run Snyk Agent Scan on everything before you trust it.
All reviews and gear ratings are at gearscope.xyz.