Start Here

This is the shortest guided path through the repo. Install the catalog, try one skill, then use OpenSpec when a change needs a durable proposal, task list, and validation trail.

What a skill is

Skills are portable prompt-and-workflow bundles for coding agents. Install them once, then call them with /skill-name inside Claude Code, Gemini CLI, OpenCode, Cursor, GitHub Copilot, Codex, and other agentskills.io-compatible agents.

Workflow MapFrom repository bundle to install, invocation, and validation across supported agent harnesses.

1. Install The Catalog

Install the full catalog

Globally installs repo skills across supported agent harnesses.

Copy

npx skills add github:wyattowalsh/agents --all -y -g --agent antigravity --agent claude-code --agent codex --agent crush --agent cursor --agent gemini-cli --agent github-copilot --agent grok --agent opencode

Install one starter skill

Use this when you want to test the workflow before installing everything.

Copy

npx skills add github:wyattowalsh/agents --skill honest-review -y -g --agent antigravity --agent claude-code --agent codex --agent crush --agent cursor --agent gemini-cli --agent github-copilot --agent grok --agent opencode

2. Choose The Packaging Path

Claude Code Plugin Use the native .claude-plugin marketplace when you want Claude Code to install and update the whole repo bundle.

Codex Plugin Use the native .codex-plugin manifest and repo marketplace when you want Codex to load bundled skills and MCP config.

OpenCode Project Config Use repo opencode.json for runtime plugins; keep TUI-only plugins in your live tui.json surface.

Skills CLI Use npx skills add for Gemini CLI, Antigravity, Copilot, OpenCode, Cursor, Crush, and other agents.

Run wagents update or npx skills update after upstream changes to refresh recorded sources, and use wagents skills sync --dry-run before --apply when you want additive cross-harness reconciliation for repo + curated external skills.

OpenCode plugin boundary

OpenCode runtime plugins are tracked in repo opencode.json; TUI-only plugins and shortcuts live in ~/.config/opencode/tui.json. Current OpenCode shortcut config uses keybinds, not stale keymap.sections. Plannotator stays scoped to the plan agent, while scheduler jobs, Claude credential reuse, Langfuse credentials, and local workflow-state plugins require explicit opt-in instead of being enabled by repo defaults.

Grok Build Plannotator

Grok has no npm plugin like OpenCode’s @plannotator/opencode. Run wagents grok plannotator install once for the CLI and curated skills (extras on by default). Home sync keeps hooks and overlays enabled by default; opt out with --no-plannotator-hooks on sync_agent_stack.py. Verify with wagents grok doctor.

OpenCode session recovery

When one OpenCode session loops on OpenAI server_error but fresh openai/gpt-5.5 pure and plugin-loaded runs succeed, diagnose the session before changing model defaults. Preserve request IDs, inspect the session DB for incomplete assistant turns and encrypted reasoning parts, back up ~/.local/share/opencode/opencode.db, and archive the affected session instead of repeatedly resuming it.

3. Try A Starter Skill

honest-review Start here if you want stronger code review, bug-finding, and risk assessment.

wargame Use this when the problem is strategic and you want to compare paths before building.

mcp-creator Start here when you need to build a new FastMCP server or expose tools cleanly.

frontend-designer Use this when the work is user-facing and you need stronger UI implementation decisions.

4. Use OpenSpec For Non-Trivial Changes

Run the repo wrapper before changing public asset formats, downstream tooling, generated docs, validation behavior, or multiple coordinated surfaces:

uv run wagents openspec doctor

To materialize upstream OpenSpec skills and commands for supported downstream tools, run:

uv run wagents openspec init --apply

Generated .claude, .cursor, .opencode, .github, .agent, .crush, .codex, and .gemini OpenSpec artifacts are local/generated unless a specific file is promoted to repo-owned source.

5. Invoke It Inside Your Agent

1. Open your supported coding agent.

2. Call the skill with /skill-name.

   Example: /honest-review

3. Give it the concrete task or target.

   Example: /honest-review review this diff for correctness and regressions

6. Use The Right Docs Path

Skills Browse the catalog when you know the kind of specialist you need but not the exact skill name.

Contributing Use this before changing skills, agents, MCP config, docs generators, external skill metadata, or distribution files.

OpenSpec Workflow Use this when repo or downstream-tooling work needs a spec/change workflow.

CLI Reference Use the wagents CLI to scaffold assets, validate the repo, package skills, and regenerate docs.

MCP Overview Use this when you need the configured MCP landscape, client sync coverage, or example client snippets.

GitHub Use the repo directly when you want the raw source, issues, or contribution flow.

7. Know What Is Generated

Generated vs curated docs

Most pages in this site are generated from repository assets via wagents docs generate. OpenSpec project state lives under openspec/. This page is hand-maintained on purpose so the onboarding path stays short, stable, and easier to refine than the generated catalogs.