🦸 Super Agent

Install and run AI-powered developer skills across Claude, Gemini, and Google Antigravity.

⚡ Turn prompts into reusable, installable skills ⚡ Works from CLI in seconds ⚡ Built for developers

🚀 Quick Start

npx @workingpayload/agent-skills install

✨ Features

• 🧠 55 expert-level developer skills with named tools, concrete steps, and edge case coverage
• 🤖 Multi-LLM support (Claude, Gemini, Google Antigravity)
• 📦 CLI-first workflow with install, route, doctor, and score commands
• 🔄 Versioning & health checks for installed skills
• 🎯 Selective install — pick only the skills you need
• 🧩 File-based skill architecture with YAML frontmatter
• 🧪 Quality tooling — linter, auto-scorer, adversarial test harness
• 📊 Machine-readable skill index for programmatic routing

📦 Installation

Use instantly (recommended)

npx @workingpayload/agent-skills install

Or install globally

npm install -g @workingpayload/agent-skills

Install specific skills only

npx @workingpayload/agent-skills install -s CodeSage,TestCrafter,BugHunter Pro

Install for Gemini

npx @workingpayload/agent-skills install -t gemini

Install for Google Antigravity

npx @workingpayload/agent-skills install -t antigravity

Skills are installed as directory-based packages (SKILL.md + optional scripts/references) to ~/.gemini/antigravity/skills/. The Antigravity agent automatically matches skills to your tasks via semantic triggering.

🧪 Usage

Use skills in Claude Code

After installing, invoke skills as slash commands:

/codesage          → "Review this function for security issues"
/bughunter-pro     → "Debug this failing test"
/testcrafter       → "Write unit tests for UserService"
/python-master     → "Fix this async race condition"
/react-ranger      → "Build a search component with debounce"
/fastapi-furious   → "Create a CRUD API for products"
/dockmaster        → "Optimize my Dockerfile"
/kubecrafter       → "Write a Deployment manifest with HPA"

Use skills in Google Antigravity

After installing with -t antigravity, skills are available automatically via semantic triggering — no slash commands needed. The Antigravity agent reads skill descriptions and matches them to your task:

"Review this function for security issues"   → agent activates CodeSage skill
"Debug this failing test"                    → agent activates BugHunter Pro skill
"Generate a codemap for this repo"           → agent activates CodeMap skill

Skills are installed as directories under ~/.gemini/antigravity/skills/, each containing a SKILL.md and optional scripts/, references/, and assets/ subdirectories.

Find the right skill

npx @workingpayload/agent-skills route "review my Python code"
# → codesage (score: 12), python-master (score: 12)

npx @workingpayload/agent-skills route "design database schema"
# → dataweaver (score: 20), modelsensei (score: 20)

🧠 CLI Commands

| Command | Description | | --- | --- | | install | Install skills to Claude or Gemini | | uninstall <names> | Remove installed skills | | list | List installed skills with timestamps | | route "<prompt>" | Find the best skill for a task | | doctor | Check health of installed skills | | validate | Lint all skill files | | score | Auto-score skills on quality metrics | | test | Run adversarial tests on all skills | | build-index | Regenerate skills.json routing index |

⚙️ Options

| Flag | Description | | --- | --- | | -t, --target | Target platform: claude, gemini, or antigravity | | -s, --skills | Comma-separated skill names to install | | --force | Overwrite existing skills | | --dry-run | Preview without writing files | | --validate | Validate skills before installing | | --json | Machine-readable output (score, route) | | --interactive | Interactive prompt mode (route) | | -n, --top-n | Number of routing results (default: 3) |

🗂️ Where Skills Are Installed

~/.claude/commands/                # Claude Code slash commands
~/.gemini/skills/                  # Gemini CLI skills (directory-based)
~/.gemini/antigravity/skills/      # Google Antigravity skills (directory-based)

Each skill becomes a slash command (Claude/Gemini) or a directory-based skill (Antigravity):

~/.claude/commands/codesage.md                  → /codesage (Claude Code)
~/.gemini/skills/codesage/SKILL.md              → auto-activated by Gemini CLI
~/.gemini/antigravity/skills/codesage/SKILL.md  → auto-triggered by Antigravity agent

🩺 Health Checks

npx @workingpayload/agent-skills doctor

Checks:

• Install directory exists
• Meta file is valid
• All tracked skills are present on disk
• Installed skills match source (stale detection)
• No orphan files
• Frontmatter is valid

📊 Quality Tooling

# Lint all skills (CI gate)
npx @workingpayload/agent-skills validate

# Auto-score on 6 measurable criteria (0-30 scale)
npx @workingpayload/agent-skills score

# Run adversarial tests
npx @workingpayload/agent-skills test

# Generate machine-readable index
npx @workingpayload/agent-skills build-index

Scoring Criteria

| Criterion | What it measures | | --- | --- | | Tool Coverage | Named tools/commands in backticks | | Code Examples | Fenced code blocks with real examples | | Edge Cases | Real-world failure scenarios covered | | Actionability | Concrete numbered workflow steps | | Structure | Frontmatter, overview, workflow, output format | | Conciseness | Optimal line count (80-150 lines) |

🧩 Skill Format

Every skill file follows this structure:

---
name: codesage
description: Performs structured code reviews with severity-tiered findings...
---

# CodeSage

## Overview
Expert code review covering security, correctness, performance...

## Workflow
### 1. Identify Language and Context
- Detect language, framework, runtime...

### 2. Static Analysis Pass
Run: `ruff check`, `mypy --strict`, `bandit -r .`...

## Edge Cases
1. Auto-generated code (Prisma, OpenAPI) — don't flag style...
2. False positive from linter — verify before reporting...

## Output Format
- Severity tier (CRITICAL/HIGH/MEDIUM/LOW/NOTE)
- Code snippet with fix
- Verification command

🧠 Skills Catalog

Code Quality

| Skill | Command | Description | | --- | --- | --- | | CodeSage | /codesage | Structured code reviews with severity tiers, OWASP security checks, and language-specific linters | | RefactorX | /refactorx | Safe refactoring using Fowler's patterns, strangler fig, and cyclomatic complexity reduction | | LogicLens | /logiclens | Analyzes complex code for cyclomatic complexity, then refactors with guard clauses and strategy pattern | | SpeedSmith | /speedsmith | Performance optimization: caching, code splitting, image optimization, bundle size CI gating | | FunctionForge | /functionforge | Pure, testable utility functions using FP principles — immutability, Result types, pipe/compose |

Testing

| Skill | Command | Description | | --- | --- | --- | | TestCrafter | /testcrafter | Unit/integration tests with Jest, Vitest, Pytest, JUnit using AAA pattern and test pyramid | | FlowTester | /flowtester | E2E tests with Playwright/Cypress, Page Object Model, and CI-friendly configuration | | CoverageMax | /coveragemax | Identifies untested paths, generates tests to reach 80%+ coverage with pytest-cov, Jest, Istanbul | | EdgeGuard | /edgeguard | Finds edge cases using boundary analysis, property-based testing, fuzzing, and mutation testing | | MockSmith | /mocksmith | Schema-driven mock data with Faker.js/factory_boy, MSW API stubs, and edge-case datasets |

Security

| Skill | Command | Description | | --- | --- | --- | | SecureScan | /securescan | SAST security auditing: OWASP Top 10, CWE mapping, Semgrep/Bandit/CodeQL | | SecretSniffer | /secretsniffer | Scans codebases and git history for hardcoded credentials with TruffleHog/gitleaks | | InputShield | /inputshield | Input validation and sanitization: Zod/Joi/Pydantic, XSS/SQLi prevention, CSP headers | | AuthCraft | /authcraft | Auth systems: OAuth 2.0/PKCE, JWT lifecycle, bcrypt/Argon2, OWASP ASVS compliance |

DevOps

| Skill | Command | Description | | --- | --- | --- | | DockMaster | /dockmaster | Optimized Dockerfiles with multi-stage builds, Trivy scanning, Compose, and Helm charts | | KubeCrafter | /kubecrafter | Production-ready K8s manifests: Deployments, HPA, StatefulSets, Pod Security Standards | | PipelinePro | /pipelinepro | CI/CD with GitHub Actions/GitLab CI, OIDC auth, blue-green/canary deploys, rollback | | EnvWizard | /envwizard | Secure env var management: .env.example, startup validation, vault integration | | CronGenius | /crongenius | Cron expressions, distributed job scheduling, Celery Beat, systemd timers, EventBridge |

Frontend

| Skill | Command | Description | | --- | --- | --- | | UISmith | /uismith | Accessible UI components: Atomic Design, WCAG AA, ARIA, design tokens, RTL support | | FlexiLayout | /flexilayout | CSS layout debugging: Flexbox/Grid, container queries, fluid typography, responsive design | | RenderBoost | /renderboost | Core Web Vitals optimization: LCP, CLS, INP, lazy loading, Service Worker caching | | SEOMancer | /seomancer | Technical SEO: JSON-LD, meta tags, Open Graph, canonical/hreflang, site migration | | AccessPlus | /accessplus | RBAC/ABAC authorization + WCAG 2.1 accessibility with screen reader testing |

Database

| Skill | Command | Description | | --- | --- | --- | | DataWeaver | /dataweaver | Schema design (1NF-3NF), EXPLAIN ANALYZE, reversible migrations with Alembic/Flyway/Prisma | | QueryPulse | /querypulse | SQL optimization: index design, N+1 detection, deadlock diagnosis, CTE materialization | | DataPolish | /datapolish | Data cleaning with pandas/polars/Great Expectations: profiling, dedup, type coercion | | ModelSensei | /modelsensei | DDD data models, ORM mappings, N+1 prevention, outbox pattern, hierarchical data |

Architecture

| Skill | Command | Description | | --- | --- | --- | | Architekt AI | /architekt-ai | System design: C4 model, ADRs, CQRS, event sourcing, zero-downtime DB migrations | | ServiceSplitter | /servicesplitter | Monolith-to-microservices: DDD bounded contexts, event storming, strangler fig, sagas | | StackWise | /stackwise | Tech stack evaluation with weighted rubric, ADR output, library health metrics | | TaskOrchestrator | /taskorchestrator | Async task queues: Celery/BullMQ/Temporal, dead-letter queues, idempotency, backpressure |

Documentation

| Skill | Command | Description | | --- | --- | --- | | DocuGenie | /docugenie | OpenAPI 3.x specs, JSDoc/TSDoc/Google-style docstrings, and developer guides | | ReadMeCraft | /readmecraft | README files: standard-readme spec, badges, install/usage sections, multilingual | | ChangeScribe | /changescribe | Changelogs: Keep a Changelog format, Conventional Commits, semver bump rules |

Git & Workflow

| Skill | Command | Description | | --- | --- | --- | | CommitCraft | /commitcraft | Conventional Commits v1.0.0 messages from git diff analysis | | PRMentor | /prmentor | PR reviews with severity tiers, scope checks, IaC review, and merge readiness checklist | | IssueForge | /issueforge | Structured bug reports and feature requests for GitHub Issues and Jira |

Debugging & Observability

| Skill | Command | Description | | --- | --- | --- | | BugHunter Pro | /bughunter-pro | Hypothesis-driven debugging with reproduction, isolation, and regression tests | | ErrorMedic | /errormedic | Stack trace diagnosis, error taxonomy, source map correlation, race condition fixes | | TraceMind | /tracemind | Distributed tracing with OpenTelemetry, structured logging, SLO-based alerting | | PerfDetective | /perfdetective | Performance profiling: flamegraphs, py-spy, async-profiler, GC pause analysis |

Language-Specific

| Skill | Command | Description | | --- | --- | --- | | Python Master | /python-master | Full Python lifecycle: ruff, mypy, pytest, uv/poetry, memory leak diagnosis | | React Ranger | /react-ranger | React with TypeScript: hooks, RSC, Vitest/RTL, a11y, bundle analysis | | FastAPI Furious | /fastapi-furious | FastAPI: project scaffold, Alembic, WebSockets, file uploads, BackgroundTasks | | Extension Extender | /extension-extender | Manifest V3 browser extensions: Chrome/Firefox, CSP, storage, service workers |

Other

| Skill | Command | Description | | --- | --- | --- | | AlgoMaster | /algomaster | Algorithm optimization: Big-O, DP, greedy, graph, data structure selection | | AutoScriptor | /autoscriptor | Automation scripts in Bash/Python/PowerShell with ShellCheck and dry-run mode | | BoilerCore | /boilercore | Project scaffolding with ecosystem conventions, CI/linter setup, Docker template | | FeatureSmith | /featuresmith | Requirement decomposition: user stories, story points, feature flags (LaunchDarkly/Unleash) | | PolyglotShift | /polyglotshift | Code translation between languages preserving idioms and type systems | | PromptForge | /promptforge | LLM prompt engineering: CoT, ReAct, few-shot, multi-modal, function-calling | | CodeMap | /codemap | Generates structured codemap files at checkpoints for instant AI agent codebase navigation | | ScrapeMaster | /scrapemaster | Web scraping: Playwright/Cheerio/BeautifulSoup, robots.txt, retry/backoff |

🤖 Supported Models

• Claude (Anthropic)
• Gemini (Google)
• Google Antigravity (agent-first IDE)

🧱 Create Your Own Skill

1. Create a directory with a skill.md file:

MySkill/skill.md

2. Add frontmatter and content:

---
name: myskill
description: Describe what this skill does in 50+ characters...
---

# MySkill

## Overview
What this skill specializes in...

## Workflow
### 1. First Step
Concrete instructions with named tools...

## Edge Cases
1. Real-world failure scenario...

## Output Format
How results should be structured...

3. Validate and install:

npx @workingpayload/agent-skills validate
npx @workingpayload/agent-skills install --force

💥 Why This Exists

Prompt engineering is repetitive.

Agent Skills turns prompts into:

• reusable skills with named tools and concrete steps
• installable CLI commands for Claude, Gemini, and Google Antigravity
• quality-gated, version-tracked AI workflows

🚀 Roadmap

• [x] 55 expert-level skills with edge case coverage
• [x] CLI with install, route, doctor, score commands
• [x] Quality tooling (linter, scorer, adversarial tests)
• [x] Machine-readable skill index (skills.json)
• [x] GitHub Actions CI pipeline
• [x] Google Antigravity IDE integration (directory-based skills with semantic triggering)
• [ ] Skill marketplace
• [ ] Web UI playground
• [ ] Skill chaining
• [ ] Embedding-based semantic routing

🤝 Contributing

PRs welcome!

1. Fork repo
2. Add or improve a skill
3. Run npx @workingpayload/agent-skills validate to check quality
4. Submit PR — CI will score your skill automatically