A curated, open-source skill library for agentic developer tools.
Give your coding agent reusable playbooks for planning, debugging, QA, security, docs, data work, browser automation, and shipping.
Open Skill Marketplace β search 598 skills, preview SKILL.md inline, explore the bundle graph, copy install commands.
Launch Highlights
awesome-agent-skill is now a searchable marketplace of 598 portable SKILL.md playbooks for AI coding agents, with source folders for Cursor and generated Claude Code copies.
| Highlight | Why it matters |
|---|
| Full Asgard AI Platform pack | 314 methodology and domain skills imported with references, examples, scripts, and samples |
| Persona bundles | Focused installs for data science, AI engineering, marketing, finance, legal, research, and Taiwan commerce |
| Multi-agent compatibility | Works with Cursor, Claude Code, Codex, OpenCode, Gemini CLI, and other SKILL.md-aware agents |
| Static marketplace | Browse skills, preview instructions, explore bundle graphs, and copy install commands |
Why This Exists
Most agents start every task from a blank prompt. Skills give them reusable operating procedures: when to ask for clarification, how to run a review, how to triage tests, how to handle PDFs, how to build a quick analysis, how to use browser QA, and how to ship safely.
awesome-agent-skill packages those workflows as portable SKILL.md files for modern coding agents.
Skill Marketplace
Live demo: charlieviettq.github.io/awesome-agent-skill
| Feature | What you get |
|---|
| Task advisor | Describe your task, get skills + bundle + install command |
| Browse & filter | Domain, tier, risk, quality, bundle filters |
| Bundle graph | Interactive force graph of bundle β domain β skill |
| Skill preview | Read SKILL.md on the page without opening GitHub |
Regenerate locally: python3 scripts/generate-catalog.py then serve catalog/.
What You Get
| Area | Includes |
|---|
| Agent workflow | Specs, planning, TDD, incremental implementation, code review, verification |
| Agent systems | MCP, RAG, tool contracts, context-window management, agent evaluation |
| Browser QA | gstack-style browsing, QA, canary checks, benchmarks, screenshots |
| Reliability and security | CI gates, launch checklists, observability, API security, skill supply-chain audit |
| Data and content | Analysis, visualization, documents, spreadsheets, PDFs, presentations |
| Role playbooks | Engineering, research, product, operations, orchestration, language experts |
Works With
Portable SKILL.md folders for agentic coding tools:
| Tool | Install path | Notes |
|---|
| Cursor | .cursor/skills/ | Source of truth in this repo |
| Claude Code | .claude/skills/ | Generated from Cursor skills |
| Codex CLI | .claude/skills/ or project skills dir | Copy compatible skill folders |
| OpenCode | .opencode/skills/ or .cursor/skills/ | Follow your clientβs skill discovery path |
| Gemini CLI / other agents | Project skills directory | Same SKILL.md format; verify client docs |
Copy only the domains you need. Reload the agent session after installing skills.
Compatibility Matrix
| Tool | Install path | Reload required | Commands support |
|---|
| Cursor | .cursor/skills/ | Reload window / new chat | Project rules optional |
| Claude Code | .claude/skills/ | Restart session | Optional .claude/commands/ |
| Codex CLI | project skills dir | Restart session | Follow client docs |
| OpenCode | .opencode/skills/ or .cursor/skills/ | Restart session | Follow client docs |
| Gemini CLI | .gemini/commands/ + skills dir | Restart session | Optional .gemini/commands/ |
Skills remain the source of truth. Commands in .claude/commands/ and .gemini/commands/ are optional wrappers for validate/sync/install workflows.
Quickstart (3 steps)
1. Choose a bundle β browse the Skill Marketplace or ask the advisor:
git clone https://github.com/charlieviettq/awesome-agent-skill.git
cd awesome-agent-skill
python3 scripts/skillhub.py recommend "ship safely with tests and CI"
2. Install into your project (preview first with --dry-run):
python3 scripts/skillhub.py install-bundle ship-ready /path/to/project --format cursor --dry-run
python3 scripts/skillhub.py install-bundle ship-ready /path/to/project --format cursor
Or install the CLI package: pip install -e . then skillhub install-bundle starter . --format cursor.
3. Reload your agent (Cursor: reload window / new chat; Claude Code: restart session).
More install options
# Curated bundles (see registry/bundles.json)
bash scripts/install/install-bundle.sh starter /path/to/project --format both
bash scripts/install/install-bundle.sh agent-builder /path/to/project --format cursor
# Single domain or skill
bash scripts/install/install-domain.sh core-workflow /path/to/project --format cursor
bash scripts/install/install-skill.sh core-workflow/verify-before-done /path/to/project --format cursor
See docs/skillhub-cli.md and docs/comparison.md (how this differs from a plain awesome list).
Marketplace: python3 scripts/generate-catalog.py β deployed via GitHub Pages from catalog/.
Manual install (Cursor):
mkdir -p /path/to/project/.cursor/skills
rsync -a .cursor/skills/core-workflow/ /path/to/project/.cursor/skills/core-workflow/
rsync -a .cursor/skills/security-appsec/ /path/to/project/.cursor/skills/security-appsec/
rsync -a .cursor/skills/writing-docs/ /path/to/project/.cursor/skills/writing-docs/
Install Claude Code skills:
mkdir -p /path/to/project/.claude/skills
rsync -a .claude/skills/ /path/to/project/.claude/skills/
Reload your agent session after copying skills.
Optional Commands
Optional slash-command wrappers (skills remain source of truth):
| Command | Claude Code | Gemini CLI |
|---|
| Validate skills | /validate-skills | validate-skills |
| Sync Claude output | /sync-skills | sync-skills |
| Install domain | /install-domain | copy from scripts/install/ |
| Install bundle | /install-bundle | copy from scripts/install/ |
Copy .claude/commands/ or .gemini/commands/ into your project to enable.
Formats
| Agent surface | Path | Status |
|---|
| Cursor | .cursor/skills/**/SKILL.md | Source of truth (nested paths, e.g. knowledge-work/data/analyze) |
| Claude Code | .claude/skills/<flat-name>/SKILL.md | Generated flat names via scripts/claude-skill-map.json |
Claude uses flat directory names (analyze, va-data-scientist) mapped from nested Cursor paths. Cursor does not need flat copies at .cursor/skills/analyze/ β the full path is the skill id for install and agent discovery.
Cursor β Claude (normal workflow after editing .cursor/skills/):
python3 scripts/convert-to-claude.py --in-repo --force --write-map --prune-orphans
Claude β Cursor (recovery when you edited .claude/skills/ only):
python3 scripts/convert-to-cursor.py --in-repo --only-newer
python3 scripts/convert-to-claude.py --in-repo --force --write-map --prune-orphans
Check parity:
python3 scripts/audit-skill-parity.py
python3 scripts/validate-skills.py --parity
Skill Map
Full index: SKILL_INVENTORY.md
Repository Layout
.
βββ .cursor/skills/ # Cursor skill format, source of truth
βββ .claude/skills/ # Claude Code skill format, generated from Cursor skills
βββ .claude/commands/ # Optional Claude slash commands
βββ .gemini/commands/ # Optional Gemini command wrappers
βββ scripts/ # Conversion, validation, install, metrics
βββ docs/ # Contributor guides, distribution, release cadence
βββ .github/workflows/ # Skill validation CI
βββ SKILL_INVENTORY.md # Human-readable skill index
Highlights
| Folder | Good For |
|---|
core-workflow/ | Spec-first implementation, planning, TDD, verification, reviews |
ai-agent-systems/ | MCP servers, RAG systems, agent evals, tool schemas |
gstack/ | Browser QA, ship workflows, design review, scrape flows |
voltagent/ | Role-based subagent playbooks |
security-appsec/ | API security, secure design, skill supply-chain checks |
reliability-ops/ | CI gates, SLOs, launch readiness, postmortems |
writing-docs/ | PDF, DOCX, XLSX, PPTX, prose polish |
visualization/ | Charts, figures, infographics, data storytelling |
Design Principles
- Portable: skills are plain folders with
SKILL.md.
- Composable: copy one domain or the full pack.
- Public-safe: no secrets, customer data, or private org assumptions.
- Agent-first: written as workflows an agent can follow, not as static articles.
- Reviewable: skills stay small; long references belong in
reference.md or examples.
Updating Skills
Edit the Cursor source:
$EDITOR .cursor/skills/core-workflow/spec-driven-development/SKILL.md
Regenerate Claude output (and prune stale nested copies under .claude/skills/):
python3 scripts/convert-to-claude.py --in-repo --force --write-map --prune-orphans
Run validation:
python3 scripts/validate-skills.py --parity
Generate metrics snapshot:
python3 scripts/repo-metrics.py
Private Skills
Keep team-specific or sensitive skills in your own repository under paths such as .cursor/skills/private/ and .claude/skills/private/. This public pack is intentionally generic.
Contributing
Contributions welcome. See .github/CONTRIBUTING.md, docs/skill-writing-guide.md, and docs/review-rubric.md.
Recent changes: CHANGELOG.md | Release cadence: docs/RELEASE_CADENCE.md
Submit to Awesome Lists
If you maintain an awesome-list or agent-tools roundup, consider linking this repo under categories such as AI agents, Cursor, Claude Code, MCP, or developer tools. Suggested blurb:
awesome-agent-skill β 170+ portable agent skills (planning, QA, security, MCP, browser automation, data/docs) for Cursor and Claude Code.
External catalogs we track: EXTERNAL_SKILLS.md | Submission tracker: docs/distribution.md
License
MIT. See LICENSE. Use it, fork it, and adapt it for your own agents.
