vibe-engineering

33 engineering discipline skills for Claude Code, OpenAI Codex & Gemini CLI + a CLI for CI/CD enforcement. Extracted from real-world multi-agent system development: not theoretical best practices, but patterns that survived 3 weeks of intensive production development with 205+ test files, 11 agents, and 50+ session observations.

"Vibe coding" meets engineering rigor. Every skill here exists because skipping it caused real pain, and every skill was re-checked in v2.0 against what current models and agent harnesses now handle on their own.

Two commands. Instant engineering discipline.

/plugin marketplace add ash1794/vibe-engineering
/plugin install vibe-engineering@vibe-plugins

Your agent picks the right skill from its description: research before design, quality gates before shipping, evidence before claiming "done."

What's new in v2.3

• vibe-cli works on macOS. It runs under the stock bash 3.2, and its scans no longer pass silently with BSD grep, which used to report CLEAN with a secret staged. --json output is now valid on every command. CI checks that a staged secret is blocked.
• Sharper skill triggers: nine skills now say when to use them in their descriptions, so agents pick them more reliably.

Earlier releases:

• v2.2 added vibe-slop-filter for prose that reads as machine-written, and vibe-devil-advocate-review now reviews any artifact from a named senior expert's standards, reporting only evidenced findings.
• v2.1 added vibe-workstream-orchestration, vibe-publication-leak-guard, and vibe-flake-root-cause from a multi-agent production build, plus panel mode, deploy provenance, wind-down handovers, and performance budgets in existing skills.
• v2.0 re-audited every skill against current Claude, GPT, and Gemini models and consolidated 38 skills into 29. If you're upgrading from v1.x, read the migration notes.

Full history: CHANGELOG · Releases

What is this?

A skill collection in the open Agent Skills format with 33 skills that enforce engineering discipline across any project, plus a lightweight CLI (vibe-cli) for CI/CD pipelines and automation:

• Research & Decision-Making (3 skills): think before building
• Quality Gates & Validation (8 skills): catch issues before they ship
• Knowledge & Continuity (2 skills): never solve the same problem twice
• Parallel & Multi-Agent Development (3 skills): scale your work safely
• Testing Patterns (6 skills): test what matters, not just what's easy
• Deployment & Operations (5 skills): ship with confidence
• Gap Analysis & Remediation (2 skills): audit and systematically close gaps
• Process (3 skills): stay focused, build to the right standard
• Meta (1 skill): skill routing

What the skills don't do

Modern models and harnesses already handle some things well, and a skill that duplicates them only adds noise. v2.0 removed skills for context-window management (harnesses compact automatically), output formatting (models format well by default), always-on skill bootstrapping (skills trigger from their descriptions), and private task queues (use your issue tracker). See CHANGELOG for the full list and migration.

Installation

Claude Code

From inside a Claude Code session: