vibestack

47 opinionated AI coding workflows. One install. Works in Claude Code, Cursor, and Kiro.

vibestack is a portable skill pack for AI coding agents. Slash commands like
/office-hours, /ship, /investigate, /tdd, /review install once and
work across every agent that supports the Agent Skills open
standard — Claude Code, Cursor, Kiro,
and a growing list of others. Same SKILL.md source, three folders, no
vendor lock-in.

Try it in 30 seconds

git clone https://github.com/timurgaleev/vibestack ~/.claude/skills/vibestack
~/.claude/skills/vibestack/install

Interactive — ./install shows a write plan and installs into detected
agents on Enter (or a for all, e to choose per-target). Open a new
session of whichever agent you chose, type /office-hours, and you'll
see this:

LEARNINGS: none yet

Before we dig in — what's your goal with this?

  Building a startup (or thinking about it)
  Intrapreneurship — internal project at a company, need to ship fast
  Hackathon / demo — time-boxed, need to impress
  Open source / research — building for a community or exploring an idea
  Learning — teaching yourself to code, vibe coding, leveling up

Pick a mode and /office-hours walks you through targeted prompts — six
forcing questions for startup mode, design-thinking flow for builder
mode. The output is a saved design doc you can hand to /plan-eng-review
next.

This is the shape of every skill in vibestack: opinionated, structured,
no LLM-flavored mush. If /office-hours clicks, the other 45 will too.

Why vibestack?

• Multi-agent, no lock-in. Built on the Agent Skills open
  standard. The same skill files
  install into Claude Code (~/.claude/skills/), Cursor
  (~/.cursor/skills/), and Kiro (~/.kiro/skills/). Switch agents
  without re-learning your workflow.
• Opinionated, not curated. Not an "awesome list" of community
  prompts. Every skill ships an explicit workflow, output template, and
  cross-model review pattern. The opinions come from real shipping
  practice, not vibes.
• Composable. Skills chain into each other: /office-hours →
  /plan-eng-review → /tdd → /review → /ship. The chains are
  documented; you don't have to invent them.
• Local-first. No telemetry, no remote logging, no analytics
  endpoint. State lives in ~/.vibestack/. Your learnings, designs,
  and reviews stay on your machine.
• Boring tech. Bash + awk for the install pipeline. No template
  engines, no runtime dependencies. git pull && ./install is the
  whole update story.

How vibestack compares

Install options

./install                          # Interactive: asks per target (recommended)
./install --target=all             # All three, non-interactive
./install --target=claude          # Claude Code only
./install --target=cursor,kiro     # Pick a subset
./install --yes                    # All three, skip prompts (CI-friendly)
./install --dry-run                # Preview every output, write nothing

vibestack works with any agent that implements the Agent Skills standard.
v1.4 ships native install paths for three:

Codex CLI, Gemini CLI, OpenCode, Antigravity, and Windsurf also support
the spec via .agents/skills/ — adding them as native targets is
tracked for v1.5+.

What ./install modifies on your machine

./install is idempotent — re-runs produce identical bytes. To remove
everything, see Uninstall.

Cross-target compatibility (verified)

The Agent Skills standard guarantees portability of the SKILL.md file
shape. Claude-Code-specific runtime extensions (the ${CLAUDE_SKILL_DIR}
env var, per-skill PreToolUse hooks, the Agent and AskUserQuestion
tools) are NOT covered by the spec.

For 43 of 47 skills (pure-prose workflows), this is fine — modern LLMs
map "ask the user" or "dispatch a subagent" to whatever native equivalent
exists in the host agent. Empirically verified across all three targets.

For the 4 hook-bearing safety skills (/careful, /freeze, /guard,
/investigate), the SKILL.md installs into Cursor and Kiro but their
hooks do not fire identically to Claude Code. Verified against
Cursor 2026.05.07-42ddaca and Kiro CLI 2.2.2:

The install prints a one-line warning when hook-bearing skills are
installed into Cursor or Kiro. Full empirical results:
docs/agent-skills-compatibility-audit.md.
Re-verification procedure for future versions:
docs/hook-verification.md.

⚠️ If you rely on /careful, /freeze, /guard, or /investigate
as a real safety net (not just an LLM nudge), use them in Claude Code.
Cursor gives you partial protection via its own sandbox; Kiro gives none.

Uninstall

~/.claude/skills/vibestack/uninstall                              # Claude only
~/.claude/skills/vibestack/uninstall --target=all                 # All three
~/.claude/skills/vibestack/uninstall --target=cursor              # Cursor only
~/.claude/skills/vibestack/uninstall --target=all --delete-state  # Full wipe

Mirrors ./install's --target= semantics. Removes the rendered
SKILL.md file, .vibe-render.json sidecar, bin/sub-doc symlinks,
and the per-skill directory (if empty) for each chosen target. Removes
the vibe-* binaries from ~/.vibestack/bin/. Asks before deleting
~/.vibestack/ (your local learnings, analytics, project state) —
keeps it by default. Pass --delete-state for a non-interactive
full state wipe.

Update

cd ~/.claude/skills/vibestack && git pull && ./install

vibestack distributes via git, no package manager. Pulling and re-running
the install updates every chosen target.

What's in the box

47 skills in seven categories. Full reference: docs/skills.md.

Product & Planning

Code Quality & Shipping

QA & Testing

Design

Operations

Session & Context

Safety & Scope Control

Tooling & Integrations

How skills work

Each skill is a SKILL.md file in your agent's skills directory. Your
agent discovers them automatically and makes them available as /name
commands. The install script writes the rendered file plus symlinks to
sub-docs and hook scripts — the source stays in this repo, so
git pull && ./install is all you need to update.

vibestack uses a small render step at install time
(bin/vibe-render-skill) to expand {{include lib/snippets/X.md}}
directives. This lets shared boilerplate live in one place
(lib/snippets/) and get composed into each skill at install. Source
files without include directives produce byte-identical output — no
behavior change for skill authors who don't use the feature.

Adding your own skills

mkdir -p skills/my-skill
cat > skills/my-skill/SKILL.md << 'EOF'
---
name: my-skill
description: |
  What this skill does.
allowed-tools:
  - Bash
  - Read
triggers:
  - trigger phrase
---

## Skill instructions here
EOF

./install

Then use /my-skill in your agent. See CONTRIBUTING.md
for the full contributor guide.

Data written locally

vibestack writes a small amount of state to your machine.
No data leaves your machine. vibestack has no telemetry, no
analytics endpoint, no remote logging.

Disabling:

• Skill invocation log: don't wire the vibe-skill-track hook, or set VIBESTACK_TRACK=0 if it is wired.
• Hook decision audit: simply don't set VIBESTACK_DEBUG=1.
• Learnings: don't run /learn, or delete ~/.vibestack/projects/<slug>/learnings.jsonl.

Wiring vibe-skill-track (optional skill-usage analytics — Claude Code only):

Add this entry to ~/.claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "~/.vibestack/bin/vibe-skill-track"
          }
        ]
      }
    ]
  }
}

Then ~/.vibestack/analytics/skill-usage.jsonl will record one line per
explicit /skill invocation. Useful as input to /devex-review and
/retro so you can see which skills you actually use.

Limitation (by design): this hook captures explicit /skill-name
invocations only. Skills that the LLM auto-invokes by description-match
without you typing / are not logged. Claude Code does not
currently expose a deterministic skill-start event. Cursor and Kiro
have their own hook frameworks; equivalent wiring is a v1.5+ candidate.

More

• ETHOS.md — five principles guiding skill design
• CHANGELOG.md — version history, including removed skills
• CONTRIBUTING.md — how to add a skill
• docs/skills.md — full skills reference with descriptions and triggers
• docs/agent-skills-compatibility-audit.md — per-agent compatibility matrix
• docs/hook-verification.md — manual hook verification procedure for Cursor/Kiro
• docs/external-tools.md — what vibestack does not bundle
• LICENSE — MIT

Star it, fork it, ship it

If vibestack saves you time, star the repo
— it's the simplest signal that opinionated workflows beat ad-hoc
prompting. Issues and PRs welcome.