← workbenchInstall

Any agent, or a plugin

Skills are the Agent Skills open format (<name>/SKILL.md) and AGENTS.md is cross-tool — nothing here is locked to one agent.

<([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])> <([{<([{<()>}])>}])>

Tool-agnostic — npx skills

Claude Code, Codex, Cursor, Copilot, Windsurf, and ~30 more. The skills CLI drops the SKILL.md files into each agent's own location.

# the Agent Skills open format — installs into whichever agent(s) you pick
npx skills@latest add tejastn10/workbench

# Codex also: point it at the conventions
ln -sf ~/workbench/AGENTS.md ~/.codex/AGENTS.md

Claude Code plugin

Managed & versioned — skills + the commands/ slash commands + the .mcp.json servers in one bundle.

# Claude Code — skills + slash commands + MCP servers, managed & updatable
claude plugins marketplace add tejastn10/workbench
claude plugins install workbench@tejastn10
claude plugins update workbench          # later

Details: docs/plugin.md

Or clone & symlink

Editable skills, git pull to update.

# clone
git clone git@github.com:tejastn10/workbench.git ~/workbench

# link every skill directory -> ~/.claude/skills/<name>  (editable; git pull to update)
~/workbench/scripts/install-skills.sh
~/workbench/scripts/install-skills.sh --list          # what's available
~/workbench/scripts/install-skills.sh --project DIR   # copy into a repo's .claude + .github

# point global instructions at the conventions
ln -sf ~/workbench/AGENTS.md ~/.codex/AGENTS.md
# and reference ~/workbench/AGENTS.md from ~/.claude/CLAUDE.md

Per agent

Claude Code

Install the plugin (skills + commands + MCP), or symlink skill dirs with install-skills.sh. git pull / claude plugins update to refresh.

VS Code agent mode

Reads the same <name>/SKILL.md format from the user profile or .github/. AGENTS.md is read natively. MCP via .vscode/mcp.json.

Codex

Reads AGENTS.md natively (root + nested). No skills directory — reference a skill's path from AGENTS.md, or paste it in. MCP via ~/.codex/config.toml.

Per project, once

Copy CONTEXT.md into the project and fill in the glossary, stack, and landmines.
Add a project .out-of-scope/ as decisions get ruled out.
If the project's rules differ from AGENTS.md, note the deviation in its CONTEXT.md.

Build your own review skill

The review skills are generated from a real corpus of your PR comments, not best-practice lists. On the machine with repo access, run the distill-review-style skill — it pulls your comments with gh, categorises what you flag / let slide / how you phrase it, and emits a new pr-review/<stack>-pr-review.md.

Full walkthrough: docs/USAGE.md