@webpresso/agent-kit-sandbox
v0.21.4
Published
One command scaffolds a repo so every AI coding agent (Claude Code, Codex CLI, Cursor, Windsurf, Gemini, OpenCode) shares the same context, hooks, and quality gates. The `wp` CLI: blueprints, skills catalog, lore commits, tech-debt lifecycle, audit compos
Maintainers
Readme
@webpresso/agent-kit
Plug-and-play setup for AI coding agents. Run one command and every agent in the repo gets the same instructions, skills, hooks, planning files, and quality gates. MIT. Experimental v0.x.
Install
Requires Node.js 24 or newer.
Install from the public npm registry:
npm install -g @webpresso/agent-kit
wp setupThat's the product.
No private registry setup is required.
If you do not want a global install, run it one-shot instead:
npm exec --yes --package @webpresso/agent-kit@latest -- wp setupwp setup is safe to run again. It refreshes the webpresso-owned pieces and
preserves consumer-owned files.
What it gives you
- One repo brain — major coding agents read the same operating contract.
- Skills that travel — repo skills show up across supported agent surfaces.
- Hooks that help — generated hooks steer common work through repo quality gates.
- Blueprints by default — planning files and templates are ready when the task needs them, and Blueprint markdown stays the canonical plan while OMX handoff files remain derived metadata.
- Agent-friendly checks — tests, lint, typecheck, E2E, and audits are easy to run and cite.
- Context-efficient evidence by default —
wp_*MCP wrappers return compact test/lint/typecheck/audit summaries, andwp setupincludesrtkin its default workstation preset set.
Why it exists
AI-agent repos usually grow six copies of the same thing:
- one instruction file for Claude,
- another for Codex,
- another for Cursor,
- separate hooks,
- separate skills,
- separate planning conventions.
Those copies drift. webpresso makes the repo feel like one product again:
wp setupWhy agents keep more useful context
Coding agents waste context in two predictable ways: duplicated repo guidance and verbose tool output. Agent Kit attacks both:
- Default compact quality evidence:
wp_test,wp_lint,wp_typecheck,wp_qa,wp_e2e, andwp_auditare MCP-first wrappers that return summary-first JSON, clipped raw output, and budget metadata such asbytesandtokensSaved. Seedocs/qa-output.md. - Default RTK shell filtering lane:
wp setupincludesrtkin its default preset set. The setup command skips RTK in CI and whenWP_SKIP_RTK=1is set. Seedocs/add-ons.md. - Default context-mode lane:
wp setupincludescontext-modein its default preset set. The setup command skips context-mode in CI and whenWP_SKIP_CONTEXT_MODE=1is set, and the published package still does not bundle the external tool. Seedocs/migration/context-mode-default.md.
The result: agents spend more of the window on code, plans, decisions, and errors that matter, and less on repeated instructions or thousand-line command logs.
Add-ons
Most repos should start with the default setup. Extra integrations and their
default/opt-in behavior are documented in docs/add-ons.md.
Package references
If you need config subpaths or dependency references, use the appendix:
docs/markdown-fact-check.md.
Docs
Status
Experimental v0.x. Public APIs may change before v1.
License
MIT — see LICENSE. Vendored catalog skills and runtime integration licenses are documented in THIRD-PARTY-NOTICES.md.
