Vibes

Transforms Claude Desktop into a conversational development environment using distributed MCP servers.

vibes

A multi-platform skill and agent framework for AI coding assistants.

What is vibes?

Vibes provides a shared set of skills, agents, and reference documentation for AI coding assistants. Write your agent instructions once in AGENTS.md, and every supported platform — Claude Code, GitHub Copilot, Codex CLI — gets the same guidance through symlinks and platform-specific configuration.

Skills teach agents how to perform specific tasks (fetch library docs, manage GitHub PRs, review code). Agents define roles with scoped tools and system prompts. MCP servers provide live documentation so agents work from current APIs instead of stale training data.

Quick Start

git clone https://github.com/jonhill90/vibes.git
cd vibes

Open the repo in your editor. The platform reads its instruction file automatically — no build step, no install.

MCP Servers

Three MCP servers provide live documentation. They are pre-configured in .mcp.json (Claude Code), .vscode/mcp.json (VS Code), and .codex/config.toml (Codex CLI). No API keys required.

ServerPurpose
context7Library and framework docs (npm, PyPI, crates, etc.)
microsoft-learnAzure, .NET, M365, and Microsoft docs
deepwikiGitHub repository wikis and documentation

Platform Setup

PlatformInstruction FileMCP Config
Claude CodeCLAUDE.md (symlink → AGENTS.md).mcp.json
GitHub Copilot.github/copilot-instructions.md (symlink → AGENTS.md).vscode/mcp.json
Codex CLIAGENTS.md (read natively).codex/config.toml

For Codex skill discovery, this repo exposes .agents/skills (symlink → .github/skills).

Skills

Skills are invocable instruction sets that teach an agent how to perform a specific task. Each lives in .github/skills/<name>/SKILL.md.

Documentation

SkillTriggerDescription
primer/primerOrient in any codebase — structure, docs, key files, current state
context7/context7Fetch live library/framework docs via MCP (Python, Bash, PowerShell variants)
ms-learn/ms-learnQuery official Microsoft docs for Azure, .NET, M365

Platform Integration

SkillTriggerDescription
gh-cli/gh-cliManage GitHub via CLI — PRs, issues, workflows, releases
az-devops/az-devopsManage Azure DevOps — repos, pipelines, boards, work items
linear/linearManage Linear via CLI — issues, teams, projects, Git workflow
tmux/tmuxRun persistent interactive CLI sessions via tmux for agent tooling and REPLs
obsidian/obsidianRead, write, search, and manage Obsidian vault notes
youtube-transcript/youtube-transcriptFetch YouTube video transcripts and metadata

Authoring & Validation

SkillTriggerDescription
create-skill/create-skillGuide for creating new skills with scripts and references
validate-skill/validate-skillValidate a SKILL.md against the spec
lint-agents/lint-agentsCheck agent definition files for correct YAML frontmatter

Agents

Agents are specialized roles with scoped tools and a system prompt. Defined in .github/agents/<name>.md.

AgentPurposeTools
code-reviewerCode quality and security review after changesRead, Grep, Glob, Bash
researcherCodebase investigation and pattern analysisRead, Grep, Glob

Architecture

graph TD
    AGENTS["AGENTS.md<br/><i>Single source of truth</i>"]

    AGENTS -->|symlink| CLAUDE["CLAUDE.md<br/>Claude Code"]
    AGENTS -->|symlink| COPILOT[".github/copilot-instructions.md<br/>GitHub Copilot"]
    AGENTS -->|reads natively| CODEX["Codex CLI"]

    GH[".github/<br/>skills/ agents/ docs/ instructions/"]
    GH -->|symlink| CLAUDE_DIR[".claude/<br/>skills/ agents/ hooks/ references/ rules/"]
    GH -->|symlink| CODEX_DIR[".codex/<br/>agents/ config.toml"]
    GH -->|symlink| AGENTS_DIR[".agents/<br/>skills/"]

    MCP["MCP Servers<br/>context7 · microsoft-learn · deepwiki"]
    MCP -.->|live docs| CLAUDE
    MCP -.->|live docs| COPILOT
    MCP -.->|live docs| CODEX

Key decisions:

  • .github/ is the single source. Platform directories (.claude/, .codex/, .agents/) contain symlinks and platform-specific config/policy.
  • One instruction file for all platforms. AGENTS.md is symlinked so every tool reads the same guidance.
  • Progressive disclosure. Skill metadata (name + description) is always in context. The SKILL.md body loads on trigger. Reference files load on demand.
  • Codex uses native surfaces. Skills are discovered via .agents/skills/; runtime policy/config lives under .codex/ (config.toml, rules/*.rules).
  • Platform-specific rules stay separate. .claude/rules/ and .github/instructions/ use their own formats rather than a forced shared one.
  • This repo's hooks always exit 0. The three scripts in .claude/hooks/ use non-blocking warnings (systemMessage) rather than exit 2 blocking. Frontmatter and symlink checks warn Claude after edits; transcript backup runs silently on compaction.

Reference Documentation

Platform-agnostic docs live in .github/docs/ and are symlinked into .claude/references/.

DocumentPathTopic
Best Practices.github/docs/best-practices.mdOperational patterns for agent development
Context Engineering.github/docs/context-engineering.mdToken budget management, progressive disclosure
TDD Workflow.github/docs/tdd-workflow.mdRed-Green-Refactor process for agents

Workflow

Explore → Plan → Red → Green → Refactor → Commit
  1. Explore — Run /primer to orient, then dig deeper as needed
  2. Plan — Surface tradeoffs and get alignment
  3. Red — Write failing tests that define success
  4. Green — Write minimum code to pass
  5. Refactor — Clean up while tests stay green
  6. Commit — Clean, descriptive messages

For non-code changes (docs, config), skip Red/Green/Refactor and go from Plan to Commit. See AGENTS.md for the full instruction set.

Adding Skills and Agents

New skill:

mkdir -p .github/skills/my-skill
# Create SKILL.md with name + description frontmatter
# Validate: /validate-skill .github/skills/my-skill

New agent:

# Create .github/agents/my-agent.md
# YAML frontmatter (name, description, tools) + markdown system prompt
# Validate: /lint-agents

See /create-skill for the full skill authoring guide and AGENTS.md for naming conventions.

Further Reading

Related Servers