gh-toolkit

Use when the user wants to publish a package or deploy a site and asks for the release runbook, the publish steps, or how to ship to npm / Vercel. Triggers on "publish to npm", "npm publish", "release this package", "deploy to vercel", "vercel --prod", "ship it", "cut a release", "publish runbook", "how do I release", "push the new version live". Lays out the exact credentialed steps — one bypass-2FA npm automation token in ~/.npmrc, per-repo publish --access public with npx verify, vercel --prod with a token, and the GitHub release check — then chains to gh-launch-post. NEVER runs the credentialed steps itself; it documents them for the human to run.

Use when the user wants to kick off a new repo or feature build with Claude Code and asks for a mission prompt, a /goal, a "DO NOT STOP until" directive, an autonomous build brief, a starter command, or a non-gameable spec to drive an unattended agent. Triggers on "write me a /goal", "mission prompt for this repo", "goal prompt", "build brief", "autonomous build", "don't stop until vX", "acceptance criteria for the agent", "spec-first build", "make a Claude Code goal", "set up an agent to build X end to end". Produces a <4000-char /goal condition plus a starter command, with numeric criteria the agent cannot game, while leaving npm publish, deploys, and credentialed steps to the human.

Use when the user wants to add a language switcher, a translate button, an i18n toggle, or multi-language support to a page or component. Triggers on "translation button", "language switcher", "language selector", "i18n toggle", "add languages to this page", "multi-language", "localize this", "RTL support", "Arabic support", "lang switcher", "translate widget". Specs a reusable component supporting EN, IT, AR, NL, ZH-CN, ES, FR, DE, PT-BR — with Arabic as RTL and the active language driven by a ?lang= URL param (never localStorage) — and ships a drop-in vanilla HTML/JS snippet.

Use before tagging a release or merging important work, when the user wants a thorough, multi-agent, adversarial review that won't rubber-stamp the code. Triggers on "adversarial review", "pre-release review", "review before tagging", "multi-agent review", "red-team this", "tear this apart", "find every bug before we ship", "verify findings", "review with multiple reviewers", "is this really ready to release". Runs N reviewers across distinct dimensions, then adversarially verifies each finding before acting — fixing real issues with regression tests, recording declined findings with reasons. Run it as the gate before `v1.0.0`.

Use when coordinating GrowthHackers' multi-agent company / agent org and the user mentions the agent company, the 14 agents, the build engine, the agent bus, routing work to an agent, or making a structural/organizational decision across the agents. Triggers on "coordinate the agents", "agent org", "14-agent company", "route this to an agent", "update the bus", "agent-ceo", "W1..W13", "BUILD-ENGINE", "AGENT-ORG", "which agent owns this", "checkpoint the agents", "structural decision for the company". Requires reading BUILD-ENGINE.md + AGENT-ORG.md (the bus) before any structural decision, routing to agent-ceo and workers W1–W13, updating the bus, and checkpointing.