Workshop

Ready-to-use AI agents and skills for Claude Code, Codex, Cursor, and Google Antigravity, extracted from real projects and packaged as installable plugins. You get review agents that catch problems in specs, tests, and code, plus workflow skills for handoffs and reports.

Install

This repo doubles as a plugin marketplace: a catalog Claude Code can install plugins from directly.

In a Claude Code session:

/plugin marketplace add giostriquer/workshop
/plugin install workbench@workshop
/plugin install toolkit@workshop   # optional utilities

Codex install:

codex plugin marketplace add giostriquer/workshop
codex plugin add workbench@workshop
codex plugin add toolkit@workshop   # optional

For Cursor, the plugin ships in the Cursor plugin format (.cursor-plugin/). Import this repo as a Team Marketplace (Teams/Enterprise, admin): Dashboard → Settings → Plugins → Team Marketplaces → Add Marketplace → Import from Repo, point it at giostriquer/workshop, then install workbench (and optionally toolkit) from Customize in the sidebar.

For Google Antigravity, each plugin folder carries a native plugin.json manifest, which is what makes it installable. Antigravity discovers plugins by scanning two locations, so copy or link plugins/workbench (and optionally plugins/toolkit) into either: your workspace's .agents/plugins/ directory, for that workspace only, or ~/.gemini/config/plugins/, for every workspace. There is no registry file: a plugin folder is found because it sits in a scanned directory.

Global Rules Adoption - Optional

To install the workshop's shipped global agent configuration (its CLAUDE.md, AGENTS.md, rules, and Claude output styles) on a machine, additively, without overwriting what is already there: run /adopt-global-rules after installing toolkit. On a machine with no plugin installed, the same installer runs standalone:

npx github:giostriquer/workshop --dry-run   # plan; drop --dry-run to apply

The plugins

workbench: use right away

Five read-only review agents, nine everyday skills, and the seven-skill workbench process layer: ready immediately after install, nothing to configure.

Agents: they inspect and report, never edit your files:

Agent	Reviews
spec-reviewer	design specs and plans
code-quality-reviewer	a diff's maintainability and structure
test-quality-reviewer	test code
pattern-reviewer	code-pattern conformance
ci-watcher	the branch's PR CI

Everyday skills:

Skill	Does
file-pr	files the branch's PR, tends it to green-and-mergeable
fix-ci	watches CI, fixes red in-session
handoff-goal	hands a defined goal to a fresh autonomous session
claim-check	deep verdict on a ticket / hunch / premise
qa-sweep	team-scale QA over a broad surface, corroborated
empirical-proof	proves a finished change at the running app
code-quality-review	strict structure-first review of a diff
get-pr-comments	triages PR feedback into an action list
route-work	the model × effort reference table + hard routing invariants

Workbench: the process layer, implementing the workbench flow:

Skill	Does
audit	user-sized investigations, confirm-the-flags gate
brainstorming	design dialogue ending at your route pick
test-driven-development	TDD, default where a test harness exists; repo conventions take precedence
systematic-debugging	root cause before fixes
verification-before-completion	evidence before any "done" claim
receiving-code-review	rigor on arriving review feedback
using-workbench	session-start flow orientation + on-demand map
self-audit	retrospective on the process that ran the session (user-invoked only)

Five workbench skills derive from obra/superpowers by Jesse Vincent (MIT), adapted per docs/decisions/workbench-system.md: no hooks, no dispatcher, descriptions as honest triggers. Upstream drift tracking (workbench-drift, .claude/skills/) is repo-local maintenance tooling, not shipped in the plugin.

Details in plugins/workbench/README.md.

toolkit: optional, install when you want it

Utilities kept out of workbench so integrators control token load: every installed skill's listing rides in each session's context:

Skill	Does
html-report	report → polished dark HTML page (from a doc, or from session findings)
arch-map	visual architecture map when no doc exists
ui-demo-video	Playwright walkthrough video + verification frames
writing-skills	TDD applied to authoring skills (derived from obra/superpowers, MIT)
adopt-global-rules	installs the workshop's shipped global CLAUDE.md / AGENTS.md, rules, and Claude output styles onto a machine, additively (user-invoked only)
me-human	act as a human user dogfooding a system for real work (user-invoked only)

Details in plugins/toolkit/README.md.

Going deeper

• docs/workbench-flow.md: the workbench system's canonical mental model (with an arch-map rendering).
• docs/skills/ is the usage handbook, with one page per shipped skill covering what it does, when to reach for it, and the questions people actually hit. Start at its index.
• docs/decisions/ is the rationale layer, explaining why each agent and skill exists and the calls made along the way. The specs themselves are self-contained.