Portable agent skills, curated skillsets, MCP presets, and routing templates for AI coding agents.
AgentSkills gives an agent reusable operating instructions for real work: planning, implementation, research, testing, documentation, release, creator operations, and security review. The core content is readable Markdown and YAML; host-specific setup stays in adapters and setup guides.
Latest published release: v0.2.2. The default branch also contains an
unreleased development addition for local YouTube production.
| You need | Start with |
|---|---|
| One focused workflow | Install one skill with gh skill install |
| A complete workflow family | Install a curated skillset |
| Guidance for a new repository | Use agentskills-project-onboarding |
| Setup for a specific host | Read the setup guides |
| The full catalog and decision path | Open the AgentSkills Wiki |
Preview a skill before installing it:
gh skill preview jeremylongworth-source/AgentSkills game-threejs-developmentInstall a pinned release for Codex:
gh skill install jeremylongworth-source/AgentSkills game-threejs-development --agent codex --scope user --pin v0.2.2Change --agent and --scope for the host and install target you use. See
compatibility for the current host matrix.
Clone the repository, preview the local install, then install it:
git clone https://github.com/jeremylongworth-source/AgentSkills.git
cd AgentSkills
.\scripts\install-skillset.ps1 game-dev -DryRun
.\scripts\install-skillset.ps1 game-devThe PowerShell installer is a Codex adapter. For other hosts, use the skillset manifest as a shopping list and follow the relevant host setup guide.
Start with agentskills-project-onboarding. It helps an agent inspect the
repository, choose the smallest useful skillset, plan routing, identify MCP
needs, and define verification steps before installation.
Use the smallest skillset that matches the job. These are the most common starting points:
game-devfor game design, implementation, assets, QA, and launchfrontend-productfor product shaping, UI quality, accessibility, analytics, and release readinessengineering-deliveryfor issue planning, PR review, regression, and release riskbackend-apifor API contracts, schemas, auth, service boundaries, and backend testsdevops-cloud-releasefor deployment, CI, configuration, rollback, and production readinessquality-testingfor unit, integration, E2E, accessibility, performance, and regression worktechnical-documentationfor READMEs, API docs, release notes, migrations, and onboardingagentops-evaluationfor benchmarks, scenarios, before/after reports, scoring, and prompt regressionskill-security-auditfor review-only install and publish security checkscreator-content-enginefor content strategy, scripts, calendars, captions, and repurposingcreator-brand-dealsfor pitches, media kits, sponsorships, rights, and campaign recapscreator-youtube-local-productionfor human-reviewed local narrated video production
See the full skillset catalog and the wiki decision guide for the complete list.
The current manifests are: agentops-evaluation, ai-transformation-governance,
all, analytics-finance, backend-api, business-analysis,
creator-ai-production, creator-analytics-reporting, creator-brand-deals,
creator-business-ops, creator-community, creator-content-engine,
creator-monetization, creator-reputation-risk,
creator-youtube-local-production, data-analytics-bi,
devops-cloud-release, engineering-delivery, executive-command-center,
founder-fundraising-ir, frontend-product, game-dev,
html5-game-publishing, llm-skill-authoring, operating-cadence,
owner-operator-os, product-research, quality-testing,
research-validation, revenue-growth, sales-marketing,
skill-security-audit, support-success, and technical-documentation.
- Atomic skills under
skills/ - Curated manifests under
skillsets/ - MCP presets and server snippets under
mcp/ - Routing templates under
agents/ - Host and context starters under
templates/ - Install, validation, listing, and release helpers under
scripts/ - Scenario tests and expected routing under
tests/ - Setup, authoring, evaluation, security, and release docs under
docs/
The current default-branch development scope validates:
- 167 skill files
- 34 skillsets
- 124 routing scenarios
- all skillset install dry-runs
- fresh Codex home smoke tests for
game-devandall - docs consistency, secret, and local-path scans
Run the full local gate before publishing:
.\scripts\release-check.ps1Useful focused checks:
python scripts\validate-skill-files.py --repo-root .
python scripts\validate-skillsets.py --repo-root .
python scripts\validate-scenarios.py --repo-root .
python scripts\validate-docs.py --repo-root .
gh skill publish --dry-runSkills are instructions and may include scripts, references, and MCP configuration. Preview them before installation. Keep public publishing, account changes, production changes, customer communication, legal or financial decisions, and other high-impact actions human-reviewed unless you have explicitly approved the action.
AgentSkills favors portable workflow logic, explicit output contracts, scenario tests, and reviewable safety boundaries. Read vendor neutrality before adding host-specific behavior.
- Setup guides
- Examples
- Compatibility
- Troubleshooting
- Authoring guide
- Bundle authoring
- Evaluation
- Release process
- Changelog
- Contributing
- Security policy
- Code of Conduct
The AgentSkills Wiki is the long-form orientation layer for choosing workflows, installing across hosts, understanding portability, and navigating the full catalog.
Keep changes small, reviewable, and vendor-neutral unless they belong in a host adapter. Good contributions include focused skills, realistic scenario tests, setup improvements, validation checks, and bundle proposals with clear users, artifacts, and safety boundaries.
Run .\scripts\release-check.ps1 before opening a pull request. See
CONTRIBUTING.md for the contribution contract.
AgentSkills is available under the MIT License.
If AgentSkills is useful to your workflow, sponsor the project or star the repository.