A modular collection of skills and workflows for AI coding agents, designed to be harness- and model-agnostic.
The workflow system in vmode is inspired by the trigger-and-principle architecture of poteto-mode (from cursor/plugins), refined into an editor-agnostic, lightweight form without vendor lock-in, proprietary model couplings, or complex external dependencies.
- Python 3.10+ (Python 3.11 recommended). Used for collection verification (
setup-vstack/scripts/verify.py), installation tooling, and evaluation test suites. - Node.js & npx (Node 18+). Required to run
npx skills addfor installing skills into your agent environments. - Hunk (
hunk-reviewskill). Used for interactive local diff reviews before opening pull requests. See modem-dev/hunk. - Cua Driver (
cua-driverskill). Optional runtime dependency for native desktop GUI automation on macOS, Windows, and Linux. See trycua/cua. - GitHub CLI (
gh). Required for PR watchdog automation, checks polling, and issue publishing.
Install this repository with the open agent skills CLI:
npx skills add -g vuon9/vstackThe skills tool opens an interactive flow where you can choose which skills
and agents to install. The collection is managed and refreshed by
setup-vstack (skills.json is the
single source of truth for what gets installed).
A deterministic check validates the collection: every skills.json entry,
the ## Required skills list in vmode, and the
repo-local skills agree (name, source, scope, description, README).
python3 setup-vstack/scripts/manage.py verify # exit 0 when clean
python3 setup-vstack/scripts/install.py --dry-run # preview install actionsCI runs the same verification plus the unit test suite on every push and PR
(.github/workflows/validate.yml).
- Free to install: Using the first command you see in the Install part to choose skills and install whatever you wanted
- Wanted to try
vmode: It's my preferred way to start working on things, installvmodeand read its README to having more context
- vmode. The manual development-workflow mode (behavior-first, review and PR discipline, watchdog/general/research roles)
- setup-vstack. Install and refresh this collection from
skills.json - automate-me. Capture personal working conventions and preferences into an agent
-modeskill - shipping-darwin-apps. Build and ship iOS/macOS apps (signing, notarization, TestFlight, DMG)
- gh-workflows. Reusable GitHub Actions workflows
- bro. Restate the last message in plain language (as-is)
- code-teach. Explain a body of work, cursor's
teachrenamed (as-is) - how. Explain how a subsystem works (edited: Cursor-specific wording normalized)
- unslop. Remove AI tells from writing (as-is)
- why. Explain why a change (edited: Cursor-specific wording normalized)
- mattpocock/skills
- diagnosing-bugs
- grill-me
- grill-with-docs
- grilling
- handoff
- prototype
- teach
- to-prd
- to-spec
- wayfinder
- obra/superpowers
- brainstorming
- dispatching-parallel-agents
- executing-plans
- receiving-code-review
- requesting-code-review
- subagent-driven-development
- systematic-debugging
- test-driven-development
- verification-before-completion
- writing-plans
- writing-skills
- emilkowalski/skills
- apple-design
- vercel-labs/skills
- find-skills
- trycua/cua
- cua-driver
- modem-dev/hunk
- hunk-review
Each skill folder includes a README.md with human-facing use cases. SKILL.md
stays focused on instructions that agents should load at runtime.
MIT