Awesome Agent Skills

English | 繁體中文

A curated directory of externally maintained Agent Skills, skill collections, and supporting tools for AI-agent workflows.

[!IMPORTANT] This is a links-only directory. Skills stay in their source repositories. Listing pull requests must not add SKILL.md, scripts, copied skill folders, or platform-specific indexes to this repository.

Scope

This directory includes:

• Agent Skills — directly usable, reusable instruction packages with a public implementation.
• Skill Collections — repositories containing multiple usable Agent Skills.
• Tooling & Integrations — plugins, CLIs, MCP integrations, and related projects built for agent workflows.

This repository is not a package registry, a host for community skill files, a security audit, or an endorsement of the linked projects. Each entry remains maintained and licensed by its upstream owner.

Browse

Agent Skills

Development & Code

• agent-architecture — Guides agent architecture discovery and evidence-based diagnosis, producing requirements, capability methods, tool and state contracts, diagrams, and acceptance criteria. Type: Skill · Platforms: Codex, Claude Code
• agent-context-lifecycle — Preserves project decisions, reconciles affected documents, and prepares session handoffs with explicit delivery evidence. Type: Skill · Platforms: Codex, GitHub Copilot
• Cloudish — Deploys a Dockerfile, source folder, or container image to Cloudish and reports the live URL, using an API key the agent creates for itself. Type: Skill · Platforms: Cross-platform
• ManualMode — Reserves bounded tasks from real project work for manual coding practice and reports verification metadata to ManualMode. Type: Skill · Platforms: Cross-platform
• Shipvela — Publishes static websites through the hosted OAuth MCP connector with owner approval, deployment tracking, and build logs. Type: Skill · Platforms: Codex, Claude Code
• Openhook — Creates webhook inboxes, inspects callbacks, and forwards events to local handlers through MCP or REST. Type: Skill · Platforms: Cross-platform
• reverse-engineer-anything — Inspect shipped binaries and JavaScript/Electron apps with REA evidence; deep native analysis requires separately installed Hopper, Ghidra, or IDA. Type: Skill · Platforms: Codex
• UIZZE anti-ui-slop — Builds a product-specific UI design contract and applies a pre-ship finish gate using real interface references. Type: Skill · Platforms: Cross-platform
• what-could-break — Traces what an uncommitted change can break outside its own diff and requires running real code to prove the safety claim instead of describing the risk. Type: Skill · Platforms: Codex, Claude Code

Data & Analysis

• AuditorProcessual — Guides Brazilian legal case analysis, evidence tracing, official-source research, and labor-law petition drafting. Type: Skill · Platforms: Cross-platform
• credit-risk-model — Guides credit-risk model development, validation, reporting, and strategy analysis. Type: Skill · Platforms: Cross-platform
• KnowSift — Compiles mixed research sources into certificate-backed knowledge documents that preserve evidence, scope, viewpoints, disputes, and rejected claims. Type: Skill · Platforms: Codex, Claude Code
• TrendCite — Collects public Hacker News, GitHub, RSS/Atom, and Reddit signals and turns them into scored, source-traceable content opportunity briefs with confidence and counterpoints. Type: Skill + CLI · Platforms: Claude Code
• Xquik x-twitter-scraper — Routes X research, extraction, monitoring, and confirmation-gated publishing through REST, MCP, SDK, or webhook workflows. Type: Skill · Platforms: Cross-platform

Communication & Writing

• essay-writer — Plans, drafts, researches, edits, and reviews essays while preserving voice, evidence, and uncertainty. Type: Skill · Platforms: Codex, Agent Skills-compatible agents
• Novel Writing — Guides fiction planning, drafting, and revision with viewpoint, scene-causality, dialogue, and style-preservation checks. Type: Skill · Platforms: Codex
• publora-post-ideas — Offers a choice of angles for a social post, drafts the selected one, and can schedule it through Publora. Type: Skill · Platforms: Claude Code, Agent Skills-compatible agents
• xLetter — Gives Agents a reusable mailbox for asynchronous text messages and public contact cards. Type: Skill · Platforms: Codex, OpenClaw

Creative & Media

• figure-rebuild — Reconstructs paper figures and technical diagrams as editable PowerPoint shapes, live text, and connectors, with audited LaTeX assets. Type: Skill + CLI · Platforms: Codex
• runapi-cli-skill — Teaches agents to run image, video, audio, and language-model jobs through the RunAPI CLI. Type: Skill · Platforms: Cross-platform

Productivity & Organization

• Agent Coordinator — Coordinates complex work through a resumable local dependency graph, declared write scopes, reconciliation before retries, and repeatable closeout checks. Type: Skill · Platforms: Codex
• GFT Map — Turns linked Agent conversations into local, editable Doc and Map context that people can correct and later Agents can read. Type: Skill · Platforms: Codex, Claude Code
• Goal Instruction Writer — Previews and drafts durable agent Goals with acceptance evidence, drift controls, proportional verification, and explicit stop conditions. Type: Skill · Platforms: Codex
• Hyperconsciousness — Discovers installed skills, local notes and encrypted knowledge through scoped grants. Type: Skill · Platforms: Cross-platform
• Syncbook — Prepares human-approved profiles, finds reciprocal collaborators, and coordinates through scoped agent access. Type: Skill · Platforms: Cross-platform
• wiki — Creates and queries indexed Markdown knowledge bases with deterministic CLI support for agents. Type: Skill · Platforms: Codex, Claude Code

Skill Collections

• Agent Toolkit — Equips coding assistants with a reusable catalog of Agent Skills, personas, MCP templates, and multi-agent loop and swarm workflows. Type: Collection · Platforms: Claude Code, Cursor, OpenCode, GitHub Copilot, Windsurf, Pi
• Code2Skill — Turns authorized application source code into runnable Function, MCP tool, workflow Skill, and offline-test packages, with separate flow and source review skills. Type: Collection · Platforms: Codex, Claude Code, Kimi Code
• Community Agent Skills — Guides agents through inbox triage, file search, meeting scheduling, subscription tracking, meal planning, backups, task capture, translation, batch file operations, and Cloudflare infrastructure management, with confirmation before irreversible actions. Type: Collection · Platforms: Cross-platform
• CreatorSkills — Guides agents through content-creator workflows for YouTube scripting, thumbnail concepts, SEO, and audience personas. Type: Collection · Platforms: Cross-platform
• EditRoast Open Skills — Provides two reusable agent skills for chapter-based full-manuscript critique and proofreading, with local file preparation and resumable progress. Type: Collection · Platforms: Claude Code, Codex
• Fillfolio portfolio MCP skills — Guides agents through read-only portfolio, cash, spending and budget reviews using an authorized Fillfolio MCP connection. Type: Collection · Platforms: Codex
• mblode/agent-skills — Guides agents through UI and typography audits, code review, documentation, PR maintenance, and npm releases with 27 skills. Type: Collection · Platforms: Claude Code, Codex
• NotFair — Guides agents through SEO, GEO, paid-media, and analytics workflows using live marketing data and approval-gated changes. Type: Collection · Platforms: Cross-platform
• OrkasVideoStudio — Routes coding agents through 14 skills for planning, composing, editing, generating, and assembling videos. Type: Collection · Platforms: Codex, Claude Code
• Research Agent Skills — Guides agents through the full academic research lifecycle: literature and systematic reviews, hypothesis generation, experimental design, field-specific data analysis, paper writing, citation verification, journal submission and peer review. Type: Collection · Platforms: Cross-platform
• Slashskills — Guides agents through software design, implementation, release checks, and documentation workflows. Type: Collection · Platforms: Codex, Claude Code, Cursor
• Solaris Dev Shop — Routes a task to one accountable specialist skill and supplies that specialist’s skill files. Type: Collection · Platforms: Cross-platform
• Suede Creator Skills — A multi-domain collection covering agent orchestration, code review, evaluation, product, design, and growth workflows. Type: Collection · Platforms: Cross-platform
• The Enso Method — Guides coding agents through requirements, design, implementation and review while recording business assumptions with evidence and confidence. Type: Collection · Platforms: Claude Code, Codex

Tooling & Integrations

• Agent QA — Runs natural-language web and mobile QA workflows through a CLI, MCP server, and three evidence-oriented Agent Skills. Type: CLI + MCP + Collection · Platforms: Codex, Agent Skills-compatible agents
• Divan — Guides Codex through natural-language software tasks, implementation, review, and verification. Type: Plugin + Collection · Platforms: Codex
• go-public — Guides agents through Git-history audits for secrets and personal data, human-reviewed fixes, and rescanned exports before publication. Type: Plugin + Skill · Platforms: Claude Code
• OpenSocrates — Guides agents through checking assumptions, comparing options, and evaluating evidence with reusable reasoning procedures. Type: Plugin + Skill · Platforms: Claude, Claude Code, Codex, Antigravity
• QFactory — Guides agents to find and adapt reusable workflows from its catalog, then run approved workflows locally through Q-Core (website). Type: Plugin + Skill · Platforms: Claude Code, Codex
• rhost — Runs ordinary commands on SSH-reachable Linux hosts with local-like process semantics. Type: CLI + Plugin + Skill · Platforms: Agent Skills-compatible agents
• SandBase — Connects AI agents to unified model and tool APIs through a local CLI-managed MCP bridge and Agent Skill. Type: CLI + MCP + Skill · Platforms: Cross-platform
• SkillPreflight — Inspects third-party Agent Skills before installation and surfaces static risk findings, token estimates, and quality scores for agent-assisted review. Type: CLI + Skill · Platforms: Cross-platform
• TweetClaw — Provides supervised X research, publishing, media, follower export, giveaway, and monitoring workflows. Type: Plugin + Skill · Platforms: OpenClaw, Agent Skills-compatible agents
• UNSLOP — Audits generated UI against design-system contracts and pinned rules, then writes fix plans using the full plugin setup (website). Type: Plugin + Skill · Platforms: Claude Code, Codex

Using a Listed Project

1. Open the linked canonical source and read its current documentation.
2. Review its SKILL.md, scripts, dependencies, requested permissions, and external services before use.
3. Follow the upstream installation instructions for your agent platform.
4. Confirm that its license, platform support, data handling, and security assumptions fit your environment.

Compatibility and availability can change; the upstream repository is the source of truth.

Contributing

For a normal listing pull request:

1. Add or update one upstream project in the appropriate section.
2. Edit both README.md and README_ZH.md, keeping the name, URL, type, and platform metadata in sync.
3. Change only those two files; do not vendor the upstream skill into this repository.

Entry format:

- [Project name](canonical public source) — One concise, factual description. `Type: Skill` · `Platforms: Cross-platform`

Read CONTRIBUTING.md for eligibility, ordering, disclosure, pull-request scope, and commit-signing requirements.

References

• Agent Skills specification
• Build skills with Codex
• Extend Claude with skills

License and Disclaimer

The documentation and assets in this repository are available under the MIT License. Linked projects use their own licenses and terms. Inclusion does not imply endorsement, compatibility guarantees, or security review.